infiniflow/ragflow
40.1
Weak · 13 November 2026
706.5k
lines of production code
Go
with TypeScript
6
measurements over time
What this system is
Features
API Key Management Modal with 16-Key Limit
A new Chat API Key Modal component has been introduced to allow users to manage their API keys. This modal displays a list of existing tokens with options to copy them or remove them, and includes a 'Create New Key' button that is automatically disabled once the user has reached the limit of 16 API keys.
web/src/components/api-service/chat-api-key-modal · high confidence
Add Ant Design compatibility and common type definitions
The web application now includes explicit TypeScript interfaces to support UI components and shared state management. A new \antd-compat.ts\ file defines types for pagination, file uploads, table row selection, and form instances, ensuring type safety for Ant Design integration. Additionally, \common.ts\ introduces base interfaces for pagination state, search strings, and modal props, providing a standardized structure for component interactions across the interface layer.
web/src/interfaces · high confidence
Add BGPT form configuration UI
The BGPT agent form now includes input fields for configuring the number of top results (top\_n), an optional API key, and an optional lookback period in days (days\_back). This form also displays the query variable and output configuration, allowing users to adjust these parameters when setting up a BGPT agent.
web/src/pages/agent/form/bgpt-form · high confidence
Add CrawlerForm component for agent configuration
A new CrawlerForm component has been added to the agent form interface, allowing users to configure crawler-specific settings. This form includes fields for specifying a proxy URL and selecting an extraction type, utilizing a Zod schema for validation and integrating with the existing form context for state management.
web/src/pages/agent/form/crawler-form · high confidence
Add DocumentDownloadButton component for agent file retrieval
A new DocumentDownloadButton component has been introduced to allow users to download files associated with agents. This UI element displays the file name and type (such as PDF, Word, Excel, or text) and triggers a download via the updated downloadAgentFile API, which now handles agent-specific attachment retrieval.
web/src/components/document-download-button · high confidence
Add Execute SQL operator with bounded connection parameters
The agent workflow now includes an Execute SQL operator that lets users configure a database connection (type, host, port, credentials) and a SQL statement, then test the connection directly from the form. The form enforces bounds on the port (1–65535) and max records returned (1–10,000) to prevent unbounded queries, and it displays the test results in an output panel.
web/src/pages/agent/form/exesql-form · high confidence
Add Firecrawl web scraping integration for RAGFlow
Users can now import web content into RAGFlow using the Firecrawl integration. This new tool adds Firecrawl as a selectable data source in the RAGFlow UI, allowing users to configure their API key and scrape single URLs, crawl entire websites, or process batch URLs. The integration supports multiple output formats including markdown, HTML, links, and screenshots, and handles content cleaning, metadata extraction, and document chunking for RAG processing.
tools/firecrawl · high confidence
Add Go agent canvas service with session management, persistence, and cancellation
The internal/service package introduces the Go implementation of the agent canvas runtime. This includes the core AgentService for loading and executing agent workflows, session persistence to store chat history and thinking traces, and a cancellation tracker to manage active runs. The diff also adds comprehensive test coverage for agent session behavior, including cancellation, persistence, and deferred failure handling.
internal/service · high confidence
Add Go document engine for OceanBase and SeekDB
Introduces a new Go-based document engine for OceanBase and SeekDB databases, implementing core CRUD operations, hybrid search, and vector indexing. The engine handles document normalization, manages table creation and vector column indexing, and supports full-text and hybrid search capabilities. It includes specific handling for SeekDB index refresh and validates minimum version requirements for both OceanBase (4.3.5.1) and SeekDB (1.3.0.0 for index refresh).
internal/engine/oceanbase · high confidence
Add Google Scholar search form to agent configuration
A new form component has been added to the agent page, allowing users to configure Google Scholar search parameters. This form includes fields for setting the number of results (top N), sorting criteria, a date range (year low/high), and an option to include patents. It also supports query variables and displays the resulting output schema, enabling users to integrate Google Scholar data into their agent workflows.
web/src/pages/agent/form/google-scholar-form · high confidence
Add Keenable web search tool configuration form
Users can now configure the new Keenable web search tool within the agent builder. This change introduces a dedicated form component that allows users to input an API key, select a search mode (Web or Site) via a searchable dropdown, specify a target site URL, and adjust the top-N result count. The form validates inputs using Zod and integrates with the existing agent workflow to capture these settings.
web/src/pages/agent/form/sofya-form · high confidence
Add Langfuse configuration settings
Users can now configure Langfuse integration directly from the User Settings page. A new Langfuse card allows users to view their project, open the configuration dialog to set the public key, secret key, and host, and save or delete the configuration.
web/src/pages/user-setting/setting-model/langfuse · high confidence
Add PubMed agent configuration form
A new form component for configuring PubMed agents has been added, allowing users to set the search query, top N results, and an email address. The form uses shadcn-ui components for the interface and Zod for schema validation, integrating with the existing agent form infrastructure to manage state and output definitions.
web/src/pages/agent/form/pubmed-form · high confidence
Add Querit Contents tool configuration form
A new form component for the Querit Contents tool has been added to the agent builder. This form allows users to configure the tool by providing an API key, specifying target URLs, selecting the output format (Text, Markdown, or HTML), setting a crawl timeout between 1 and 60 seconds, and toggling the inclusion of extra metadata. The interface uses a search-enabled selector for format selection and includes validation for all input fields.
web/src/pages/agent/form/querit-contents-form · high confidence
Add RAGFlow Chat plugin for ChatGPT-on-WeChat
Users can now extend ChatGPT-on-WeChat with a new \ragflow\_chat\ plugin that integrates the RAGFlow API to provide retrieval-augmented generation capabilities. This addition introduces a new plugin component that handles conversational interactions by retrieving relevant data from external knowledge sources, allowing WeChat and other supported platforms to enrich chat responses with specific, retrieved information.
tools/chatgpt-on-wechat · high confidence
Add RagLineChart component for displaying line charts
A new \RagLineChart\ component has been added to the web application, built on the \recharts\ library. This component renders a responsive line chart that visualizes data points defined by \xAxis\ (string) and \yAxis\ (number) properties. It includes standard chart features such as a Cartesian grid, tooltips, and optional legends, providing a reusable UI element for displaying time-series or categorical data trends within the interface.
web/src/components/line-chart · high confidence
Add StringTransform operator form
Users can now configure a new StringTransform operator within the agent flow. This form allows selecting a transformation method (Split or Merge), specifying a reference variable for splitting, writing a script, and choosing delimiters, with the resulting output type automatically adjusted based on the selected method.
web/src/pages/agent/form/string-transform-form · high confidence
Add Vastbase G100 document engine support
Users can now store and query documents using the Vastbase G100 database as a document engine. This new Go-based implementation connects via the PostgreSQL-compatible wire protocol (using \lib/pq\) and supports both PG and B compatibility modes. It handles chunk and memory document storage with vector search capabilities (HNSW indexes), implements a delete-then-insert upsert pattern for data consistency, and ensures data format parity with the existing Python connector to allow shared table usage.
internal/engine/vastbase · high confidence
Add WenCai operator form
A new form component for the WenCai operator has been added to the agent configuration interface. This form allows users to configure the operator by specifying a query string, selecting a query type from a searchable dropdown, and setting a top-N limit for results, with the output structure defined by the operator's initial values.
web/src/pages/agent/form/wencai-form · high confidence
Add You.com search tool configuration form
A new form component for the You.com search tool has been added to the agent configuration interface. This form allows users to input an API key, select search result freshness (Any, Day, Week, Month, or Year) using a searchable dropdown, and configure the number of top results. It also displays the expected output variables for the tool.
web/src/pages/agent/form/youcom-form · high confidence
Add calendar component with range selection styling
A new Calendar component has been added to the UI library, wrapping the react-day-picker library to provide a consistent date selection interface. This component supports date range selection, featuring specific visual styles for the start and end dates of a range (rounded corners and distinct background colors) and includes navigation chevrons for moving between months.
web/src/components/originui/calendar · high confidence
Add data compilation form to agent configuration
Users can now configure data compilation steps within the agent creation flow. A new form component has been added to the agent canvas, allowing users to select a compilation template group and configure LLM settings. The form includes validation to ensure required fields are filled and handles errors related to unavailable templates or models.
web/src/pages/agent/form/compilation-form · high confidence
Add folder creation dialog with input validation
Users can now create new folders via a dedicated dialog in the file manager. The dialog includes a form that validates the folder name, ensuring it is not empty and does not contain forward slashes (/), providing immediate feedback if the input is invalid.
web/src/pages/files/create-folder-dialog · high confidence
Add loop operator node form
The loop operator node now includes a dedicated configuration form allowing users to define loop variables, set termination conditions, and specify the maximum loop count. This form supports dynamic variable selection, logical operators for conditions, and input modes for constant or variable values, with validation provided by a Zod schema.
web/src/pages/agent/form/loop-form · high confidence
Add note node to agent canvas
Users can now add note nodes to the agent canvas for annotations. These nodes feature a resizable container with a header for naming and a text area for content, with form changes automatically synchronized to the canvas state.
web/src/pages/agent/canvas/node/note-node · high confidence
Add pipeline run sheet with file upload for agent workflows
A new PipelineRunSheet component has been added to the agent page, providing a modal interface for running pipeline operations. This sheet includes an UploaderForm that allows users to upload files (limited to one file) via a direct upload component and submit the run, integrating with the existing run dataflow hooks to handle loading states and execution.
web/src/pages/agent/pipeline-run-sheet · high confidence
Added C++ bindings for RE2 and Darts-clone libraries
The internal C++ binding layer now includes the RE2 regular expression engine and the Darts-clone double-array trie library. This adds support for high-performance regex compilation, DFA/NFA-based searching, and fast string lookups via the new \internal/binding/cpp/re2\ and \internal/binding/cpp/darts\ directories.
internal/binding/cpp/darts, internal/binding/cpp/re2, internal/binding/cpp/util · high confidence
Added Go implementation for citation prompt templates
The Go agent now includes a dedicated component for generating citation instructions, mirroring the canonical Python templates. This addition introduces a \CitationPrompt\ function for standard citation rules and a \CitationPlusPrompt\ function that renders a post-stream grounding template, allowing the agent to correctly format and inject source citations into its output.
internal/agent/component/prompts · high confidence
Added Google Search operator form configuration
A new form component for the Google Search operator has been introduced, allowing users to configure search parameters directly within the agent workflow. This form enables the specification of a query variable, an API key, pagination settings (start index and number of results), and optional filtering by country and language. The component integrates with the existing form system to validate inputs and manage state for this specific retrieval tool.
web/src/pages/agent/form/google-form · high confidence
Added MCP import dialog UI
Users can now import MCP configurations via a new dialog in User Settings. This change introduces the frontend form and dialog components (import-mcp-form.tsx and index.tsx) that allow selecting a platform (defaulting to RAGFlow) and uploading a JSON file, with validation and submission handling wired to the parent modal.
web/src/pages/user-setting/mcp/import-mcp-dialog · high confidence
Added RAG text analyzer binding with CGO support
The internal binding layer now includes a new RAG (Retrieval-Augmented Generation) text analyzer. This change introduces a Go wrapper (\rag\_analyzer.go\) that interfaces with a native C++ tokenizer library via CGO, enabling features such as tokenization, fine-grained analysis, position tracking, and language-specific stemming. A corresponding stub implementation (\rag\_analyzer\_nocgo.go\) is provided to ensure the codebase builds successfully on systems where CGO is disabled, returning errors that indicate the native dependency is required for actual analysis.
internal/binding · high confidence
Added WarpCrop alignment regression tool
A new Python script, \tools/render\_diff/warp\_align.py\, has been added to the \tools/render\_diff\ directory. This tool validates the geometric accuracy of the Go \WarpCrop\ implementation by comparing its perspective de-skewing results against an independent OpenCV-based reference. It generates skewed text quads, runs the Go rendering pipeline, and calculates pixel Mean Squared Error (MSE) to ensure the Go code reproduces the same crop geometry as the reference, isolating the warp logic from model-dependent recognition steps.
_tools/render\diff · high confidence
Added chat widget demo pages
The example/chat\_demo area now includes two new HTML files, index.html and widget\_demo.html, which serve as demonstration pages for a floating chat widget. These pages embed an iframe pointing to a local chat service and include JavaScript to handle widget visibility, toggling, and scroll passthrough, allowing users to preview the widget's integration and behavior on a sample website.
_example/chat\demo · high confidence
Added chat widget page for embedded deployment
A new page component has been introduced at the widget route that renders the FloatingChatWidget. This page is specifically styled with transparent backgrounds and zero margins/padding to ensure the chat interface blends seamlessly into an external host page when embedded.
web/src/pages/next-chats/widget · high confidence
Added chunk result bar UI components for knowledge base data flow
Introduced two new React components in the data-flow result view: ChunkResultBar, which provides a toggle for switching between 'Full' and 'Ellipse' text display modes and a button to create new chunks (when not readonly), and CheckboxSets, which offers a 'Select All' checkbox and a conditional 'Delete' button for managing selected chunks. These components support the visualization and configuration of parsing configurations within the knowledge base data pipeline.
web/src/pages/dataflow-result/components/chunk-result-bar · high confidence
Added horizontal timeline component for data flow visualization
A new \TimelineDataFlow\ component has been introduced in the data flow result page to visualize the parsing configuration of the knowledge base. This component renders a horizontal timeline using the \CustomTimeline\ UI element, displaying pipeline steps as nodes. It allows users to interact with the timeline by clicking on steps, which triggers a callback function to update the active state and potentially navigate or display details for the selected pipeline step.
web/src/pages/dataflow-result/components/time-line · high confidence
Added migrate-canvas tool for normalizing agent DSL files
A new Go-based CLI tool has been added to normalize agent canvas DSL JSON files using the \dsl.NormalizeForCanvas\ function. The tool supports pretty-printing normalized output to stdout, validating changes against golden files in CI mode, and batch-processing directories. This ensures consistent formatting and structure for agent definitions during migration or development workflows.
tools/migrate-canvas · high confidence
Added mobile detection and toast notification hooks
New React hooks have been added to the application: \use-mobile.tsx\ provides a \useIsMobile\ hook to detect screen width changes relative to a 768px breakpoint, and \use-toast.tsx\ implements a state-driven toast notification system inspired by react-hot-toast, supporting adding, updating, dismissing, and removing toasts with a limit of one visible toast at a time.
web/src/components/hooks · high confidence
Added retrieval documents selection component for next-search
Introduced a new RetrievalDocuments component in the next-search area, providing a popover-based multi-select interface for users to choose specific documents for testing. The component integrates with existing knowledge request hooks to display document lists, supports filtering via search input, and manages selection state to trigger testing workflows.
web/src/pages/next-search/retrieval-documents · high confidence
Agent canvas editor rewritten with new architecture and components
The agent editor page has been rebuilt from the ground up, introducing a new state management layer via \useGraphStore\ (Zustand) and a comprehensive history manager for undo/redo operations. The UI now leverages \@xyflow/react\ for the canvas, \react-hook-form\ with Zod for operator forms, and \shadcn/ui\ components, replacing the previous implementation. This change adds new operator forms (such as SearXNG and Tokenizer), implements a canvas checklist for validation, and introduces features like global variable sheets, pipeline log sheets, and version dialogs within the agent context.
web/src/pages/agent · high confidence
Agent form now supports dynamic prompts and direct tool/MCP synchronization
The agent configuration form has been refactored to support dynamic, multi-turn prompts (User/Assistant roles) via a new \DynamicPrompt\ component, replacing the previous single-string prompt field. Additionally, tools and MCP servers are now managed directly on the canvas and synchronized to the form state without being stored within the form's primary data payload, ensuring that tool configurations are preserved and updated independently of the main form submission. The form also includes new UI components for managing these tools and agents, with specific handling for code tools and structured output settings.
web/src/pages/agent/form/agent-form · high confidence
Automated Go component and tool parity matrix generation
A new developer tool (\tools/gen-component-parity\) has been added to automatically generate the component and tool registry inventory for the Go agent. This tool scans the live Go component and tool registries to produce a Markdown or TSV parity matrix, replacing the previously manually-curated table. It covers Canvas DAG components (Universe A) and eino ReAct tools (Universe B), handling cases where components require non-empty parameters for construction. Tests have been added to verify the structural integrity of the generated Markdown and TSV outputs.
tools/gen-component-parity · high confidence
Begin operator form refactored with webhook security and dynamic input support
The Begin operator form has been rebuilt using shadcn UI components and a new Zod-based validation schema, introducing a dedicated Webhook mode alongside the existing Conversational and Task modes. In Webhook mode, users can now configure security settings including IP whitelists (with IPv4/IPv6 CIDR validation), rate limiting, JWT authentication, and HMAC signatures, while defining request schemas for query, headers, and body. The form also supports dynamic input fields with optional values and options, and ensures unique keys for inputs to prevent configuration errors.
web/src/pages/agent/form/begin-form · high confidence
Canvas debug (dry-run) mode with result preview and log tracking
Users can now run a side-effect-free debug of their ingestion pipelines to preview parsed chunks and view component logs without persisting data to the knowledge base. The debug context skips indexing and embedding, limits the preview to the first few chunks, and captures a detailed, per-component progress log in Redis that the front-end renders as a timeline. The run result is attached to the log's END marker so the UI can display the parsed output directly.
internal/ingestion/task · high confidence
Chat channels can now connect to agents and assistants
Users can now link chat channels to both assistants (dialogs) and agents via a new connection dialog. The added channel card displays the connected target (agent or assistant) and provides a 'Connect' button to open the selection modal, which uses a searchable dropdown to choose from available assistants and agents. Additionally, WhatsApp channels now support a QR code view for connection status, and the channel list UI has been updated to reflect these connection states.
web/src/pages/user-setting/chat-channel/component · high confidence
Configurable token-count cap with sentence-boundary re-split for title chunker
The title chunker form now supports a configurable token-count cap (defaulting to 512, with a range of 128–8000) and implements sentence-boundary re-splitting logic to ensure chunks do not break mid-sentence when the cap is reached. The form UI includes inputs for this cap, switches for including heading content and treating root chunks as headings, and dynamic hierarchy options that adjust based on the selected method (Hierarchy vs. Group). The underlying hook handles backward compatibility by migrating legacy \levels\ and \rules\ data structures to the new \hierarchyRules\ and \groupRules\ format, ensuring existing configurations are preserved during the upgrade.
web/src/pages/agent/form/title-chunker-form · high confidence
Dataset-level knowledge compilation logging and deduplication
The knowledge compilation pipeline now records dataset-level ingestion logs for all task types (Wiki, Tree, Graph, Mindmap, Timeline, PageIndex) and implements a batch-level deduplication system using LLM-based merge decisions. This ensures that when multiple documents contribute to the same dataset, duplicate entities are merged and distinct ones are preserved, with the results properly scoped to the dataset and reflected in the navigation tree.
_internal/ingestion/knowledge\compile · high confidence
Docker deployment configuration overhaul and Go stack introduction
The Docker deployment files have been restructured to support a new Go-based backend alongside the existing Python services. A new \.env-go\ configuration file and \docker-compose-go.yml\ define the Go stack, which uses Kvrocks as its cache and queue backend instead of Valkey/Redis, and exposes additional ports for the Admin server and Go-specific services. The main \docker-compose.yml\ and \.env\ files have been updated to include NATS and ClickHouse services, expose MCP (Model Context Protocol) ports, and support the new Go deployment profiles. The configuration now allows users to choose between the Python and Go stacks via Docker Compose profiles, with the Go stack requiring specific environment variables for Kvrocks connectivity.
docker · high confidence
Embed dialog now persists widget customization settings
The embed dialog component now supports saving and restoring full widget configuration state, including theme, embed type, avatar visibility, and color accents. Users can customize the appearance of their shared chat or agent widgets, and these settings are persisted so they are automatically restored when the dialog is reopened.
web/src/components/embed-dialog · high confidence
Event sourcing and replay capabilities added to the harness
The harness now includes an internal event-sourcing subsystem that records graph execution events (LLM calls, tool usage, state changes) into an append-only log. This provides a \LocalFileEventStore\ for durable, JSONL-based persistence with automatic segment rotation and garbage collection, alongside a \MemoryEventStore\ for in-memory operations. The system supports event filtering, streaming, and seeking by logical clock, and includes a context-based \EventRecorder\ to integrate with the agent's callback manager, enabling detailed tracing and replay of agent runs.
internal/harness/events · high confidence
Go Agent Canvas: Multi-turn ReAct Agent, MCP Tool Integration, and Artifact Handling
The Go agent canvas now includes the core Agent component, enabling multi-turn ReAct reasoning via the eino library. This component supports configurable tools, including Model Context Protocol (MCP) servers, and correctly handles sub-agents. It also introduces artifact collection from tool responses (such as CodeExec), rendering them as markdown image or download links in the chat stream. The Begin component serves as the canvas entry point, injecting query and user\_id into the shared state and mapping DSL inputs. A unified Component interface and BeOutput helper standardize the runtime contract for all canvas nodes. Also in this area: 17 commits (8 fixes) modifying internal; Agent tool budgets, sampler parameter gating, and canvas permission enforcement; Sandbox configuration simplification and type coercion unification; Agent Invoke variables now support literal values and Retrieval enforces dataset permissions; Improved agent error readability and parallel resume reliability; and 1 more.
internal/agent/component · medium confidence · unverified
Go Elasticsearch engine implementation with refresh-aware chunk insertion and metadata filter pushdown
The internal Elasticsearch engine is now implemented in Go, introducing a new chunk storage and retrieval layer. Chunk insertion distinguishes between API reads and ingestion pipelines: \InsertChunks\ waits for an index refresh to ensure immediate visibility for users querying data, while \InsertChunksNoRefresh\ omits the refresh parameter to avoid latency overhead during ingestion. The engine supports specialized index mappings for memory messages and skills, and includes a metadata filter translator that pushes down complex filter clauses (including operators like \in\, \not in\, and range comparisons) directly into Elasticsearch DSL queries.
internal/engine/elasticsearch · high confidence
Go NLP service: dataset navigation, query building, and reranking logic
The Go NLP service now includes a dataset navigation system that manages hierarchical document clusters via a dedicated index, a query builder that precompiles regexes for stop-word removal and special-character escaping to improve retrieval performance, and a reranker that normalizes scores from diverse providers (e.g., NVIDIA, Cohere) to a consistent \[0, 1\] range while preferring natural chunk text over tokenized content for cross-encoder scoring. Also in this area: 5 commits (1 fix) modifying internal/service; 7 commits (2 fixes) modifying internal; Fix multi-tenant retrieval scope and update dataset navigation integration test; Prevent duplicate search results for repeated tenant IDs; Knowledge graph type filtering now uses a new type\_kwd column with legacy fallback; and 2 more.
internal/service/nlp · medium confidence · unverified
Go admin server introduces role-based access control and model provider management
The Go-based admin server now supports role-based access control (RBAC) and model provider management. Administrators can create, update, and delete roles, grant and revoke permissions, and manage model providers and instances. The server also includes endpoints for user management, system settings, and ingestion task management. Tests verify the functionality of the admin handler and service layers.
internal/admin · high confidence
Go agent DSL parsing, normalization, and reset logic
The Go agent runtime now includes a dedicated DSL package that mirrors the Python backend's canvas handling. This adds helpers to safely extract component input forms, parameters, and names from the agent canvas map, along with a normalizer that repairs React-Flow edge handles, derives missing graph layouts from components, and converts legacy loop/iteration node types. It also introduces reset functions that clear per-session history and execution paths, and wipe per-run state (history, retrieval, memory) while restoring environment variables to their declared defaults, ensuring the Go agent behaves consistently with the Python version when starting new sessions or resetting canvases.
internal/agent/dsl · high confidence
Go agent audio synthesizer wired to model provider service with KVROCKS caching
The Go agent's audio package now implements a production-ready TTS synthesizer backed by the per-tenant model provider service. The new \model\_provider\_synthesizer.go\ routes TTS requests through a \ModelProviderFunc\ callback (plugged in at boot) and adds a KVROCKS-based cache layer that mirrors the Python \synthesize\_with\_cache\ behavior (SHA-256 keying, 7-day default TTL, \RAGFLOW\_TTS\_CACHE\_TTL\_SECONDS\ override). The \tts\_dispatch.go\ file provides a \TTSDispatcher\ interface and \NewTTSDispatchFunc\ to translate audio requests into the model provider's \AudioSpeech\ call shape, handling built-in engine selectors (gtts/edge-tts/custom) by falling back to the tenant's default TTS model. The \tts.go\ file defines the \Synthesizer\ interface, engine constants, and error types (including \ErrTTSEngineNotConfigured\ and \ErrSynthesizeEmpty\), while tests in \tts\_dispatch\_test.go\, \model\_provider\_synthesizer\_test.go\, and \tts\_test.go\ verify cache key stability, TTL env overrides, dispatch field mapping, and stub behavior.
internal/agent/audio · high confidence
Go agent canvas port with checkpointing and cancellation support
The Go agent canvas implementation in internal/agent/canvas is now available, providing a Go port of the Python canvas engine. This includes core DSL compilation (compile.go), state management (canvas.go), and history/memory decoding (decode.go). A new Redis-backed checkpoint store (checkpoint\_store.go) enables workflow interruption and resumption, while a dedicated cancellation mechanism (cancel.go) allows users to stop running workflows via Redis keys. The port also introduces compile-time guards to reject incompatible nodes (like UserFillUp) in resume mode and handles legacy component names for backward compatibility. Also in this area: 17 commits (8 fixes) modifying internal; Sandbox configuration simplification and type coercion unification; Simplifies XunFei Spark provider configuration and request handling; Stabilized workflow terminal merging and message ID generation; Improved agent error readability and parallel resume reliability.
internal/agent/canvas · medium confidence · unverified
Go backend database access layer and migration logic
The Go backend introduces a new data access object (DAO) layer in the \internal/dao\ package, providing database operations for conversations, connectors, search, and sync tasks. It also includes a comprehensive migration system that handles schema changes (such as renaming columns and adding unique indexes), backfills legacy data into new tenant model tables, and splits conversation history into dedicated message and reference tables.
ragflow · high confidence
Go document engine supports Elasticsearch, Infinity, OceanBase, SeekDB, and SereneDB
The internal document engine now supports multiple storage backends. Elasticsearch is fully functional for search, indexing, and document operations. Infinity is available as a placeholder implementation pending the official Go SDK. OceanBase and SeekDB share a SQL-based implementation. SereneDB is also supported as a new document-store engine. The engine is initialized at startup based on configuration and provides a unified interface for chunk and metadata operations.
internal/engine · high confidence
Go implementation of NER and relation extraction for ingestion
The ingestion pipeline now includes a Go-based Named Entity Recognition (NER) and relation extraction engine in the \internal/ingestion/compilation/extractor\ package. This implementation mirrors the existing Python logic to ensure identical output across both code paths. It leverages a C++ ThincNER engine (via cgo) for entity detection and applies pure-Go regex and dependency-tree rules to extract typed relations (such as \founded\_by\, \works\_for\, \located\_in\, and \born\_in\) across seven languages (English, Chinese, German, French, Spanish, Portuguese, and Japanese). A \wire\ package has also been added to centralize the registration of these ingestion components, ensuring they are correctly initialized by command-line entry points.
internal/ingestion/compilation · high confidence
Go implementation of the chunk service and vector hydration
The internal chunk service is now implemented in Go, providing the core logic for managing document chunks including creation, updates, deletion, and retrieval. This implementation introduces a new vector hydration mechanism that fetches embedding vectors for chunks on demand, supporting multiple storage backends such as Elasticsearch (parsing tab-separated or array vectors) and Infinity (skipping search as vectors are included in results). The service also includes comprehensive unit tests for vector parsing, chunk validation, and knowledge base embedding key resolution. Also in this area: 5 commits (1 fix) modifying internal/service; 17 commits (8 fixes) modifying internal; Chunk service access control and search field fixes; Agent canvas permissions and session management behavior changes; Admin heartbeat API returns structured data and user creation no longer auto-provisions LLM configs.
internal/service/chunk · medium confidence · unverified
Go sandbox providers now accept admin-panel settings and environment variables
The sandbox providers (Aliyun, E2B, and others) now read their configuration from both the admin-panel settings JSON and environment variables, using the same lowercase field names as the Python side. This allows operators to configure providers via the UI or via environment variables, with the same validation and defaults applied.
internal/agent/sandbox · high confidence
Go server configuration and runtime variable management
The Go server now uses a centralized configuration system (config.go) that loads settings from YAML files and environment variables, supporting multiple backends including MySQL, Elasticsearch, MinIO, NATS, ClickHouse, and Kvrocks. A new runtime variable management system (variable.go) handles secret keys with Redis-backed persistence and atomic creation. The server also introduces local variables for runtime-modifiable settings like server name (local\_variables.go) and an enterprise edition server interface (server\_ee.go) for OpenTelemetry tracing. Tests verify Kvrocks cache engine support and prevent MCP API key leakage in configuration logs.
internal/server · high confidence
Initial Nginx configuration for RAGFlow deployment
This change introduces the core Nginx configuration files required to serve the RAGFlow application. It includes \nginx.conf\ for global settings (such as a 1GB max body size), \proxy.conf\ for reverse proxy headers and timeouts, and two server block configurations: \ragflow.conf\ for standard HTTP access and \ragflow.https.conf\ for HTTPS with SSL certificate handling and HTTP-to-HTTPS redirection. These files define how static assets, the admin API, and the main application API are routed to the backend services.
docker/nginx · high confidence
Initial Storybook configuration for the web application
Storybook is now configured for the web frontend, enabling developers to view and interact with UI components in isolation. The setup includes a Webpack 5 framework, integration with Tailwind CSS and Less styling, and a global decorator that wraps stories in a TooltipProvider to ensure consistent UI behavior during development.
web/.storybook · high confidence
Initial release of the RAGFlow Python SDK
The Python SDK for RAGFlow is now available, providing a programmatic interface to interact with the RAGFlow API. The package exposes core modules for managing datasets, chats, sessions, documents, chunks, agents, and memory through the main \RAGFlow\ client class. It includes runtime type validation via \beartype\ (when installed) and supports standard HTTP operations for creating, listing, and deleting resources.
_sdk/python/ragflow\sdk · high confidence
Introduce Agent Harness Go framework for stateful multi-agent applications
The internal/harness package now provides a Go framework for building stateful, multi-agent applications with LLMs. It includes a graph-based execution engine (graphengine) using Pregel-style BSP execution, and an Agent Development Kit (agentcore) that supports ReAct agents, middleware, workflows, checkpointing, human-in-the-loop interrupts, and streaming. The harness exposes a public API via harness.go, re-exporting core types like StateGraph, CompiledGraph, Checkpointer, and Agent interfaces, and is documented in README.md with quick-start examples for minimal StateGraphs and ReAct agents.
internal/harness · high confidence
Introduce Agent Log Sheet for workflow execution visibility
A new Log Sheet component has been added to the agent interface, providing a slide-out panel that displays a detailed timeline of the agent's workflow execution. This feature visualizes the sequence of nodes and tool calls (such as search and retrieval operations) with their status, elapsed time, and input/output details, allowing users to inspect the internal reasoning and data flow of an agent run directly from the chat interface.
web/src/pages/agent/log-sheet · high confidence
Introduce ArtifactForceGraph component with safe tooltips and entity count overlay
The new ArtifactForceGraph component renders knowledge-base graphs using react-force-graph-2d, featuring a node-count overlay (showing returned vs. total entities) and a tooltip system that safely escapes HTML in node descriptions to prevent injection. It also adds visual highlighting for hovered or pinned nodes, dims unrelated elements, and applies a weak center-gravity force to keep the layout stable.
web/src/components/artifact-force-graph · high confidence
Introduce Go Agent Harness framework with graph engine and agent core
Adds a new Go-based Agent Harness framework in \internal/harness\ that provides a Pregel-style graph execution engine (\graphengine/\) and a comprehensive Agent Development Kit (\agentcore/\). The graph engine supports stateful computation with nodes, edges, and checkpointing, while the agent core layer builds on it to offer ReAct agents, middleware, tool abstraction, workflow orchestration, and a push-based \AgentLoop\ for managing turns, preemption, and stop signals. This location specifically contributes the core loop implementation, checkpoint bridging, and deterministic agent transfer logic. Also in this area: 1 commit (0 fixes) modifying internal/harness, internal/router, internal/server; 17 commits (8 fixes) modifying internal; Stabilized cancellation tests in the internal harness.
internal/harness/core · low confidence · unverified
Introduce Go Agent Harness framework with graph engine and agent core
Adds a new Go-based Agent Harness framework in \internal/harness\ that provides a Pregel-style graph execution engine (\graphengine/\) and a comprehensive Agent Development Kit (\agentcore/\). The graph engine supports stateful computation with nodes, edges, and checkpointing, while the agent core layer builds on it to offer ReAct agents, middleware, tool abstraction, workflow orchestration, and a push-based \AgentLoop\ for managing turns, preemption, and stop signals. This location specifically contributes the core loop implementation, checkpoint bridging, and deterministic agent transfer logic. Also in this area: 7 commits (2 fixes) modifying internal.
internal/harness/graph/types · medium confidence · unverified
Introduce Go-based DOCX parser with native office\_oxide backend
The internal/deepdoc/parser/office package now provides a Go implementation for parsing DOCX files, replacing the previous Python-based extraction. It uses the office\_oxide library (via CGO) to extract structured blocks (paragraphs, tables, images) and maps them to the shared Section representation, ensuring headings are identified as titles and tables are rendered as HTML. The implementation includes a fallback for non-CGO builds and comprehensive tests to verify parity with the Python reference output. Also in this area: 6 commits (1 fix) modifying internal/ingestion; 17 commits (8 fixes) modifying internal; Improved PDF text preview quality and reference caption handling; Fixes for figure caption formatting and bounding box clamping in document analysis; Fix PDF caption merging and TOC/header/footer removal for page ranges.
internal/deepdoc/parser/pdf/inference, internal/deepdoc/parser/pdf/table · medium confidence
Introduce Go-based RAGFlow CLI with admin mode and virtual filesystem
A new Go implementation of the RAGFlow command-line interface is added to the \internal/cli\ package, providing a full-featured alternative to the existing Python CLI. The CLI supports two operational modes: standard user mode and a new administrator mode (\--admin\), enabling privileged operations such as managing users, roles, permissions, and system services. It features a virtual filesystem that allows intuitive navigation and management of datasets and files using familiar commands like \ls\, \cat\, and \search\. The tool includes a custom recursive descent parser for SQL-like syntax, ensuring compatibility with the Python CLI's command structure while offering enhanced error handling and output formatting options.
internal/cli · high confidence
Introduce KnowledgeBase context for dataset state management
Added a new React context (\KnowledgeBaseProvider\) and associated hooks (\useKnowledgeBaseContext\, \useOwnerTenantId\) to manage dataset-related state. This allows components within the dataset area to access the current knowledge base object and its loading status, as well as the tenant ID, without needing to pass these props through every level of the component tree.
web/src/pages/dataset/contexts · high confidence
Introduce Memory management page with create, edit, and delete capabilities
Users can now manage memories via a new dedicated page that lists existing memories and supports creating, editing, and deleting them. The page includes a filter bar for searching and filtering by owner, memory type, and storage type, along with pagination for large lists. A modal dialog allows users to create new memories or edit the name of existing ones, while a dropdown menu on each memory card provides quick access to rename and delete actions, including a confirmation dialog for deletions.
web/src/pages/memories · high confidence
Introduce Retrieval Testing interface for datasets
Adds a new Retrieval Testing page under the dataset section, allowing users to evaluate retrieval performance. The page features a split layout with a testing form on the left (including similarity thresholds, rerank settings, metadata filters, and a question input) and a results panel on the right displaying retrieved chunks with similarity scores, highlighted content, and document metadata. Users can submit queries via a button or by pressing Enter, and filter results by document.
web/src/pages/dataset/testing · high confidence
Introduce RunSheet component for agent debugging
A new RunSheet component has been added to the agent run-sheet page, providing a modal sheet interface for testing and debugging agent flows. It integrates with the graph store to retrieve and update node form inputs, specifically handling the 'begin' operator parameters, and passes these inputs to a DebugContent component for visualization and execution.
web/src/pages/agent/run-sheet · high confidence
Introduce SereneDB as a new document-store engine
Added a new Go-based SereneDB engine implementation in the internal/engine/serenedb package, providing a PostgreSQL-wire alternative to the existing Elasticsearch and Infinity backends. This engine supports full-text search via a BM25 inverted index, vector similarity search using an IVF index, and hybrid fusion queries, while maintaining compatibility with the existing RAGFlow document and metadata schemas. The implementation includes a new integration test suite to validate end-to-end chunk lifecycle and search behavior against a live SereneDB instance. Also in this area: 6 commits (1 fix) modifying internal/engine; Fixes for multi-tenant search scope, Infinity connection pooling, and Redis memory reporting.
internal/engine/serenedb · medium confidence · unverified
Introduce SwitchForm component with dot-notation variable paths
Added the SwitchForm component (web/src/pages/agent/form/switch-form) to manage agent switch conditions. This form now supports dot-notation field paths (e.g., \.field.subfield\) in Switch conditions, allowing users to target nested fields of object outputs. The implementation includes a \VariableWithPath\ picker that combines a dropdown base with an optional dotted-path suffix, and hooks to synchronize form changes with the canvas store.
web/src/pages/agent/form/switch-form · high confidence
Introduce UserFillUpForm for configurable user input collection
A new UserFillUpForm component has been added to the agent builder, enabling users to define custom input fields (such as text, file, or options) that are collected before the agent processes a request. The form supports optional guiding questions (tips) and, when file inputs are detected, integrates layout recognition options. This change introduces the specific UI and state-management logic for this input-collection node within the agent workflow.
web/src/pages/agent/form/user-fill-up-form · high confidence
Introduce centralized TypeScript interfaces for the database layer
The \web/src/interfaces/database\ directory now contains a comprehensive set of TypeScript interfaces that define the shape of data exchanged with the backend. This change introduces type definitions for core application domains including agents (DSL, operators, templates), chat (messages, references, dialogs), datasets (documents, chunks, parsing configs), and knowledge base structures (navigation, skills, graph entities). It also adds interfaces for LLM providers, MCP servers, file management, ingestion events, and system status, providing a unified and strongly-typed contract for the frontend to interact with the database and API responses.
web/src/interfaces/database · high confidence
Introduce dataset compilation utility functions for navigation, skills, and diffs
This change adds a new set of utility functions in the dataset compilation module to support the new data compilation layer. Specifically, it introduces \nav-tree.ts\ to build navigation tree data structures, handle search result nesting, and mount structure graphs under document nodes; \skill-tree.ts\ to build, filter, and count skill tree nodes for the skills list view; \parse-progress-msg.ts\ to parse backend progress logs into structured entries; and \parse-wiki-diff.ts\ to parse wiki diff strings into hunks and changed lines. These utilities provide the data transformation logic required by the UI components for navigation, skills, and wiki diff display.
web/src/pages/dataset/compilation/utils · high confidence
Introduce dedicated SimilaritySliderFormField component
A new \SimilaritySliderFormField\ component has been added to manage similarity configuration within forms. It exposes two controls: a \similarity\_threshold\ slider (0–1) and a \keywords\_similarity\_weight\ slider (0–1) that explicitly balances full-text versus vector weights. The component enforces a \[0,1\] range for the keyword weight, normalizes values to two decimal places, and displays the complementary vector weight as 1 minus the keyword weight.
web/src/components/similarity-slider · high confidence
Introduce dedicated dataflow result page with chunk management and pipeline visualization
The dataflow-result location now provides a dedicated interface for reviewing and managing knowledge base parsing results. Users can view a visual timeline of the pipeline steps (parser, chunker, etc.) and interact with the parsed content through a split-pane layout: a document preview on the left and a chunk editor on the right. This location introduces specific capabilities for chunk management, including the ability to select, edit, delete, and switch the availability status of individual chunks. It also supports dynamic text display modes (full vs. ellipsis) and highlights within the document preview. The implementation relies on new React Query hooks for scoped pipeline log detail fetching and utilizes a new set of constants and interfaces to define the pipeline DSL structure and search parameters.
web/src/pages/dataflow-result · high confidence
Introduce dedicated parsing editor components for dataflow results
Added a new \parse-editer\ component suite (including \hook.ts\, \index.tsx\, \interface.ts\, \json-parser.tsx\, and \object-parser.tsx\) to the dataflow-result page. This new implementation provides a structured editor for parsing configurations, supporting both JSON/array-based and text/object-based content types. It introduces features such as dynamic field name support, inline editing via contentEditable divs, chunk selection with checkboxes, and automatic scrolling to newly created chunks, replacing the previous ad-hoc or less structured editing approach.
web/src/pages/dataflow-result/components/parse-editer · high confidence
Introduce global conversation variable management sheet
A new Global Variable Sheet has been added to the agent configuration interface, allowing users to view, add, edit, and delete global conversation variables. This feature includes a modal form for defining variable names, types (including support for arrays and objects), default values, and descriptions, with the sheet persisting these changes to the agent's DSL configuration.
web/src/pages/agent/gobal-variable-sheet · high confidence
Introduce metadata management UI components
This change adds the core React components for the dataset metadata feature, including the main manage modal, column definitions for the metadata table, and a sub-modal for editing values. It introduces support for multiple metadata value types (string, time, number, list) with specific UI controls like a date picker for time fields, and provides the interface and constant definitions required to manage metadata fields, values, and their types within the dataset context.
web/src/pages/dataset/components/metedata · high confidence
Introduce new NextMessageInput component with manual resize and agentic thinking controls
A new \NextMessageInput\ component has been added to the application, providing users with a draggable handle to manually resize the chat text area and a dropdown to select from six thinking levels (including a new 'Agentic' mode). The component also integrates file upload capabilities, web search toggles, and audio input support, replacing the previous input implementation.
web/src/components/message-input · high confidence
Introduce new user settings sidebar with collapsible mobile layout
The user settings page now uses a dedicated sidebar component that displays navigation items (Model Providers, Data Sources, Chat Channels, MCP, Team, Profile, API) with icons and labels. On mobile devices, the sidebar collapses to an icon-only rail to save space, while on larger screens it expands to show full labels. The sidebar includes a logout button and displays the system version and theme switcher in the footer.
web/src/pages/user-setting/sidebar · high confidence
Introduce session management components for Agent Explore
Added four new components in the Agent Explore page to handle session interaction: SessionList displays a searchable list of sessions with a count of filtered results; SessionCard renders individual sessions with visual distinction for the selected state and new sessions; SessionDropdown provides a delete action for both temporary and persistent sessions; and SessionChat manages the chat interface, ensuring streaming answers remain bound to the correct session and handling message hydration.
web/src/pages/agent/explore/components · high confidence
Introduce wiki left-panel with contents and graph tabs
A new WikiLeftPanel component has been added to the dataset compilation view, providing a two-tab interface for navigating wiki content. The 'Contents' tab features a hierarchical topic navigation bar with breadcrumb support, search, and artifact listing, while the 'Graph' tab displays a force-directed graph of knowledge-base nodes with a search dropdown to select and highlight specific entities.
web/src/pages/dataset/compilation/wiki-left-panel · high confidence
Introduces Go agent runtime infrastructure
The internal/agent/runtime package now provides the foundational Go runtime for the agent canvas, including a component registry with category-aware lookups, context-based state management for cross-package access, and helper utilities for progress tracking, timeouts, and safe JSON serialization. This runtime layer enables the Go agent to execute workflows and manage component lifecycles, serving as the core execution engine for the ported agent capabilities. Also in this area: 17 commits (8 fixes) modifying internal; Permission checks now enforce canvas context for dataset access; Simplifies XunFei Spark provider configuration and request handling; New Go-based Agentic RAG chat mode with configurable templates and per-dialog failover; Improved agent error readability and parallel resume reliability.
internal/agent/runtime · medium confidence · unverified
Introduces a new custom modal component with programmatic API
Replaces the previous UI library modal implementation with a new custom modal component built on Radix UI. This change adds a programmatic API (\createPortalModal\) that allows modals to be rendered outside the main React component tree via DOM portals, supporting features like type-specific variants (warning, info, success, error, confirm), customizable sizes, and keyboard accessibility.
web/src/components/ui/modal · high confidence
Introduces new Agent Form Sheet UI with inline node renaming
The agent configuration interface now uses a new Form Sheet component (web/src/pages/agent/form-sheet) that renders operator-specific forms via a centralized configuration map. This update adds the ability to inline-edit node titles directly within the sheet header, supports single-step debugging for compatible operators, and integrates with the new shadcn-ui-based design system. The sheet also handles context providers for agent forms and manages visibility to prevent layout shifts.
web/src/pages/agent/form-sheet · high confidence
Introduces shared parser type definitions and DLA/TSR label contracts
The deepdoc parser pipeline now uses a new \internal/deepdoc/parser/type\ package to centralize shared types, interfaces, and constants for all format-specific parsers (PDF, DOCX, XLSX, etc.). This change introduces the \ParseResult\ struct, which now includes \PipelineMetrics\ for tracking box removal and merge stages, and \DLARegions\ for layout detection data. It also defines the \TextBox\ and \TableItem\ structures with new fields to support page-spanning tables, OCR scoping, and TSR table annotations. Additionally, it establishes the wire contract for the DeepDoc backend by defining the 10-class DLA and 6-class TSR label taxonomies, ensuring parity with the Python implementation.
internal/deepdoc/parser/type · high confidence
Introduces structured constants for agent operators, pipelines, and document generation
This change adds new constant definitions to the agent workspace, centralizing configuration for the UI. It defines output formats and features for the Document Generator (PDF, DOCX, TXT, Markdown, HTML), establishes output format mappings and default model fields for the ingestion pipeline (PDF, spreadsheets, images, email, etc.), and provides initial value templates for various agent operators such as Retrieval, Rewriting, Categorization, and Excel processing. These constants serve as the source of truth for default node configurations and available options in the agent canvas and pipeline forms.
web/src/pages/agent/constant · high confidence
Iteration operator form now supports configurable max concurrency and dynamic output variables
The iteration operator's configuration form has been updated to allow users to set a maximum concurrency level (0–32) via a slider and to define multiple output variables with specific names and types. Users can now add, remove, and map output variables to child node references, with the system automatically inferring the output type (e.g., Array\<T\>) based on the selected variable. This change introduces a new \DynamicOutput\ component and associated form logic to manage these outputs, replacing the previous static or less flexible output handling.
web/src/pages/agent/form/iteration-form · high confidence
Memory page layout and API action constants introduced
The Memory page now includes a dedicated layout component that renders a sidebar alongside the main content area, providing the structural foundation for memory-related views. Additionally, a new constants file defines specific API action types (FetchMemoryDetail, FetchMemoryMessage, FetchMessageContent) used to identify data fetching operations within this section.
web/src/pages/memory · high confidence
Native DeepDoc OCR and layout detection now use a pure-Go backend
The internal/deepdoc/native package introduces a new in-process Go implementation for DeepDoc's text detection (DB), layout analysis (DLA), and image processing, replacing the previous external or Python-bound approach. This change ports core algorithms—including bilinear resizing, contour extraction, polygon rasterization, and the Clipper offset unclip logic—to pure Go, ensuring bit-exact alignment with the Python reference pipeline for coordinate accuracy and score consistency. The backend now uses FlatBuffer (.ort) models for inference and includes comprehensive regression tests to validate geometric fidelity against Python goldens.
internal/deepdoc/native · high confidence
New API access key management page in user settings
A new API settings page has been added to the user settings section, providing a dedicated interface for managing API access keys. This page renders the existing API content component, allowing users to view and manage their API credentials directly from their account settings.
web/src/pages/user-setting/setting-api · high confidence
New API documentation and chat statistics components
This change introduces a new set of components within the chat-overview-modal area to enhance the API service experience. It adds a dedicated API documentation viewer (api-content.tsx) that loads and renders the HTTP API reference markdown, featuring a toggleable table of contents (markdown-toc.tsx) with smooth-scrolling anchor navigation (anchor.tsx) and deferred rendering to improve initial page load performance. Additionally, it includes a backend service API card (backend-service-api.tsx) for displaying the service URL and API key management, and a new statistics dashboard (stats-chart.tsx) that visualizes chat metrics such as PV, UV, speed, and token usage via line charts.
web/src/components/api-service/chat-overview-modal · high confidence
New API service hooks for token management and chart statistics
A new \hooks.ts\ file has been added to the \api-service\ component directory, introducing several React hooks to streamline API interactions. The \useOperateApiKey\ hook provides functionality to create, list, and remove system tokens, while \useSelectChartStatsList\ retrieves and formats chart statistics data from the React Query cache. Additionally, helper hooks like \useShowTokenEmptyError\ and \useShowBetaEmptyError\ display specific error messages for missing tokens or beta access, and \useShowEmbedModal\ orchestrates the logic to fetch tokens before displaying an embed modal.
web/src/components/api-service · high confidence
New CLI filesystem interface for managing skills, datasets, and files
The CLI now includes a virtual filesystem layer that exposes datasets, skills, and local files through a unified Unix-like interface. Users can list, search, and view content using standard commands like \ls\, \cat\, and \search\. A new \install-skill\ command allows installing skills from local paths, GitHub, ClawHub, or skills.sh into specific spaces, while \uninstall-skill\ removes them. All skill installations are automatically scanned for security threats (such as data exfiltration, prompt injection, and destructive commands) and enforced via trust levels (builtin, trusted, community), requiring explicit \--force\ confirmation for high-risk community sources.
internal/cli/filesystem · high confidence
New Datasets page with creation, renaming, and filtering capabilities
The Datasets page has been rebuilt to provide a dedicated interface for managing knowledge bases. Users can now create new datasets via a dialog that enforces required fields (name, embedding model, and parser/pipeline selection) and supports the Enter key for submission. Existing datasets can be renamed through a dedicated modal, and the list view includes an owner filter to narrow results by creator. The page also features pagination, search, and empty-state handling to improve navigation and usability.
web/src/pages/datasets · high confidence
New DuckDuckGo and ArXiv search operators added to Agent forms
Users can now configure DuckDuckGo and ArXiv search operators within the Agent workflow. The new DuckDuckGo form allows selecting a specific search channel (e.g., Web, News, Images) and setting the number of results, while the ArXiv form enables searching academic papers with options to sort by submission date, last update, or relevance. Both forms utilize a consistent UI pattern with search-selectable dropdowns and output variable mapping.
web/src/pages/agent/form/duckduckgo-form · high confidence
New ES to OceanBase migration tool for RAGFlow
A new CLI tool is available in tools/es-to-oceanbase-migration to migrate RAGFlow data from Elasticsearch to OceanBase. The tool handles schema conversion, vector data mapping, batch import, and includes resume capability and data consistency validation. It supports ES 8+ via the search\_after API, auto-detects vector field dimensions, and generates detailed migration reports.
tools/es-to-oceanbase-migration · high confidence
New EditTag component with IME-aware input handling
A new EditTag component has been introduced to manage tag creation and deletion. It allows users to add tags via an input field that confirms entries on Enter press, while correctly ignoring Enter keys during IME composition to prevent premature submission of half-typed characters. Tags can be removed individually, and the component includes basic styling and tooltip support for tag display.
web/src/components/edit-tag · high confidence
New FileIcon component for document thumbnails and type indicators
A new FileIcon component has been added to the web interface to display file-type indicators. It prioritizes showing a document thumbnail (fetched via authenticated requests) when available; otherwise, it falls back to a specific SVG icon based on the file extension (e.g., mp3, svg), and finally to a generic Lucide file icon if no specific icon is found. This component is now used to visually represent files in the application.
web/src/components/file-icon · high confidence
New Go DAO implementations for API tokens, chat sessions, and canvas templates
The internal/dao package now includes new Go data-access objects for API tokens (APITokenDAO), chat sessions (ChatSessionDAO), and canvas templates (CanvasTemplateDAO), along with seeding logic for built-in agent templates. These additions provide the database layer for managing API keys, conversation history, and template configurations in the Go backend, supported by unit tests that verify CRUD operations, tenant isolation, and template seeding idempotency.
internal/dao · high confidence
New Go agent canvas API route registration
The internal router now registers the full set of agent canvas endpoints in Go, including agent CRUD, versioning, session management, file operations, component introspection, and webhook triggers. This replaces the previous scattered registration and ensures all agent routes are consistently wired through the central router setup.
internal/router · high confidence
New Go agent tools for search, code execution, and external data retrieval
The agent canvas now includes a suite of new Go-based tools: an agentic search tool supporting hybrid, vector, and BM25 retrieval modes with keyword narrowing; an AkShare tool for fetching Chinese A-share stock news from East Money; an ArXiv tool for searching academic preprints; a BGPT tool for structured scientific paper search; and a CodeExec tool that dispatches Python and JavaScript code to a sandbox provider. Each tool is accompanied by unit tests covering parsing, schema validation, and integration with the canvas runtime. Also in this area: 17 commits (8 fixes) modifying internal; Canvas dependency checks and PubMed email fallback; Sandbox configuration simplification and type coercion unification; Dataset permission checks and Wikipedia extract truncation fix; Improved agent error readability and parallel resume reliability.
internal/agent/tool · medium confidence · unverified
New Go chat package for LLM invocation and message fitting
The internal/agent/chat package has been added in Go to provide a dependency-light interface for LLM chat invocations and message fitting. It defines the Invoker and StreamingInvoker interfaces, along with request/response types (Request, Response, Tool, ToolCall) that support native tool calling and streaming. The package also implements message fitting logic (FitMessages) to trim conversation history based on context length budgets while preserving image-only turns and handling multi-modal content, mirroring the previous Python implementation. A package-level singleton allows production and test code to inject or retrieve the default invoker without import cycles.
internal/agent/chat · high confidence
New Go common library with shared utilities and constants
This change introduces a new Go \internal/common\ package that consolidates shared utilities, constants, and types previously scattered or Python-specific. It provides name validation and duplicate-name generation for resources, standardized error codes with a \CodedError\ type for consistent API responses, and formatting helpers for bytes, numbers, and time. The package also includes HTTP client wrappers with SSRF protection and configurable LLM debug logging, JSON unmarshaling for flexible string/list parameters, and query-rewrite prompt building and parsing for knowledge graph search. Tests are included to verify the behavior of these components. Also in this area: 17 commits (8 fixes) modifying internal; Opt-in support for private database hosts via ALLOW\_ANY\_HOST; Sandbox configuration simplification and type coercion unification; Simplifies XunFei Spark provider configuration and request handling; New configuration options for MCP private hosts and MonkeyOCR; and 2 more.
internal/common · medium confidence · unverified
New Go model drivers for 302.AI, Aliyun, AnonRouter, API-Route, and Anthropic
The \internal/entity/models\ package adds Go implementations for several new model providers. The 302.AI driver supports chat (streaming and non-streaming), embeddings, and reasoning configuration. The Aliyun (Tongyi-Qianwen) driver implements chat with specific handling for Qwen3 thinking defaults and tool-call behavior for qwen-flash. AnonRouter and API-Route are added as thin wrappers around the existing OpenAI-compatible model driver to support their respective hosted gateways. The Anthropic driver implements the Messages API, handling authentication via the x-api-key header, system prompts, and multimodal content.
internal/entity/models · high confidence
New Go-based connectors for Airtable, Asana, Azure Blob Storage, Azure DevOps, and BigQuery
The syncer now includes native Go connectors for Airtable, Asana, Azure Blob Storage, Azure DevOps, and Google BigQuery. These connectors allow users to ingest data from these sources with checkpoint-based resume support, enabling reliable incremental synchronization that resumes from the point of interruption. The implementation includes robust configuration validation, credential handling, and specific logic for each platform (such as Azure DevOps' custom base URL support and BigQuery's query-based ingestion).
internal/syncer/connector · high confidence
New Go-based data source syncer framework
A new Go-based data source syncer has been introduced to replace the legacy Python sync worker. This framework manages the full lifecycle of sync and prune tasks, including scheduling via NATS/JetStream, task concurrency control, and connector validation. It provides a standardized interface for connectors to read external sources and normalize data into documents, while handling checkpoint persistence, resume capabilities, and document pruning. The syncer includes a robust task coordinator, a shared batch job executor for fair dispatching, and a connector lock mechanism to prevent concurrent writes for the same connector and knowledge base.
internal/syncer · high confidence
New Go-based storage engine with multi-backend support
The application now includes a Go implementation of the storage engine in the internal/storage package, providing a unified interface for object storage operations. This new backend supports multiple cloud providers, including AWS S3, Google Cloud Storage (GCS), Aliyun OSS, and MinIO, alongside an in-memory backend for testing. The implementation introduces a storage factory for initialization and management, along with comprehensive unit and integration tests to ensure reliability across different storage types.
internal/storage · high confidence
New IndentedTree component for rendering knowledge graphs
A new IndentedTree component has been added to render knowledge graphs using the @antv/g6 library. It visualizes tree-structured data with a left-to-right indented layout, supports canvas dragging, zooming, element dragging, and node collapse/expand, and automatically resizes and refits the graph when the container dimensions change.
web/src/components/indented-tree · high confidence
New Infinity vector search engine implementation in Go
The internal engine location now includes a complete Go implementation of the Infinity vector search engine, replacing the previous Python-based approach. This new engine handles chunk storage, metadata management, document indexing, and search operations using the Infinity Go SDK. Key features include connection pooling with configurable timeouts, HNSW vector indexing with LVQ encoding, meta-field filtering with SQL translation, and chunk transformation logic that maps ingestion field names to Infinity's schema. The implementation also includes specialized handling for skill indexes and supports both single and bulk document operations. Also in this area: 6 commits (1 fix) modifying internal/engine; Fix chunk insertion failures and document-scoped list filtering in Infinity engine; Fixes for multi-tenant search scope, Infinity connection pooling, and Redis memory reporting; Knowledge graph type filtering now uses a new type\_kwd column with legacy fallback; Admin heartbeat API returns structured data and user creation no longer auto-provisions LLM configs.
internal/engine/infinity · medium confidence · unverified
New Invoke and GitHub operator forms for agent configuration
Users can now configure Invoke and GitHub operators directly within the agent canvas. The new Invoke form allows setting the request URL (supporting variable placeholders), HTTP method, timeout, headers, proxy, and HTML cleaning options, along with managing custom variables via a dialog and table. The GitHub form provides fields for specifying a query and a top-N limit, enabling structured data retrieval within agent workflows.
web/src/pages/agent/form/invoke-form · high confidence
New JSON Editor component with theming and i18n support
A new JSON editor component has been introduced in the web application, wrapping the jsoneditor library to provide a consistent user experience. It features a 'cloud9\_night' visual theme via custom LESS stylesheets and supports dynamic language switching (English and Chinese) by recreating the editor instance when the application locale changes. The component exposes configuration options for modes (tree, code), indentation, and expansion state, ensuring that JSON data is displayed and edited with proper syntax highlighting and internationalized labels.
web/src/components/json-edit · high confidence
New JSON Schema Builder with visual editing and validation
The jsonjoy-builder component now includes a comprehensive set of new features for managing JSON schemas. Users can visually edit schemas through a split-pane interface that displays a tree-based visual editor alongside a raw JSON source view, with a resizable divider and fullscreen mode. A new Schema Inferencer allows users to generate a schema automatically from sample JSON input. Additionally, a JsonValidator dialog enables real-time validation of JSON data against the current schema, providing immediate feedback on syntax and structural errors. The editor also supports adding, editing, and deleting fields with type selection (string, number, boolean, object, array) and required status toggling.
web/src/components/jsonjoy-builder · high confidence
New LLM setting form components for configurable model parameters
The \web/src/components/llm-setting-items\ directory now contains a new set of UI components (\LlmSettingFieldItems\, \LLMFormField\, \SliderInputSwitchFormField\) that provide a structured form for configuring LLM parameters such as temperature, top\_p, max\_tokens, and thinking mode. These components integrate with \react-hook-form\ and Zod schemas to manage validation, support a 'freedom' toggle to switch between preset and custom parameter values, and ensure slider inputs respect their defined step precision.
web/src/components/llm-setting-items · high confidence
New Lexical-based Markdown editor with Mermaid support
The wiki's Markdown editor has been replaced with a new WYSIWYG editor built on Meta's Lexical framework, replacing the previous implementation. This change introduces a rich-text editing experience that supports standard Markdown formatting, tables, and lists, while also adding native support for rendering and editing Mermaid diagrams directly within the editor. A raw Markdown source view is available via Monaco Editor, and the entire editor component is fully theme-aware, adapting to the application's light and dark modes.
web/src/lib/editor · high confidence
New Loop and Parallel workflow nodes for agent canvases
The agent canvas now supports iterative and fan-out execution patterns via two new workflowx components: a Loop node that repeatedly runs a sub-workflow until a condition is met (with configurable max iterations, stream modes, and checkpointing), and a Parallel node that processes a slice of inputs concurrently with bounded concurrency and per-item interrupt/resume support. These nodes are exposed through the \internal/agent/workflowx\ package and wired into the canvas engine so that DSL definitions using Loop or Parallel components are executed as single composite nodes with full checkpoint and resume capabilities.
internal/agent/workflowx · high confidence
New Markdown Content Renderer for Next Search
Introduces a new MarkdownContent component for the next-search page that renders chat responses with enhanced security and formatting. The component sanitizes input using DOMPurify (allowing custom tags like 'think' and 'section'), escapes unpaired angle brackets to prevent parsing errors, and supports LaTeX rendering via rehype-katex. It also handles citation references by displaying document thumbnails and opening external web document links directly in new tabs.
web/src/pages/next-search/markdown-content · high confidence
New Markdown document preview component with prose styling
A new \Md\ component has been added to the document preview module to render Markdown files. It fetches content via a GET request, displays a loading spinner while retrieving data, and handles errors by showing a \FileError\ message. The rendered Markdown is styled using Tailwind CSS classes to apply a 'prose' typography theme, ensuring consistent text colors, link styles, and code block backgrounds for both light and dark modes.
web/src/components/document-preview/md · high confidence
New PDF document preview drawer component
A new \PdfSheet\ component and its associated \useClickDrawer\ hook have been added to the \pdf-drawer\ directory. This feature introduces a slide-out drawer (using the \Sheet\ UI primitive) that displays a \DocumentPreview\ for a specified document ID and chunk. The component handles fetching the document URL, extracting file type information, and passing highlight positions to the previewer, allowing users to view document content directly from the interface.
web/src/components/pdf-drawer · high confidence
New PDF parsing utility library for image cropping, geometry, and text analysis
The PDF parser now includes a new Go utility package (\internal/deepdoc/parser/pdf/util\) that implements core image and text processing functions previously handled by the Python backend. This adds support for cropping section images with context padding, calculating geometric overlaps and distances between text boxes, detecting garbled text via font encoding and character analysis, and identifying English-language pages. It also provides a 1D K-Means clustering implementation for layout analysis and robust parsing of position tags, including support for negative coordinates. Comprehensive unit tests are included to verify correctness against the Python reference implementation.
internal/deepdoc/parser/pdf/util · high confidence
New RAG prompt templates and Go embedding infrastructure
The system now includes a comprehensive set of new prompt templates for Retrieval-Augmented Generation (RAG) workflows, embedded directly into the Go binary via the \internal/rag/prompts\ package. These templates define behavior for summarizing knowledge (\ask\_summary\), generating precise citations with illustrative ID caveats (\citation\_prompt\), extracting keywords (\keyword\_prompt\), translating across languages (\cross\languages\\*\), filtering metadata via JSON conditions (\meta\_filter\), generating related questions (\related\_question\), and processing visual content from PDFs (\vision\_llm\_describe\_prompt\, \vision\_llm\_figure\_describe\_prompt\). The \generator.go\ module provides the \CitationPrompt\ function to enforce citation rules and append necessary caveats, while \templates.go\ and its tests ensure these assets are reliably loaded without external file dependencies.
internal/rag/prompts · high confidence
New Search application list page with create, rename, and delete capabilities
A new Search page has been introduced to manage search applications. This page provides a list view with search and owner filtering, pagination, and an empty state. Users can create new search applications, rename existing ones, and delete them via a confirmation dialog. The implementation includes the main list component, a card component for individual items, and hooks for data fetching and mutations.
web/src/pages/next-searches · high confidence
New Storybook stories for public UI components
Added Storybook documentation and interactive examples for several public-facing components, including AvatarUpload, ButtonLoading, Calendar, Collapse, ConfirmDeleteDialog, Modal, NodeCollapsible, NumberInput, RAGFlowAvatar, RAGFlowForm, and RAGFlowPagination. These stories allow developers to preview component states, explore usage patterns, and verify behavior in isolation.
web/src/stories · high confidence
New Tavily Search and Extract operator forms
Added new UI forms for the Tavily Search and Tavily Extract agent operators. The Tavily Search form allows users to configure API keys, search queries, depth (Basic/Advanced), topic (News/General), result limits, date ranges, and toggles for including answers, raw content, images, and image descriptions, with support for dynamic include/exclude domain lists. The Tavily Extract form provides fields for API keys, target URLs, extraction depth, and output format (Text/Markdown). Both forms sync their configuration to the canvas node in real-time as users edit them.
web/src/pages/agent/form/tavily-form · high confidence
New Tokenizer configuration form for agent pipelines
A new TokenizerForm component has been added to the agent form interface, allowing users to configure tokenizer-specific settings within their agent pipelines. This form provides controls for selecting search methods (displayed as badges), adjusting the filename embedding weight via a slider, and choosing specific fields for tokenization. It integrates with the existing form validation schema and output display mechanisms, enabling precise control over how text is processed before being passed to downstream operators.
web/src/pages/agent/form/tokenizer-form · high confidence
New UI components for dataset configuration and knowledge management
This update introduces a suite of new React components in the web interface to enhance dataset and knowledge base management. Users can now upload and crop profile pictures with the new AvatarUpload component, which enforces a 4 MB file size limit. The dataset configuration forms now feature dedicated fields for auto-keywords and auto-questions, allowing users to set limits via sliders. Parsing settings have been improved with a ChildrenDelimiterForm that includes a live preview of delimiter splits, and a CompilationTemplateFormField that helps users select or create compilation templates. Additionally, the interface includes a BulkOperateBar for performing actions on multiple files, a BackButton for easier navigation, and a ConfirmDeleteDialog for safer data removal.
web/src/components · high confidence
New accordion-style operator selection dropdown for agent canvas nodes
The agent canvas now uses a new dropdown component to let users add the next operator node. Operators are grouped into five categories—Foundation, Dialog, Flow, Data Manipulation, and Tools—and displayed in an accordion layout that adapts to the current node type and pipeline context. The menu respects upstream restrictions (e.g., parser/chunker limits) and prevents accidental node creation when dragging across menu items by stopping press events. Selecting an operator adds the corresponding node to the canvas and closes the menu.
web/src/pages/agent/canvas/node/dropdown · high confidence
New admin form components for user and role management
Added new React form components in the admin UI to support user and role management workflows. The \change-password-form.tsx\ provides a validated interface for resetting passwords with confirmation. The \email-form.tsx\ handles email creation with validation. The \import-excel-form.tsx\ allows administrators to upload \.xlsx\ files for bulk data import, including file type validation. The \role-form.tsx\ enables creating roles with configurable permissions across different resource types using a tabbed interface. The \user-form.tsx\ supports creating new users with email, password, and optional role assignment (for enterprise features). All forms use \react-hook-form\ with \zod\ validation and follow the existing design system.
web/src/pages/admin/forms · high confidence
New admin panel with role-based access and sandbox management
The admin interface has been rebuilt with a new layout system (root, navigation, and authorized layouts) and a dedicated login page. Administrators can now manage user accounts, view service status with detailed task-executor bar charts, and configure sandbox providers (including local, SSH, and cloud options). Enterprise editions gain access to role-based access control (RBAC) for managing roles and permissions, a registration whitelist for controlled access, and a monitoring dashboard that embeds Prometheus alerts.
web/src/pages/admin · high confidence
New agent debug form with multi-file upload support
The agent debug interface now includes a dynamic form component that renders input fields (text, paragraph, options, integer, boolean, and JSON object) based on the agent's defined parameters. A key addition is the ability to upload files directly within the debug form, allowing users to test agents that accept file inputs. The form handles validation via Zod schemas and manages file uploads with progress tracking and error handling, ensuring that file parameters can be submitted during agent debugging.
web/src/pages/agent/debug-content · high confidence
New agent import dialog with duplicate name validation
A new upload-agent-dialog component has been added to the agents page, allowing users to import agent definitions via JSON DSL files. The dialog includes a form that validates the agent name and specifically handles backend errors indicating that a name already exists, displaying a user-friendly message in the name input field instead of a generic error. The file uploader is configured to accept only JSON files and hides the folder tab for this specific import flow.
web/src/pages/agents/upload-agent-dialog · high confidence
New agent management hooks for creation, logging, and filtering
The agents hooks directory now includes three new React hooks that enhance the agent management experience. useCreateAgentOrPipeline provides the logic for creating new agents or dataflow pipelines via a modal interface. useExportAgentLogToCSV enables users to export agent conversation logs to a UTF-8 CSV file, including full message details and metadata. useSelectFilters supplies the configuration for the agent list filter bar, dynamically fetching and displaying counts for owner, canvas category, and tags.
web/src/pages/agents/hooks · high confidence
New agent share page with parameter input and theme support
A new embedded share page for agents has been added, allowing users to interact with shared agent instances. This page supports dynamic theming via URL parameters and handles locale changes without affecting the host site. It introduces a parameter dialog that appears at the start of a session to collect initial inputs, and it manages file uploads with progress tracking. The interface uses a custom message list and input component, hiding the reset button while a message is being sent.
web/src/pages/agent/share · high confidence
New agents page with templates, logs, and tag management
The agents page has been rebuilt to include a dedicated agent list with card-based display, tag management (add, edit, remove, and filter tags), and a new agent log page with detailed view and CSV export. Users can now create agents from templates or blank, with mandatory binding of datasets and LLM models when templates require them. The page also supports importing agents via JSON files and provides a unified interface for managing agent templates and compilation templates.
web/src/pages/agents · high confidence
New and updated data source configuration forms
The data source settings page now includes configuration forms for several new connectors—Feishu Wiki, Azure DevOps, Bitbucket, Confluence, Sitemap, and Xquik—alongside updated forms for existing sources like Jira, S3, and Seafile. These additions provide users with the ability to connect to these platforms, featuring specific fields such as Feishu Wiki's space and root node tokens, Azure DevOps' custom base URL and organization scoping, Bitbucket's workspace and repository slugs, Confluence's page and space indexing, Sitemap's URL filtering and PDF link following, and Xquik's query parameters. Existing connectors have also been refined: Jira now supports scoped tokens and separate cloud/server authentication, S3 allows custom region values and assume-role authentication, and Seafile offers granular library and directory sync scopes.
web/src/pages/user-setting/data-source/constant · high confidence
New base and tooltip node components for the canvas
The canvas rendering layer now includes a new BaseNode component that provides a consistent visual style (rounded card with selection highlighting) and a new TooltipNode system. This system introduces a context-aware tooltip mechanism where hovering or focusing on a node makes its associated toolbar visible via the TooltipContent component, improving discoverability of node-specific actions.
web/src/components/xyflow · high confidence
New chat channel management UI with WhatsApp QR and agent binding
The user settings page now includes a dedicated Chat Channels section where users can add, edit, and remove external messaging integrations. The interface currently supports Discord, DingTalk, Feishu (Lark), Telegram, QQ Bot, WeCom, and WhatsApp. For WhatsApp specifically, the new Add Channel modal displays a live QR code that users must scan to authenticate the connection, with status polling to confirm when the channel is connected. Additionally, the system now supports binding chat channels to either a standard chat assistant or an agent, allowing users to direct incoming messages to the appropriate backend handler.
web/src/pages/user-setting/chat-channel · high confidence
New chat page with multi-model comparison and batch conversation management
The chat interface has been replaced with a new implementation that supports multiple simultaneous chat boxes (up to three) for model comparison, accessible via a debug toggle. It introduces batch selection and deletion for conversations in the sidebar, along with a dedicated dropdown for individual conversation removal. The chat box now integrates web search capabilities, supporting multiple providers (You.com, Serply, Search1API, Querit, etc.) with validation for required API keys.
web/src/pages/next-chats/chat · high confidence
New chunk card component for data flow results
A new ChunkCard component has been added to the data flow result page, providing a UI for displaying individual data chunks. It renders chunk content with HTML sanitization (converting images to text to prevent tracking), supports image previews via a popover, and includes controls for selecting chunks and toggling their availability status.
web/src/pages/dataflow-result/components/chunk-card · high confidence
New chunk creation and editing modal with tag feature support
A new modal component has been introduced for creating and editing data-flow chunks, providing a unified interface for managing chunk content, keywords, and questions. For tag parsers, users can now assign specific tags and define their importance via a frequency weight (1–10), with logic to automatically fall back to the dataset's own tag list if no external knowledge base IDs are configured. The modal also includes functionality to delete existing chunks and toggle their availability status.
web/src/pages/dataflow-result/components/chunk-creating-modal · high confidence
New chunk processing result views with segmented navigation
The chunk page now features a segmented control allowing users to switch between 'Parsed results', 'Chunk result', and 'Result view' tabs. The 'Parsed results' tab displays a dedicated panel with page-level cards, while the 'Chunk result' tab shows a split view with parsed details on the left and chunked content cards (with an active toggle) on the right. A breadcrumb header enables navigation back to the dataset, and toolbar buttons allow copying or exporting the current view's text.
web/src/pages/chunk · high confidence
New claim visualization and representation selection components
Added a new UI layer for displaying extracted claims and evidence within the chunk representation view. The \claim-list.tsx\ component introduces a \ClaimsPanel\ that renders a scrollable list of claim cards, each showing the claim name, description, and verbatim evidence quotes (with an expandable toggle for long lists), alongside a loading state and close handler. The \representation-select.tsx\ component provides a searchable dropdown (\SelectWithSearch\) that allows users to filter and select from available structure graph templates, displaying both the template name and its kind label.
web/src/pages/chunk/representation/components · high confidence
New context menu for agent canvas nodes
Users can now right-click nodes on the agent canvas to open a context menu that provides options to duplicate or delete the selected node. This change introduces a new \NodeContextMenu\ component and a \useHandleNodeContextMenu\ hook to manage the menu's visibility and positioning, replacing the previous native browser context menu behavior with a custom UI.
web/src/pages/agent/canvas/context-menu · high confidence
New data source detail page with configuration form and log table
Users can now view and edit data source settings via a new detail page in User Settings. This page includes a dynamic configuration form for fields like prune frequency, refresh frequency, and timeout, along with a dedicated log table that displays task status, dataset links, and summary information (including sync/prune counts and error messages) with pagination.
web/src/pages/user-setting/data-source/data-source-detail-page · high confidence
New data source linking UI with auto-scroll and auto-parse controls
The dataset settings page now features a new modal for linking data sources, allowing users to search and select multiple sources via a card-based interface. Once linked, data sources appear in the settings list with a visible toggle to enable or disable auto-parsing, along with buttons to view details, rebuild, or unlink. Additionally, the interface automatically scrolls to the newly linked data source to improve visibility after selection.
web/src/pages/dataset/setting/components · high confidence
New dataset compilation view with multi-mode knowledge display
A new dataset compilation page has been added, allowing users to view and interact with compiled knowledge in several modes: LLM Wiki, Skills, Navigation Tree, Graph, Mind Map, and Timeline. The UI features a mode selector, a left panel for navigation (tree or skills), and a main content area for the selected view. It includes progress tracking for background compilation tasks, showing inflight/backlog counts and error diagnostics, and provides version history and update logs for wiki artifacts.
web/src/pages/dataset/compilation · high confidence
New dataset ingestion overview page with file and dataset log tabs
A new dataset overview page has been added to the application, providing a consolidated view of dataset ingestion status. This page features a segmented control to switch between 'File Logs' and 'Dataset Logs', allowing users to monitor specific ingestion activities. It displays key statistics including total files, download counts (running, success, failed), and processing counts (running, success, failed). Users can search, filter by status, and paginate through log entries. The page includes a detailed table showing file names, sources, pipeline titles, start dates, task types, and operation statuses, with support for sorting and viewing process logs. The implementation includes React hooks for data fetching with automatic polling when active ingestion logs are detected, and utility functions for building overview statistics and managing query keys.
web/src/pages/dataset/dataset-overview · high confidence
New dataset page layout and ingestion log visibility
The dataset section now features a dedicated wrapper page with a sidebar and context provider, along with a reusable title component. Ingestion logs are now visible in real-time via a polling hook that fetches run-scoped progress events and stops once the process reaches a terminal state. Additionally, the process log modal now displays timestamps for each event and provides a specific hint when a file is missing from object storage.
web/src/pages/dataset · high confidence
New dataset page with Go ingestion status and parser gap handling
The dataset page has been rebuilt to support the Go ingestion backend, introducing a new document table and action cell that display real-time parsing statuses (including QUEUED, RUNNING, STOPPING, and terminal states) and provide inline controls for renaming, downloading, and deleting files. The update adds a dedicated parser-change dialog for configuring document pipelines, a re-parse confirmation dialog that safely handles existing chunks, and a parser-gap validation system that prevents uploads or parses when required audio/video models are missing or file types are unsupported, surfacing clear error messages to the user.
web/src/pages/dataset/dataset · high confidence
New dataset sidebar with navigation and metadata display
A new sidebar component has been added to the dataset pages, providing a consistent navigation menu and metadata display for each knowledge base. The sidebar displays the dataset's avatar, name, file count, total size, and creation date in the header. It includes navigation links to Files, Testing, Overview, Configuration, and a new Artifacts section (linked to the compilation route). The component uses React Router for navigation and supports internationalization for all labels.
web/src/pages/dataset/sidebar · high confidence
New dedicated page for shared search results
A new \ShareSearchPage\ component has been added at \web/src/pages/next-search/share/index.tsx\ to handle the display of shared search links. This page retrieves the specific search configuration and tenant details for the shared link, and respects the locale specified in the share parameters by dynamically changing the application language. It conditionally renders the avatar and name if configured, and manages the transition between the initial search input state (\SearchHome\) and the results view (\SearchingPage\).
web/src/pages/next-search/share · high confidence
New diagnostic tools for search coverage and Markdown parsing
Added two new Go-based diagnostic tools to the \tools/\ directory. \covsweep\ performs full-document search coverage analysis by comparing source files against Elasticsearch chunks to identify missing or fused text passages, helping verify that ingestion pipelines preserve content integrity. \mdprobe\ runs the Markdown parser on a single file and reports the parsed items, allowing users to quickly determine if text loss occurs at the parsing stage versus downstream indexing.
tools/mdprobe · high confidence
New document preview modal for search results
A new document preview modal has been added to the next-search page, allowing users to view the content of cited documents directly from search results. The modal handles various file types by generating preview URLs and highlighting specific text chunks within the document. It also supports opening web pages cited in the results by linking to them in a new tab.
web/src/pages/next-search/document-preview-modal · high confidence
New editor plugins for floating toolbars, table editing, and wiki navigation
The Markdown editor now includes several new capabilities: a floating selection toolbar that appears when text is selected, providing formatting options (bold, italic, code, strikethrough, link, quote, headings) and intelligently adjusting dropdown direction based on available screen space; a table editing toolbar that appears when the cursor is inside a table, allowing users to add/remove rows and columns and adjust text alignment; a table-of-contents outline feature in the main toolbar for quick navigation between headings; and wiki link click handling that enables navigation to new Markdown files when wiki links are clicked. Additionally, a crash fix prevents the editor from hanging or crashing when importing markdown tables with extremely large rows by capping row size at 5,000 bytes.
web/src/lib/editor/plugins · high confidence
New email sending form with SMTP configuration
A new form component has been added to the agent builder, allowing users to configure and send emails. The form collects recipient details (to, cc), subject, and content, alongside SMTP server settings including server address, port, username, password, and sender name. Input fields are validated using Zod schemas, and the interface supports translation for internationalization.
web/src/pages/agent/form/email-form · high confidence
New file error display component
A new FileError component has been added to the document viewer to handle and display error states when a file fails to parse. This component renders a centered, styled error message box using the application's error theme, displaying either a custom child message or a default localized 'fileError' text, ensuring users receive clear feedback on parsing failures.
web/src/pages/document-viewer/file-error · high confidence
New file upload dialog with table column configuration and folder support
Introduces the FileUploadDialog component, enabling users to upload files and folders (preserving relative paths) and configure table parsing. The dialog allows automatic or manual selection of column roles (indexing, metadata, or both) for table files and includes an option to parse documents immediately upon creation.
web/src/components/file-upload-dialog · high confidence
New form components for agent configuration and structured output selection
This change introduces a suite of new UI components within the agent form to support advanced configuration and structured data handling. Key additions include \ApiKeyField\ and \UserIdFormField\ for sensitive input fields, \DescriptionField\ for agent metadata, and \DynamicStringForm\ for managing list-based inputs. The update also implements \QueryVariable\ and \QueryVariableList\ to allow users to select and manage query variables, featuring a \GroupedSelectWithSecondaryMenu\ that enables nested selection of structured output properties. Additionally, \SchemaDialog\ and \SchemaPanel\ provide visual editing and viewing capabilities for JSON schemas, while \Output\ and \JsonViewer\ handle the display of agent outputs.
web/src/pages/agent/form/components · high confidence
New form for configuring variable aggregation operators
Users can now configure variable aggregation operators via a dedicated form that supports dynamic grouping. The form allows adding and removing groups, each with a unique name (restricted to letters, numbers, and underscores) and a list of variables. It automatically tracks and propagates the data type of the first variable in each group to the operator's output, ensuring consistent type handling in the agent graph.
web/src/pages/agent/form/variable-aggregator-form · high confidence
New hooks for dataset compilation navigation, artifact management, and wiki editing
This change introduces a suite of new React hooks in the dataset compilation area to support the new data compilation layer. Specifically, it adds \useCompilationNav\ to manage the dataset navigation tree (including keyword filtering, child loading, and structure graph fetching), \useCompilationArtifact\ for selecting artifacts, \useCompilationSkill\ for skill tree/page data, and \useCreateDirectory\ for directory creation UI. It also adds \useWikiVersion\ for version selection, \useWikiLinkNavigation\ for managing a wiki page history stack, \useWikiEditor\ for tracking content changes, \useCommitArtifact\ for handling wiki commits with validation, and \useRunEndEffect\ to trigger data refreshes after compilation runs complete. These hooks collectively provide the state management and logic required for the new compilation interface components.
web/src/pages/dataset/compilation/hooks · high confidence
New icon assets added to the web application
A new file, \web/src/assets/icon/next-icon.tsx\, has been added to the project, introducing a collection of SVG-based icon components (including Api, Team, Profile, Password, LogOut, ModelProvider, Prompt, Wikipedia, and Keyword) for use in the user interface. This change also includes standard Apache 2.0 license headers across the new assets.
web/src/assets/icon · high confidence
New image component with caching and cache-busting support
The \web/src/components/image\ module now provides a dedicated \Image\ component and associated hooks (\useDocumentImageUrl\, \buildDocumentImageUrl\) that handle document image retrieval with an in-memory cache and optional cache-busting via a \t\ parameter. This ensures that images are fetched with proper authorization headers for internal API URLs, cached to avoid redundant network requests, and can be refreshed when the cache-busting key changes or when explicitly evicted (e.g., after an image update).
web/src/components/image · high confidence
New internal utility library for Go ingestion and processing
The \internal/utility\ package introduces a suite of Go components supporting the ingestion pipeline and document processing. This includes a \BestEffort\ helper for non-blocking error logging, a thread-safe \EmbeddingLRU\ cache, and file-type classification logic (\GetFileType\, \FilenameType\) that handles video, image, and document extensions. It also provides a stdlib-only PNG captcha renderer to prevent answer leakage, conversion utilities (including \JSONFloat64\ to preserve decimal formatting), and a Markdown-to-JSON parser for mind-map extraction.
internal/utility · high confidence
New iteration start configuration form
A new UI component has been added to the agent form to configure iteration start parameters. This component renders a list of output fields derived from the system's initial iteration start values, allowing users to define the starting state for loop operations.
web/src/pages/agent/form/iteration-start-from · high confidence
New knowledge graph visualization and deletion flow
This change introduces a new knowledge graph view for datasets, featuring a force-directed graph visualization powered by @antv/g6 that supports node grouping, tooltips, and dark theme styling. It also adds the ability to delete a knowledge graph via a confirmation dialog, which automatically navigates the user back to the dataset page upon successful deletion.
web/src/pages/dataset/knowledge-graph · high confidence
New languages and UI features added to localization files
The localization files for Arabic (ar.ts), Azerbaijani (az.ts), and Bulgarian (bg.ts) have been added to the web application. Additionally, the English (en.ts) and French (fr.ts) locale files have been updated to include new translation keys for the 'Skills' feature, which allows users to create and manage skill spaces, and the 'Memories' feature, which provides memory management options.
web/src/locales · high confidence
New list filter bar component with nested filtering and selection persistence
Introduces a new \list-filter-bar\ component that provides a unified interface for search and multi-select filtering. The component supports nested filter hierarchies, allowing users to drill down through child items, and includes a search input to filter available options. A key behavioral improvement is the persistence of unsubmitted selections: when filter options refresh (e.g., due to document parsing updates), user selections made within the popover are preserved rather than reset. The implementation includes a \FilterPopover\ for the UI, a \FilterField\ for rendering nested checkboxes, and a \useHandleFilterSubmit\ hook that manages state and resets pagination upon submission.
web/src/components/list-filter-bar · high confidence
New local development and CLI installation tooling
The repository now includes a suite of scripts in tools/scripts to streamline local development and CLI distribution. A new download\_dev\_deps.py script automates fetching the native static libraries (pdfium, pdf\_oxide, office\_oxide, and onnxruntime v1.29.0-patch2) and runtime assets required by the Go server, eliminating the need for separate dependency downloads. To support the Go DeepDoc backend, a convert\_deepdoc\_model.py script allows users to upgrade ONNX models to the opset supported by the reduced ONNX Runtime library. For CLI distribution, install.sh and install.ps1 provide one-line installation of the ragflow-cli binary for Linux, macOS, and Windows, including platform detection and SHA256 verification. Additionally, a fetch\_deepdoc\_testdata.sh script manages external test data via sparse clones, and a gen-proto.sh script automates protobuf code generation. Also in this area: 1 commit (1 fix) fixing tools; 1 commit (0 fixes) modifying tools.
tools/scripts · medium confidence · unverified
New login page with animated background and flip-card UI
The login interface has been replaced with a new implementation featuring a dynamic SVG background with animated glow and highlight effects, and a 3D flip-card component that transitions between the login and registration forms. The new form uses react-hook-form and Zod for validation, supports a 'disable\_password\_login' configuration to enforce SSO-only access, and correctly hides the sign-up toggle when registration is disabled or loading.
web/src/pages/login-next · high confidence
New markdown content component with artifact support and RTL styling
The chat interface now uses a new \next-markdown-content\ component to render messages. This update introduces support for displaying downloadable artifacts (files and images) directly within the chat stream, including a dedicated download button for artifact images. It also adds specific styling for collapsible 'thinking' and 'retrieving' blocks, ensures proper list rendering, and enables Right-to-Left (RTL) text alignment for supported languages.
web/src/components/next-markdown-content · high confidence
New markdown rendering components and UI utilities
This change introduces a new set of components for the markdown content area: a reusable \LoadingDots\ animation, a \FallbackComponent\ for route errors, and an \ImageCarousel\ that groups consecutive image references into a swipable view. The main \MarkdownContent\ component is updated to support RTL text direction, disable LaTeX math rendering for user messages, and sanitize the parsed markdown tree to address a security vulnerability ([GHSA redacted]).
web/src/components/markdown-content · high confidence
New memory configuration page with validation and advanced settings
The memory settings interface has been replaced with a new configuration page that includes basic info (name, avatar, description with 512-character limit), model selection (embedding and LLM), and advanced settings (permissions, storage type, forgetting policy, temperature, system/user prompts). The page validates that selected models still exist in the supplier list, showing errors if a model has been deleted, and revalidates persisted values once the model list loads.
web/src/pages/memory/memory-setting · high confidence
New memory message management interface
Introduces a dedicated page for viewing and managing memory messages, featuring a filterable and paginated table. Users can now view message details, toggle message status via a switch, delete messages, and view extraction task progress logs. The interface supports filtering by agent and searching within the message list.
web/src/pages/memory/memory-message · high confidence
New memory settings hook for fetching base configuration
A new \use-memory-setting.ts\ hook has been added to the memory page area, providing a \useFetchMemoryBaseConfiguration\ function. This hook retrieves the memory base configuration by combining the URL parameter \id\ with the \id\ search query parameter, utilizing React Query for data fetching and caching. It exposes the fetched memory data, loading state, and search/pagination utilities to components within the memory settings interface.
web/src/pages/memory/hooks · high confidence
New memory sidebar component with avatar and navigation
The memory sidebar now displays the memory's avatar, name, description, and creation time using the RAGFlowAvatar component, and provides navigation buttons to switch between the 'Messages' and 'Configuration' views within the memory context.
web/src/pages/memory/sidebar · high confidence
New message-item component with file, image, and audio support
The \web/src/components/next-message-item\ directory now contains a complete set of new components and utilities for rendering chat messages. This includes a main \MessageItem\ component that handles message content, RTL text direction, and theme-aware styling, alongside a \group-button\ component for message actions like copy, feedback, and file download. The update introduces a \ReferenceImageList\ that displays referenced images in a carousel with zoom capabilities, and an \UploadedMessageFiles\ component that allows users to view and play audio attachments directly from the chat interface. Supporting utilities for MIME type detection and file classification have also been added, along with comprehensive test coverage for these new features.
web/src/components/next-message-item · high confidence
New metadata filter components for manual and semi-automatic filtering
This change introduces a new metadata-filter component suite that enables users to apply metadata filters in both manual and semi-automatic modes. The manual mode allows users to define specific key-operator-value conditions (e.g., 'title equals example'), while the semi-automatic mode lets users select metadata keys and operators without specifying exact values, allowing the system to handle the matching. The components also support logical operators (AND/OR) to combine multiple conditions. Additionally, an agent-specific variant is provided that includes upstream canvas variables as filterable options, enabling dynamic filtering based on agent inputs. The implementation includes a Zod schema for validation and utility functions to flatten variable options for the agent context.
web/src/components/metadata-filter · high confidence
New model management section in provider instance settings
The provider instance card now includes a dedicated models section that lets users browse the upstream model catalog, add or remove models, and verify individual models. The UI displays model-type badges (collapsing overflow into a tooltip) and provides per-model verify buttons with status indicators. For new (draft) instances, newly listed models are automatically merged into the local list while respecting user removals, and the section defers catalog fetching until required credentials (such as API keys or base URLs) are available.
web/src/pages/user-setting/setting-model/instance-card/models-section · high confidence
New model provider settings layout with sidebar navigation and batch save
The model provider settings page now uses a new layout structure featuring a left sidebar for selecting between system default models and specific provider instances, and a right pane with a sticky header bar. Users can now search and filter available providers, with a visual indicator for those that already have configured instances. The system default model settings have been reorganized into a dedicated view with improved field handling and persistence. Additionally, provider configuration changes are now saved using a batch 'Save all' button in the provider header, replacing the previous per-card auto-save behavior, which allows users to review and commit multiple changes at once.
web/src/pages/user-setting/setting-model/layout · high confidence
New multi-source skill registry integration
The CLI now supports fetching and managing skills from multiple external registries. Users can reference skills from GitHub repositories, the ClawHub registry, and the skills.sh directory using specific URI schemes (e.g., \github.com/...\, \clawhub://...\, \skill://...\) or standard paths. The system automatically resolves these references, downloads the skill bundles, and applies trust levels based on the source (e.g., distinguishing between trusted official repositories and community sources).
_internal/cli/filesystem/skill\hub · high confidence
New next-search page components and embed functionality
The next-search page now includes a new embed dialog (embed-app-modal.tsx) that allows users to generate iframe code and open the search in a new tab, with options to hide the avatar and select the locale. The page also features a new expandable content component (expandable-content.tsx) that handles overflow and provides 'View More/Less' buttons for long content. Additionally, the search settings now use a new LLM setting defaults module (llm-setting-defaults.ts) to ensure unset parameters fall back to backend defaults rather than 0, preventing issues with nucleus sampling. The search home page (search-home.tsx) has been updated to include a RAGFlow logo with an embed icon, and the search view (search-view.tsx) now integrates these new components for a more cohesive user experience.
web/src/pages/next-search · high confidence
New originui component library for forms, selection, and data visualization
A new set of UI components has been added to the \originui\ directory to standardize the interface. This includes an \Input\ component with optional icons, a \NumberInput\ that enforces precision, step grids, and integer modes while preserving caller blur handlers, and a \PasswordInput\ with visibility toggling. Selection is handled by \SelectWithSearch\, which supports searching by label, custom values, and missing-value rendering, replacing previous selection implementations. Data presentation is enhanced with a \TimeRangePicker\ offering quick-select date ranges and a \Timeline\ component for displaying step-by-step progress. Additionally, \UnderlineTabs\ provide a tabbed interface with an underline indicator style.
web/src/components/originui · high confidence
New per-chunk result cache to optimize ingestion resume performance
A new \chunkcache\ package has been introduced to memoize expensive per-chunk model calls (such as extraction, tokenization, and tagging) during the ingestion pipeline. By storing these results in Kvrocks with a 7-day TTL and tracking them via task-scoped manifests, the system ensures that resumed tasks can reuse cached outputs instead of re-executing costly external model requests. The implementation includes robust cleanup logic to purge cache entries once chunks are durably indexed, while gracefully degrading to a no-op if the cache store is unavailable.
internal/ingestion/chunkcache · high confidence
New pipeline log sheet with dataflow timeline view
Users can now view detailed pipeline execution logs in a new sheet component that displays a visual timeline of dataflow steps. This view shows the progress percentage for each node, timestamps, elapsed times, and log messages (including error highlighting), along with an option to export the log data as JSON or navigate to the full dataflow result page.
web/src/pages/agent/pipeline-log-sheet · high confidence
New pre-commit hooks for file validation and web tooling
Added a new Python-based pre-commit hook script (\check\_files.py\) that validates staged files for JSON and YAML syntax, trailing whitespace, mixed line endings, missing trailing newlines, merge conflict markers, broken symlinks, and case conflicts, while safely skipping binary files. Also added a shell script (\web\_tools.sh\) to resolve and cache the specific versions of \oxlint\ and \oxfmt\ pinned in \web/package-lock.json\, ensuring consistent linting and formatting for web files in CI and local development.
tools/hooks · high confidence
New provider field-config schema and lookup system
The provider configuration system in the user settings has been restructured to use a centralized schema-driven approach. A new \getProviderConfig\ lookup function now resolves settings for each LLM factory by checking a specific \ProviderConfigMap\ (covering providers like Azure OpenAI, VolcEngine, Google Cloud, and MinerU), a \LocalLlmConfigs\ map (for local/compatible providers like Ollama, FunASR, and MWS), and finally falling back to a \GenericApiKeyConfig\ for any unrecognized providers. This ensures that provider-specific fields, such as the required \base\_url\ for FunASR or the \group\_id\ for certain API keys, are correctly rendered and submitted, while also standardizing credential handling across the model provider forms. Also in this area: 4 commits (2 fixes) modifying web; Google Drive sync flexibility and MinerU 4.0 support; Display built-in pipelines as read-only items in the agents list.
web/src/pages/user-setting/setting-model/provider-schema/field-config · medium confidence · unverified
New provider instance cards and model management UI
The instance-card module now includes dedicated UI components for the AIMLAPI and AWS Bedrock providers, along with a generic dynamic form dialog for adding custom models. The AIMLAPI card supports OAuth-based API key retrieval, while the Bedrock card handles region selection and multiple authentication modes (API key, IAM role, assume role). A new 'Available Models' sidebar allows users to browse and filter providers by model type (LLM, VLM, ASR, etc.), and a reusable dialog component enables the creation of custom model entries with provider-specific extra fields.
web/src/pages/user-setting/setting-model/instance-card · high confidence
New provider-schema hooks for model picker and form configuration
This change introduces a new set of React hooks in the provider-schema module to power the model provider configuration UI. \useListModelsPicker\ manages the state for the 'List Models' picker, fetching available models from the provider catalog, handling selection toggles (including an all/none toggle with reentrancy guards), and deriving the \is\_tools\ capability from model features. \useProviderFields\ builds the dynamic form field configurations, resolving per-provider text (placeholders/tooltips), handling visibility logic (such as showing the base URL only when the catalog provides options), and mapping base URL selections to region keys. \useProviderModalActions\ orchestrates the verify and submit flows, injecting the derived region and model info into the payload. A test suite for \useProviderFields\ verifies that the catalog's \url\_hint\ is correctly used as the endpoint placeholder.
web/src/pages/user-setting/setting-model/provider-schema/hooks · high confidence
New rename dialog component with optional slash validation
A new RenameDialog component has been introduced to handle renaming operations, featuring a form that accepts an initial name and a callback for the new name. The component includes a new 'forbidSlash' prop; when enabled, it prevents users from entering names containing forward slashes ('/'), addressing a validation requirement for file and folder names. The UI uses a standard dialog layout with a loading state on the save button.
web/src/components/rename-dialog · high confidence
New repository configuration files for build hygiene and AI agent workflows
The repository now includes several new configuration files to improve build efficiency and guide development workflows. A \.dockerignore\ file has been added to reduce the Docker build context by excluding local virtual environments, build artifacts, logs, and IDE configurations, which speeds up image builds. A \.rooignore\ file is introduced to reduce indexing noise and token waste for AI coding assistants by excluding generated files and caches. Additionally, a \.trivyignore\ file is added to suppress security scanning noise for static assets like images and stylesheets, and a \.gitattributes\ file ensures correct line endings for shell scripts and marks \.msg\ files as binary to prevent corruption. Finally, an \AGENTS.md\ file provides a comprehensive operating guide for the codebase, detailing the current stack, code layout, Go-specific rules, database schema reconciliation, and test tier classifications.
(repo-wide) · high confidence
New request interface definitions for web API communication
This change introduces a new set of TypeScript interfaces in the \web/src/interfaces/request\ directory to define the structure of request payloads for various web API endpoints. These interfaces cover agent debugging, chat feedback, compilation templates, document parsing, file management, flow debugging, knowledge base operations, LLM provider management, MCP server testing, and system configuration. This provides a standardized and type-safe way for the frontend to communicate with the backend for these specific features.
web/src/interfaces/request · high confidence
New reusable component for editable avatar, name, and description fields
The \web/src/components/avatar-name-description\ directory now contains a new \AvatarNameDescription\ component that consolidates the avatar upload, name input, and description textarea into a single, reusable UI block. This component introduces \EditableField\ for single-line inputs and \EditableTextarea\ for multi-line descriptions, both featuring an inline edit mode triggered by clicking the field or the pencil icon. The implementation handles form integration via \react-hook-form\, manages editing state (including auto-focus and cursor positioning), and enforces required-field validation by restoring the previous value if a required field is cleared.
web/src/components/avatar-name-description · high confidence
New reusable empty-state UI components
The application now includes a standardized set of UI components for displaying empty states across various pages. This change introduces the \Empty\ component for simple no-data messages, the \EmptyCard\ for structured content blocks, and the \EmptyAppCard\ which provides pre-configured templates for specific sections like Agents, Datasets, Chat, Search, Memory, and Skills. These components support internationalization, dark/light theme adaptation, and customizable sizes, ensuring a consistent user experience when content is missing.
web/src/components/empty · high confidence
New scripts for tokenizer validation, coverage reporting, and safe ingestor restarts
Added \scripts/gen\_tokenizer\_oracle.py\ to generate reference tokenization fixtures for verifying Go counter implementations against external libraries (tiktoken, sentencepiece, HuggingFace tokenizers); added \scripts/tokenizer\_coverage.py\ to report how many embedding models in the catalog are covered by implemented tokenizer counters versus those requiring runtime calibration; and added \scripts/restart\_ingestor.sh\ to safely restart the ingestor service by checking for in-flight NATS tasks before proceeding, preventing data loss during redeployment.
scripts · high confidence
New service layer for admin, agent, and knowledge operations
The web frontend now includes a comprehensive set of new service modules in \web/src/services\ to handle API communication for core features. This adds dedicated service files for admin operations (user, role, and service management), agent management (creation, versioning, and tracing), knowledge base and document handling (ingestion, structure, and retrieval), and data sources (including web auth flows for Google Drive, Gmail, and Box). It also introduces adapters to map backend ingestion statuses to frontend document filters and a dedicated service for chat completion streaming using native fetch to support SSE.
web/src/services · high confidence
New shadcn/ui-based component library replaces legacy UI elements
The web interface now uses a comprehensive set of new UI components (such as Accordion, AlertDialog, Avatar, Badge, Breadcrumb, Button, Calendar, Card, Checkbox, Command, and more) built on top of shadcn/ui and Radix primitives. This provides a consistent, accessible, and themeable design system across the application, replacing previous ad-hoc or third-party UI implementations with a unified, modern component architecture.
web/src/components/ui · high confidence
New shared chat page with embedded UI and external integration support
A new shared chat page has been introduced at web/src/pages/next-chats/share, enabling users to view and interact with agent conversations via external links. This page renders the chat within an EmbedContainer, supporting dynamic theming and locale synchronization from URL parameters. It features a NextMessageInput component configured for external uploads and web search, displays message references and PDF documents via a PdfSheet drawer, and includes specific UI adjustments such as hiding the reset button during loading and disabling the send button when errors occur.
web/src/pages/next-chats/share · high confidence
New shell examples for RAGFlow API usage
Added several cURL-based shell scripts in the example/http directory to demonstrate how to interact with the RAGFlow API. These examples cover creating and managing chat assistants and sessions, performing document chunk management (add, list, update, delete), handling document lifecycles including upload and parsing status polling, executing semantic and keyword retrieval, and listing datasets with optional filtering and sorting.
example/http · high confidence
New single-node debug sheet with CodeExec output grouping
A new single-debug-sheet component has been added to display debug results for individual agent nodes. It introduces a specialized layout for CodeExec nodes that groups debug outputs into business outputs, expected/actual types, and system outputs (such as errors and attachments), while other node types fall back to a generic JSON/text view. The sheet includes copy-to-clipboard support, error highlighting, and scrollable content areas to ensure large debug payloads are fully visible.
web/src/pages/agent/form-sheet/single-debug-sheet · high confidence
New streaming store and driver for single-chat conversations
A new module-level streaming infrastructure has been added to the single-chat view to ensure that Server-Sent Events (SSE) streams survive page navigation and unmounting. This includes a Zustand-based store (\store.ts\) that maintains session state independently of the React component lifecycle, a dedicated driver (\run-stream.ts\) that handles chunked responses and error detection, and supporting utilities for message merging and file attachment preservation. The implementation introduces configurable constants for performance tuning, such as a 60ms flush interval to coalesce rapid SSE chunks and a cache limit of 10 idle sessions to manage memory usage.
web/src/pages/next-chats/chat-stream · high confidence
New tabbed interface for pipeline operator configuration
A new set of components in the pipeline-operator-tabs directory introduces a tabbed UI for configuring pipeline operators (parsers, chunkers, extractors, compilers, and tokenizers). This interface renders operator-specific forms within tabs, synchronizes their values with the outer document pipeline form to prevent data loss when switching tabs, and mirrors validation errors from the outer form onto the individual operator fields. It also handles validation of parser configurations against the appropriate operator schemas, ensuring that required fields are enforced consistently across the pipeline.
web/src/components/pipeline-operator-tabs · high confidence
New token chunker configuration form in agent pipeline
A new form component for configuring the token chunker operator has been added to the agent pipeline. It allows users to set the chunk token size and overlap percentage via integer-restricted sliders, and manage delimiter lists (including child delimiters) with a preview feature. The form supports two modes: 'delimiter' for custom delimiters and 'one' for a single method, and adapts its fields based on whether it is a general or specific token chunker.
web/src/pages/agent/form/token-chunker-form · high confidence
New tool configuration forms and unified agent tool wiring
The agent form now includes dedicated configuration UIs for a wide range of search and data tools, including Querit, ArXiv, BGPT, Bing, Crawler, DuckDuckGo, Email, ExeSQL, GitHub, Google, Google Scholar, Keenable, PubMed, SearXNG, Search1API, Sofya, Tavily, WenCai, Wikipedia, Yahoo Finance, and You.com. These forms are registered in a central configuration map and rendered via a unified tool-form component that dynamically selects the correct form based on the selected tool. The Querit tool form specifically supports dynamic lists for site/country/language filtering, numeric inputs for result counts, and time-range constraints, with validation and persistence logic ensuring data integrity. MCP tool forms allow selecting specific tools from an MCP server, with changes synchronized to the agent node.
web/src/pages/agent/form/tool-form · high confidence
PDF layout parser rewritten in Go with improved column detection and cross-page handling
The PDF layout parser in the deepdoc module has been implemented in Go, replacing the previous Python-based logic. This update introduces a hybrid gap-and-KMeans column detection algorithm that better distinguishes between text columns and table cells, and correctly handles edge cases like title-bridged or gutter-less double columns. It also adds robust support for cross-page text boxes, ensuring accurate position tags and reading order for content spanning multiple pages, while fixing word boundary detection for non-Latin scripts like Cyrillic and CJK.
internal/deepdoc/parser/pdf/layout · high confidence
Per-chunk result caching and built-in pipeline registry for ingestion
The ingestion pipeline now caches expensive per-chunk model results (extractor, tokenizer) in Kvrocks with a 7-day TTL, so resumed tasks reuse cached values instead of re-calling external models. A new built-in pipeline registry loads embedded JSON templates, enforces a canonical ordering, and maps legacy parser IDs (e.g., 'naive') to current templates via aliases. Component parameters are extracted from DSLs for validation and UI forms, while canvas DSL envelopes are unwrapped and parser page caps are injected. Cleanup routines purge task-scoped cache entries once chunks are persisted, and tests verify the cache behavior, registry resolution, and parameter extraction. Also in this area: 17 commits (8 fixes) modifying internal; Improved agent error readability and parallel resume reliability; Admin heartbeat API returns structured data and user creation no longer auto-provisions LLM configs.
internal/ingestion/pipeline · medium confidence · unverified
Port agent canvas engine and runtime to Go
The agent canvas engine, including the DSL v2 runtime, 22 component types, and 13 API endpoints, has been ported from Python to Go. This change introduces the \AgentHandler\ in \internal/handler\ to manage agent canvases, component input forms, debug execution, and file attachments. It also adds OpenAI-compatible chat completion streaming support for agent runs, allowing existing OpenAI clients to interact with the agent system. The port includes the full agent file download and preview handlers, ensuring parity with the Python implementation for attachment handling and streaming.
internal/handler · high confidence
Python SDK introduces dedicated modules for Agents, Chats, and Sessions
The Python SDK now includes a new \modules\ package that provides dedicated classes for managing Agents, Chats, and Sessions. Users can now create and list agent sessions, manage chat sessions, and interact with agent-specific features like streaming responses and variable inputs. The SDK also adds support for memory management, allowing users to list, forget, and update message statuses within memory contexts. These modules work alongside existing dataset and document management capabilities to provide a more comprehensive interface for interacting with the RAGFlow platform.
_sdk/python/ragflow\sdk/modules · high confidence
Python SDK: New test and demo scripts with uv-based dependency management
The Python SDK now includes a \hello\_ragflow.py\ demo script that prints the SDK version, and a \test.py\ interactive script for testing agent features via command-line arguments or environment variables (e.g., \RAGFLOW\_API\_KEY\, \RAGFLOW\_AGENT\_ID\). To support these scripts and the SDK itself, the project has switched from Poetry to \uv\ for dependency management, introducing a \uv.lock\ file that pins dependencies such as \lxml\ 6.1.0, \pillow\ 12.0.0, and \urllib3\ \>=2.7.0.
sdk/python · high confidence
Redesigned home page with segmented application lists and dedicated dataset section
The home page has been rebuilt to feature a new layout: a welcome banner at the top, followed by a dedicated Datasets section and a segmented Applications section. The Applications section uses a segmented control to switch between Chats, Searches, Agents, and Memories, displaying the first 10 items of each type in a card list with options to rename and navigate. A 'See All' card is shown when items exist, and an empty state is displayed when lists are empty. The Datasets section similarly lists the first 6 knowledge bases with rename capabilities and an empty state.
web/src/pages/home · high confidence
Retrieval form now supports sourcing from Memories and validates required bindings
The retrieval operator form has been updated to allow users to select whether to retrieve content from Knowledge Bases or Memories via a new 'Retrieval From' toggle. When Memories are selected, the form displays a Memories selector and a User ID field; when Knowledge Bases are selected, it shows the Knowledge Base selector. The form now enforces that at least one dataset or memory is selected before saving, preventing runtime errors. Additionally, legacy DSLs using 'kb\_ids' are automatically migrated to 'dataset\_ids' on load to ensure compatibility.
web/src/pages/agent/form/retrieval-form · high confidence
Support for hierarchical Wiki topic navigation
The wiki left panel now supports navigating topics in a hierarchy. Users can drill down into sub-topics, and the panel will display child topics or artifacts based on the selected path, with a 'back to topics' option to return to the root view.
web/src/pages/dataset/compilation/wiki-left-panel/hooks · high confidence
Unified document viewer with expanded file type support
The document viewer page has been refactored to serve as a central hub for previewing a wider variety of file formats. It now natively supports EPUB, MDX, HTML, and CSV files in addition to existing PDF, image, Markdown, text, Excel, DOCX, and PPT/PPTX previews. The implementation uses a unified routing logic based on file extension to dispatch to specific previewer components, ensuring consistent rendering across all supported document types.
web/src/pages/document-viewer · high confidence
Unified search request and result types for the Go engine
The Go-based document engine now uses a shared \types\ package to define standardized structures for search operations. This introduces \SearchRequest\ and \SearchResult\ types that unify parameters such as index names, knowledge base filters, pagination, and hybrid match expressions (text, dense vector, and fusion) across different storage backends. It also adds \SearchMetadataRequest\ and \SearchMetadataResult\ for metadata-specific queries, along with \OrderByExpr\ for sorting and specific expression types like \MatchTextExpr\, \MatchDenseExpr\, and \FusionExpr\. These types provide a consistent interface for search logic and include debugging utilities to log request details.
internal/engine/types · high confidence
Webhook configuration UI for the Begin node
The Begin node form now includes a dedicated Webhook configuration section, allowing users to define security settings (Token, Basic, JWT, or anonymous access with IP whitelisting), set rate limits and maximum body sizes, and specify HTTP methods. Users can also define the request schema (query parameters, headers, and body with support for file uploads in multipart forms) and configure the response behavior, including execution mode and status codes.
web/src/pages/agent/form/begin-form/webhook · high confidence
Security
HighLightMarkdown component sanitizes content after LaTeX preprocessing
The HighLightMarkdown component now applies sanitization to the markdown content after the LaTeX preprocessing step, ensuring that entity-encoded HTML payloads (such as \<img\> or \<script\> tags) are properly stripped by DOMPurify before rendering. This change fixes a potential cross-site scripting (XSS) vulnerability where malicious HTML could bypass sanitization if it was encoded as entities and decoded by the LaTeX preprocessor. The component also integrates with the SafeImg component for image sanitization and supports both light and dark themes for syntax highlighting.
web/src/components/highlight-markdown · high confidence
Sanitize Rerun Modal Content to Prevent Stored XSS
The RerunButton component now sanitizes the modal content displayed when confirming a pipeline rerun. Previously, the step title was interpolated directly into the modal's HTML via dangerouslySetInnerHTML, which could allow stored XSS attacks if a step name contained malicious scripts. The component now uses DOMPurify to strip unsafe HTML (such as script tags and event handlers) from the step title before rendering it in the confirmation modal, ensuring user safety while preserving safe formatting.
web/src/pages/dataflow-result/components/rerun-button · high confidence
Security fix for buffer overflow in OpenCC text dictionary parsing
The OpenCC C binding in internal/binding/cpp/opencc now includes a buffer-length check in dictionary/text.c to prevent CWE-120 buffer overflows. The code detects when a line read from a text dictionary exceeds the buffer size and drains the remainder, ensuring that malformed or excessively long entries do not corrupt memory or cause crashes during conversion.
internal/binding/cpp/opencc · high confidence
Architecture
Extract admin server logic into dedicated package
The admin HTTP server startup and lifecycle management code has been moved from the main package into a new internal/server/admin package. This refactoring keeps the main entry point thin by delegating server initialization, route setup, and graceful shutdown to the dedicated admin runner, improving code organization without changing user-facing behavior. Also in this area: 17 commits (8 fixes) modifying internal.
internal/server/admin · medium confidence · unverified
Behavioural changes
1 commit (0 fixes) modifying internal/deepdoc/parser/image
A change to existing behaviour in internal/deepdoc/parser/image — 1 commit, 2 files: “feat(ingestion): recognize image tables through the DeepDoc page pipeline (\#21911)” (refs \#21911).
internal/deepdoc/parser/image · medium confidence · unverified
1 commit (0 fixes) modifying internal/deepdoc/parser/office
A change to existing behaviour in internal/deepdoc/parser/office — 1 commit, 14 files: “refactor(parser): move native PDF/Office extraction into deepdoc facade (\#20273)” (refs \#20273).
internal/deepdoc/parser/office · medium confidence · unverified
1 commit (0 fixes) modifying internal/engine/clickhouse, internal/service/billing
A change to existing behaviour in internal/engine/clickhouse, internal/service/billing — 1 commit, 2 files: “refactor: keep the enterprise code out of the community build (\#21614)” (refs \#21614).
internal/engine/clickhouse, internal/service/billing · medium confidence · unverified
1 commit (0 fixes) modifying web/src/components/virtual-message-list
A change to existing behaviour in web/src/components/virtual-message-list — 1 commit, 2 files: “perf(web): virtualize chat message lists (\#21790)” (refs \#21790).
web/src/components/virtual-message-list · medium confidence · unverified
1 commit (0 fixes) modifying web/src/ee-stub
A change to existing behaviour in web/src/ee-stub — 1 commit, 66 files: “refactor(web): move OSS no-op overlays to ee-stub (\#22321)” (refs \#22321).
web/src/ee-stub · medium confidence · unverified
1 commit (0 fixes) modifying web/src/pages/agent/setting, web/src/pages/agent/sidebar, web/src/pages/chunk/components
A change to existing behaviour in web/src/pages/agent/setting, web/src/pages/agent/sidebar, web/src/pages/chunk/components — 1 commit, 12 files: “Feat: Adjust the page layout and move the navigation bar to the left side. (\#22330)” (refs \#22330).
web/src/pages/agent/setting, web/src/pages/agent/sidebar, web/src/pages/chunk/components · medium confidence · unverified
11 commits (2 fixes) modifying internal/server/ingestor
A change to existing behaviour in internal/server/ingestor — 11 commits (2 fixs), 4 files: “fix init admin client status (\#21316)”; “feat(ingestion): emit structured lifecycle events for document-parse tasks (\#21711)”; “refactor(deps): rename ragflow\_deps directory to deps (\#21345)”; and 8 more (refs \#21316, \#21711, \#21345, \#22172, \#21244, …).
internal/server/ingestor · medium confidence · unverified
126 commits (53 fixes) modifying internal/service/document
A change to existing behaviour in internal/service/document — 126 commits (53 fixs), 45 files: “fix: write wrong type of metadata (\#17605)”; “Go: align cancel-parse error with Python for un-started documents (\#18245)”; “Refactor: REST API show the real error messages (\#22212)”; and 123 more (refs \#17605, \#18245, \#22212, \#20108, \#19013, …).
internal/service/document · medium confidence · unverified
13 commits (2 fixes) modifying internal/harness/graph/graph
A change to existing behaviour in internal/harness/graph/graph — 13 commits (2 fixs), 25 files: “refactor: use WaitGroup.Go to simplify code (\#16539)”; “refactor: replace Split in loops with more efficient SplitSeq (\#18248)”; “More resilient graph engine (\#16325)”; and 10 more (refs \#16539, \#18248, \#16325, \#18821, \#16271, …).
internal/harness/graph/graph · medium confidence · unverified
15 commits (4 fixes) modifying internal/server/api
A change to existing behaviour in internal/server/api — 15 commits (4 fixs), 2 files: “Refactor: REST API show the real error messages (\#22212)”; “CLI: model group (\#21392)”; “feat(channels): support agent workflows and isolate startup failures (\#22031)”; and 12 more (refs \#22212, \#21392, \#22031, \#21316, \#21478, …).
internal/server/api · medium confidence · unverified
16 commits (7 fixes) modifying web/src/components/document-pipeline-dialog
A change to existing behaviour in web/src/components/document-pipeline-dialog — 16 commits (7 fixs), 3 files: “Fix: Modify the style of the "Built-in" label in the ParserSelect dropdown menu. (\#20669)”; “TS: more license declarations (\#18081)”; “Feat: Use a tree select component to display different versions of the pipeline. (\#21628)”; and 13 more (refs \#20669, \#18081, \#21628, \#21943, \#21408, …).
web/src/components/document-pipeline-dialog · medium confidence · unverified
17 commits (8 fixes) modifying internal
A change to existing behaviour in internal — 17 commits (8 fixs), 146 files: “perf(knowledge\_compile): cut merge-path engine reads (PR 7) (\#22241)”; “test(ingestion): scale burst-test completion window with task count (\#22302)”; “fix(pdf): preserve cross-page table previews (\#22293)”; and 14 more (refs \#22241, \#22302, \#22293, \#22212, \#22203, …). Also in this area: Centralized permission checks and token generation.
internal/permission/response · medium confidence · unverified
2 commits (0 fixes) modifying internal/channels/core
A change to existing behaviour in internal/channels/core — 2 commits, 1 file: “feat(chatchannel): stream Feishu, WeCom and DingTalk replies (\#22041)”; “feat: make chat-channel and implement WhatsApp bot (\#17518)” (refs \#22041, \#17518).
internal/channels/core · medium confidence · unverified
2 commits (0 fixes) modifying internal/harness/prebuilt
A change to existing behaviour in internal/harness/prebuilt — 2 commits, 3 files: “Refactor harness framework (\#16271)”; “Feat: add harness-go framework —— graph engine (\#16039)” (refs \#16271, \#16039).
internal/harness/prebuilt · medium confidence · unverified
2 commits (0 fixes) modifying web/src/pages/agent/form/keenable-form
A change to existing behaviour in web/src/pages/agent/form/keenable-form — 2 commits, 1 file: “Add Keenable web search tool to the agent (\#16233)”; “Refactor: Replace RAGFlowSelect with SelectWithSearch. (\#19795)” (refs \#16233, \#19795).
web/src/pages/agent/form/keenable-form · medium confidence · unverified
20 commits (10 fixes) modifying web/src/pages/agent/version-dialog
A change to existing behaviour in web/src/pages/agent/version-dialog — 20 commits (10 fixs), 2 files: “Fix: Canvas variables are not displaying in the agent's version history. (\#21675)”; “Fix text issue (\#12221)”; “fix(web): constrain flex children in agent version history dialog (\#19903)”; and 17 more (refs \#21675, \#12221, \#19903, \#21653, \#21538, …).
web/src/pages/agent/version-dialog · medium confidence · unverified
20 commits (7 fixes) modifying internal/harness/graph/pregel
A change to existing behaviour in internal/harness/graph/pregel — 20 commits (7 fixs), 40 files: “Go: format code (\#16813)”; “fix(harness): honor node and engine retry policies in Pregel (\#18388)”; “refactor: use WaitGroup.Go to simplify code (\#16539)”; and 17 more (refs \#16813, \#18388, \#16539, \#22074, \#16261, …).
internal/harness/graph/pregel · medium confidence · unverified
22 commits (4 fixes) modifying internal/entity
A change to existing behaviour in internal/entity — 22 commits (4 fixs), 10 files: “Add Slovak & Czech language dataset support (\#19013)”; “Implement retrieval\_test in GO (\#14231)”; “GO: improve model info parsing and add model\_id/tenant context to list response (\#16804)”; and 19 more (refs \#19013, \#14231, \#16804, \#16745, \#20026, …).
internal/entity · medium confidence · unverified
3 commits (0 fixes) modifying internal/harness/replay
A change to existing behaviour in internal/harness/replay — 3 commits, 7 files: “refactor: use the built-in max/min to simplify the code (\#17059)”; “Go: add context, part5 (\#19108)”; “Feat: add event sourcing and replay to harness (\#16326)” (refs \#17059, \#19108, \#16326).
internal/harness/replay · medium confidence · unverified
3 commits (0 fixes) modifying internal/ingestion
A change to existing behaviour in internal/ingestion — 3 commits, 2 files: “refactor(memory): inject MQ engine into memory publish path (\#21244)”; “refactor(ingestion): converge global singletons into injected dependencies (\#21203)”; “feat(parser): add media references for on-demand image materialization (\#21561)” (refs \#21244, \#21203, \#21561).
internal/ingestion · medium confidence · unverified
3 commits (0 fixes) modifying web/src/components/fallback-component
A change to existing behaviour in web/src/components/fallback-component — 3 commits, 1 file: “Feat/configurable metadata display (\#13464)”; “TS: more license declarations (\#18081)”; “Refactor: UmiJs -\> Vite (\#12410)” (refs \#13464, \#18081, \#12410).
web/src/components/fallback-component · medium confidence · unverified
3 commits (0 fixes) modifying web/src/pages/agent/form/list-operations-form
A change to existing behaviour in web/src/pages/agent/form/list-operations-form — 3 commits, 1 file: “Refa: align list operations and strict mode (\#14387)”; “Feat: Set the outputs type of list operation. \#10427 (\#11366)”; “Feat:new component list operations (\#11276)” (refs \#14387, \#10427, \#11366, \#11276).
web/src/pages/agent/form/list-operations-form · medium confidence · unverified
3 commits (1 fix) modifying internal/mcp/client
A change to existing behaviour in internal/mcp/client — 3 commits (1 fix), 6 files: “Combine RAGFLOW\_MCP\_ALLOW\_PRIVATE\_HOSTS into ALLOW\_ANY\_HOST (\#21523)”; “refactor(mcp): move MCP client out of utility into internal/mcp/client (\#21150)”; “fix(agent): runtime fixes for tool naming, legacy sessions, tenant scoping and MCP SSRF (\#20558)” (refs \#21523, \#21150, \#20558).
internal/mcp/client · medium confidence · unverified
3 commits (1 fix) modifying web/src/components/compilation-template-group
A change to existing behaviour in web/src/components/compilation-template-group — 3 commits (1 fix), 26 files: “Feat: Adjust the page layout and move the navigation bar to the left side. (\#22330)”; “Feat: Create a compilation template using a dialog on the pipeline canvas page. (\#21550)”; “Fix: Prevent the saving of dataset configurations when creating compilation templates. (\#21692)” (refs \#22330, \#21550, \#21692).
web/src/components/compilation-template-group · medium confidence · unverified
4 commits (0 fixes) modifying internal/ingestion/mediacache
A change to existing behaviour in internal/ingestion/mediacache — 4 commits, 4 files: “perf(ingestion): batch PurgeTask cache reclaim into a single DEL (\#22080)”; “feat(media): harden run media cache lifecycle and limits (\#21694)”; “test(mediacache): cover real Kvrocks lifecycle and index boundary (\#21851)”; and 1 more (refs \#22080, \#21694, \#21851, \#21561).
internal/ingestion/mediacache · medium confidence · unverified
4 commits (0 fixes) modifying web/src/pages/agent/form/yahoo-finance-form
A change to existing behaviour in web/src/pages/agent/form/yahoo-finance-form — 4 commits, 1 file: “Feat: Add invoke and github operators \#3221 (\#9112)”; “Feat: Add Arxiv GoogleScholar operator \#3221 (\#9102)”; “Feat: Add FormDrawer to agent page. \#3221 (\#5323)”; and 1 more (refs \#3221, \#9112, \#9102, \#5323, \#9088).
web/src/pages/agent/form/yahoo-finance-form · medium confidence · unverified
4 commits (1 fix) modifying deps
A change to existing behaviour in deps — 4 commits (1 fix), 2 files: “ci : add ttf font to ragflow\_deps image (\#21461)”; “refactor(deps): rename ragflow\_deps directory to deps (\#21345)”; “fix: harden credential and input handling (\#21593)”; and 1 more (refs \#21461, \#21345, \#21593, \#21599).
deps · medium confidence · unverified
4 commits (1 fix) modifying internal/harness/graph/channels
A change to existing behaviour in internal/harness/graph/channels — 4 commits (1 fix), 12 files: “fix(pregel): preserve slice types during state reduction (\#22014)”; “Refactor harness framework (\#16271)”; “Refactor(harness): remove naive inline graph engine , unify graph execution under single pregel engine (\#16608)”; and 1 more (refs \#22014, \#16271, \#16608, \#16039).
internal/harness/graph/channels · medium confidence · unverified
4 commits (1 fix) modifying internal/harness/graph/store
A change to existing behaviour in internal/harness/graph/store — 4 commits (1 fix), 3 files: “Refactor harness framework (\#16271)”; “Go: add context, part5 (\#19108)”; “fix(harness): avoid re-locking when Batch handles ListNamespacesOp (\#21709)”; and 1 more (refs \#16271, \#19108, \#21709, \#16039).
internal/harness/graph/store · medium confidence · unverified
4 commits (1 fix) modifying web/src/pages/builtin-pipeline
A change to existing behaviour in web/src/pages/builtin-pipeline — 4 commits (1 fix), 4 files: “Feat: Adjust the page layout and move the navigation bar to the left side. (\#22330)”; “Feat: Split the agent list page into three separate tabs. (\#21529)”; “Feat: Display the built-in pipeline canvas in read-only mode. (\#20742)”; and 1 more (refs \#22330, \#21529, \#20742, \#21669).
web/src/pages/builtin-pipeline · medium confidence · unverified
48 commits (11 fixes) modifying internal/server/config
A change to existing behaviour in internal/server/config — 48 commits (11 fixs), 30 files: “Add ee stub function (\#21671)”; “feat(engine): Add OpenSearch 3.9 as the document engine for the Go backend (\#21926)”; “refactor(mcp): centralize configuration and transport wiring (\#20011)”; and 45 more (refs \#21671, \#21926, \#20011, \#21523, \#17881, …).
internal/server/config · medium confidence · unverified
5 commits (0 fixes) modifying web/src/pages/agent/form/wikipedia-form
A change to existing behaviour in web/src/pages/agent/form/wikipedia-form — 5 commits, 1 file: “Feat: Render WikipediaForm and BaiduForm with shadcn-ui. \#3221 (\#5564)”; “Feat: Add Arxiv GoogleScholar operator \#3221 (\#9102)”; “Feat: Add FormDrawer to agent page. \#3221 (\#5323)”; and 2 more (refs \#3221, \#5564, \#9102, \#5323, \#9090, …).
web/src/pages/agent/form/wikipedia-form · medium confidence · unverified
5 commits (1 fix) modifying internal/deepdoc/parser/pdf/pdfoxide
A change to existing behaviour in internal/deepdoc/parser/pdf/pdfoxide — 5 commits (1 fix), 9 files: “fix(pdf): rejoin line-break hyphens / strip U+FFFD in Go layout (temporary workaround) (\#20274)”; “feat(agent): Go ingestion pipeline progress mirroring and DeepDOC parser hardening (\#16795)”; “Refactor: migrate pdf\_parser.py to golang (\#16323)”; and 2 more (refs \#20274, \#16795, \#16323, \#16391, \#21950).
internal/deepdoc/parser/pdf/pdfoxide · medium confidence · unverified
5 commits (1 fix) modifying web/src/pages/skills/components
A change to existing behaviour in web/src/pages/skills/components — 5 commits (1 fix), 11 files: “fix: correct Scholar sorting and Skills fallbacks (\#22095)”; “Feat: add skills space to context engine (\#13908)”; “TS: add apache license (\#18064)”; and 2 more (refs \#22095, \#13908, \#18064, \#14595, \#13545).
web/src/pages/skills/components · medium confidence · unverified
6 commits (0 fixes) modifying internal/parser/media
A change to existing behaviour in internal/parser/media — 6 commits, 13 files: “feat(parser): hand off Markdown and HTML remote images as media refs (\#21764)”; “refactor: consolidate duplicate helpers into internal/common (\#22222)”; “feat(parser): hand off EPUB images as media refs (\#21677)”; and 3 more (refs \#21764, \#22222, \#21677, \#21694, \#21700, …).
internal/parser/media · medium confidence · unverified
6 commits (1 fix) modifying internal/engine
A change to existing behaviour in internal/engine — 6 commits (1 fix), 37 files: “refactor(engine): hoist shared ES/OpenSearch chunk helpers into esoscommon (\#22323)”; “test(nats): target the first priority lane in pull capacity tests (\#22316)”; “refactor(engine): extract shared SanitizeString into esoscommon package (\#22320)”; and 3 more (refs \#22323, \#22316, \#22320, \#22336, \#22305, …). Also in this area: 17 commits (8 fixes) modifying internal.
internal/engine/opensearch, internal/ingestion/internal · medium confidence · unverified
6 commits (1 fix) modifying internal/observability
A change to existing behaviour in internal/observability — 6 commits (1 fix), 10 files: “feat(observability): carry user\_name/tenant\_name through IngestionLog (\#21968)”; “Stop persisting process\_end\_at; recover end time from begin + duration (\#21906)”; “feat(ingestion): emit structured lifecycle events for document-parse tasks (\#21711)”; and 3 more (refs \#21968, \#21906, \#21711, \#22168, \#22197, …).
internal/observability · medium confidence · unverified
7 commits (1 fix) modifying internal/harness/graph/task
A change to existing behaviour in internal/harness/graph/task — 7 commits (1 fix), 2 files: “Refactor harness framework (\#16271)”; “Go: fix warnings (\#17738)”; “Refactor(harness): remove naive inline graph engine , unify graph execution under single pregel engine (\#16608)”; and 4 more (refs \#16271, \#17738, \#16608, \#16537, \#17059, …).
internal/harness/graph/task · medium confidence · unverified
7 commits (2 fixes) modifying internal/mcp
A change to existing behaviour in internal/mcp — 7 commits (2 fixs), 5 files: “refactor(mcp): centralize configuration and transport wiring (\#20011)”; “Port agent PRs to GO (\#16529)”; “fix(mcp): remove unused force\_refresh parameter (\#21349)”; and 4 more (refs \#20011, \#16529, \#21349, \#20539, \#22005, …).
internal/mcp · medium confidence · unverified
7 commits (3 fixes) modifying internal/server/bootstrap
A change to existing behaviour in internal/server/bootstrap — 7 commits (3 fixs), 6 files: “fix init admin client status (\#21316)”; “feat(ingestion): emit structured lifecycle events for document-parse tasks (\#21711)”; “fix(ingestor): initialize shared retrieval services (\#22172)”; and 4 more (refs \#21316, \#21711, \#22172, \#21588, \#21893, …).
internal/server/bootstrap · medium confidence · unverified
8 commits (1 fix) modifying internal/harness/graph/checkpoint
A change to existing behaviour in internal/harness/graph/checkpoint — 8 commits (1 fix), 12 files: “refactor(log): unify Go logging to zap with rotation, strip per-package levels (\#16261)”; “refactor (\#21875)”; “More resilient graph engine (\#16325)”; and 5 more (refs \#16261, \#21875, \#16325, \#16271, \#17738, …).
internal/harness/graph/checkpoint · medium confidence · unverified
8 commits (1 fix) modifying web/src/pages/admin/layouts
A change to existing behaviour in web/src/pages/admin/layouts — 8 commits (1 fix), 3 files: “refactor(ui): adjust dataset page styles (\#13452)”; “Adjust styles to match the design system (\#11118)”; “feat(web): isolate enterprise routes, admin pages and header slots in src/ee (\#21696)”; and 5 more (refs \#13452, \#11118, \#21696, \#11155, \#12410, …).
web/src/pages/admin/layouts · medium confidence · unverified
9 commits (1 fix) modifying web/src/pages/agent/form/data-operations-form
A change to existing behaviour in web/src/pages/agent/form/data-operations-form — 9 commits (1 fix), 4 files: “Feat: Add data operation node \#10427 (\#10985)”; “Feat: Globally defined conversation variables can be selected in the operator's query variables. \#10427 (\#11135)”; “Feat: If a query variable in a data manipulation operator is deleted, a warning message should be displayed to the user. \#10427 \#11255 (\#11384)”; and 6 more (refs \#10427, \#10985, \#11135, \#11255, \#11384, …).
web/src/pages/agent/form/data-operations-form · medium confidence · unverified
9 commits (4 fixes) modifying internal/engine/kvrocks
A change to existing behaviour in internal/engine/kvrocks — 9 commits (4 fixs), 6 files: “Fix: export deleteIfEqual for ee (\#21351)”; “refactor(go): default Go cache/queue backend to Kvrocks and fail fast when down (\#19804)”; “fix(kvrocks): surface transport errors from SetNX and DistributedLock.Acquire (\#20660) (\#20774)”; and 6 more (refs \#21351, \#19804, \#20660, \#20774, \#22080, …).
internal/engine/kvrocks · medium confidence · unverified
Add PDF.js worker script for local path configuration
The PDF.js worker script (pdf.worker.min.js) is now included in the public directory, enabling the PdfLoader component to reference a local worker path instead of relying on external or CDN-hosted resources.
web/public/pdfjs-dist · high confidence
Admin server availability check is disabled
The system no longer validates whether the admin server is reachable or healthy before proceeding. The \IsAdminAvailable\ function in the enterprise edition file (\admin\_status\_ee.go\) has been hardcoded to always return \true\, effectively bypassing any connectivity or status checks that might have previously prevented operations if the admin server was unavailable.
internal/server/local · high confidence
Agent canvas logic and tool configuration hooks refactored
The agent canvas hooks have been restructured to improve node management and tool configuration. The \useBeforeDelete\ hook now enforces stricter deletion rules, protecting the Begin node and unattached IterationStart nodes, while ensuring that deleting an Agent node also removes its downstream Agent and Tool nodes. A new \useCanvasChecklist\ hook provides real-time validation of canvas issues, checking for stale datasets, invalid models, and missing memories. Tool initialization logic is now centralized in \useAgentToolInitialValues\, which applies specific field filtering for operators like Querit, Tavily, and Bing. Additionally, the canvas now supports structured output from Agent nodes, allowing downstream operators to reference nested JSON schema fields, and connection dragging has been refined to prevent placeholder node creation from input handles.
web/src/pages/agent/hooks · high confidence
Authentication wrapper redirects unauthenticated users to login
A new AuthWrapper component has been added to the application's wrapper layer. This component checks the user's login status using the useAuth hook; if the user is not logged in, it automatically redirects them to the login page via redirectToLogin, ensuring that protected routes are inaccessible without authentication.
web/src/wrappers · high confidence
C++ stemmer binding refactored with proper environment lifecycle management
The C++ stemmer binding in \internal/binding/cpp/stemmer\ has been refactored to provide explicit environment creation and cleanup functions (e.g., \SN\_create\_env\, \SN\_close\_env\, and language-specific variants like \danish\_UTF\_8\_close\_env\). This change ensures that stemmer environments are properly released after use, preventing resource leaks and allowing safe reuse of stemmer instances across multiple operations.
internal/binding/cpp/stemmer · high confidence
Canvas node UI and behavior overhaul
The agent canvas node components have been rewritten to use a new visual style and interaction model. Nodes now feature a standardized toolbar, a consistent wrapper with loading indicators, and new connection handles that support dropdown menus for adding downstream steps. The UI includes a right-click context menu for duplicating and deleting nodes, and specific nodes (like Agent and Begin) display configuration details such as LLM labels and webhook URLs. Additionally, LLM and compilation template cards now validate availability against the canvas owner's tenant, displaying error states for unavailable resources.
web/src/pages/agent/canvas/node · high confidence
Centralized chat channel configuration and status handling
The chat channel settings page now uses a dedicated constants module to define the available messaging integrations, including DingTalk, WeCom, WhatsApp, and others. This change centralizes the channel metadata (names, icons, and descriptions) and provides utilities for managing form values and displaying runtime status indicators (such as Connected, Connecting, or Error) within the user interface.
web/src/pages/user-setting/chat-channel/constant · high confidence
Centralized web constants and configuration
The web application now uses a dedicated \web/src/constants\ directory to centralize configuration data previously scattered across components. This change introduces new modules for agent node definitions (including operators and globals), chat variable settings, web search providers, file type mappings, language and timezone lists, knowledge base processing types, and LLM provider icons. It also adds constants for markdown sanitization and list-deletion behavior, providing a single source of truth for these values across the frontend.
web/src/constants · high confidence
Claims and evidence panels are now gated by compilation template kind
The chunk representation view now conditionally displays claims and evidence panels only for compilation templates that actually produce claim rows (specifically \page\_index\ and \tree\). For other templates (such as \knowledge\_graph\, \mind\_map\, \timeline\, \wiki\, \session\_graph\, \session\_essence\, and \empty\), the panels are suppressed to prevent showing empty states or unexplained content. This change ensures that users only see relevant compiled data when interacting with the graph tree, improving clarity and reducing noise in the artifact page's middle column.
web/src/pages/chunk/representation · high confidence
Code operator form now uses a single output contract with legacy multi-output support
The Code operator form has been refactored to use a new single-output schema (name and type) instead of the previous multi-output structure. Existing nodes with multiple business outputs are detected and displayed with a warning, instructing users to consolidate to a single output and move field extraction to downstream nodes. The form now includes an expandable full-screen code editor, supports switching between Python and JavaScript (which loads corresponding templates), and uses a new variable management system for inputs and outputs. The change also introduces a one-way data flow to synchronize form changes to the canvas and replaces the RAGFlowSelect component with SelectWithSearch for better searchability.
web/src/pages/agent/form/code-form · high confidence
Custom canvas background styling for agent nodes
The canvas now renders a distinct background for agent nodes using a new \AgentBackground\ component. This component applies a rounded border, uses the primary text color for the grid pattern, and sets the background color via the \--bg-canvas\ CSS variable, replacing the previous default canvas appearance.
web/src/components/canvas · high confidence
Data source management page restructured with unified connector validation
The Data Source settings page has been restructured to use a new dynamic form system for adding and editing connectors. This change introduces a unified validation step that runs before the Test Connection action, ensuring required fields (such as source configuration) are valid before attempting to contact the backend, which prevents spurious errors. The page now renders available connectors using a new \AvailableSourceCard\ component and manages the add/edit flow through a refactored \AddDataSourceModal\ that leverages \DynamicForm\ for field generation. Additionally, the \useTestDataSource\ hook and its associated tests confirm that the test connection logic is now gated by form validation, improving reliability when users test new or modified data source connections.
web/src/pages/user-setting/data-source · high confidence
Dataset settings page rebuilt with pipeline support and improved validation
The dataset configuration interface has been completely rewritten to support both built-in parsing and Go-based pipeline configurations. Users can now select a pipeline, which exposes component-specific operator tabs for advanced settings, while the general form handles core attributes like name, avatar, language, permissions, and embedding models. The new system enforces stricter validation using Zod schemas, ensures that parser configuration metadata is scoped correctly to individual pipeline nodes (preventing flat metadata keys from being sent to the backend), and provides better error handling by automatically scrolling to and revealing the first validation error after an invalid submission.
web/src/pages/dataset/setting · high confidence
Dynamic variable assignment form with real-time updates
The variable assignment node now features a dynamic form that adapts its input controls based on the selected variable's data type and the chosen logical operator. Users can assign values to variables using context-aware inputs, such as a JSON editor for objects, numeric inputs for numbers, and radio buttons for booleans, while array operations allow appending or removing elements. The form also supports real-time updates to ensure that changes to assigned variables are immediately reflected in the workflow state.
web/src/pages/agent/form/variable-assigner-form · high confidence
Enforce tenant-scoped file browsing and granular permission checks
The file service now enforces strict tenant isolation and resource-level permissions for file operations. Users can only browse files within their own tenant, and access to individual files and folders is validated against the permission layer before any read, write, or delete action. This includes checking ancestry for folder operations and ensuring that only authorized users can access shared tenant resources. The changes also introduce batch permission checks for large file trees to maintain performance while ensuring security.
internal/service/file · high confidence
Extractor form now validates extraction settings and supports tag file selection
The extractor operator form in the agent canvas now enforces validation rules for extraction features: enabling keywords, questions, or tags requires a positive top\_n value, and enabling tags additionally requires selecting a tag file. The form also introduces a metadata management modal for configuring custom and built-in metadata, and adds a tag file selector that allows users to browse and select files from the file manager (supporting xlsx, xls, csv, and txt extensions).
web/src/pages/agent/form/extractor-form · high confidence
Frontend application shell and routing infrastructure overhaul
The web application's entry point and core infrastructure have been restructured. The root component now initializes a new theme provider (defaulting to dark mode), integrates React Query for data fetching, and configures responsive breakpoints. A comprehensive route configuration has been introduced, defining paths for the new login flow, agent exploration, and administrative sections, while also implementing lazy loading for page components. Additionally, global styles have been updated to enforce the Inter font family and standardize scrollbar appearance across the interface.
web/src · high confidence
Frontend codebase migrated to TypeScript and RESTful API conventions
The web application has been rewritten in TypeScript, replacing the previous JavaScript implementation. This migration introduces strict typing for utility functions, components, and data models, improving code reliability and maintainability. Additionally, the frontend API layer has been consolidated to use a unified RESTful endpoint structure (\/api/v1\), replacing the previous mixed WEB/HTTP API patterns. New utility modules have been added for handling authentication, canvas operations, chat message processing, and document parsing, ensuring consistent behavior across the application.
web/src/utils · high confidence
Frontend hooks refactored for Go backend parity and improved reliability
The \web/src/hooks\ directory has been significantly refactored to align frontend behavior with the Go backend and improve stability. Authentication is now handled via a new \useOAuthCallback\ hook that manages OAuth flows and login state. Document ingestion and cancellation logic has been overhauled: a new \document-ingest-in-flight\ hook prevents duplicate requests, while a \cancel-stop-loss\ mechanism ensures that stuck document cancellation states are resolved by falling back to normal polling cadences. Parser configuration is normalized by \parser-config-utils\ to map legacy flat keys to component-scoped structures required by the Go backend. Additionally, pagination state is now persisted in \localStorage\ via \route-hook\ to survive navigation, and the \logic-hooks\ include improved handling for empty list states after deletions.
web/src/hooks · high confidence
Go CLI and server command-line interface overhaul
The command-line interface for the Go-based RAGFlow server and CLI tool has been restructured. The server now supports a unified binary with explicit mode flags (--admin, --api, --ingestor, --syncer, --deepdoc, --migrate) and a new --deepdoc-inference-concurrency flag to control DeepDoc processing limits. The CLI tool now includes a --version flag and defaults to a 'warn' log level, which can be raised to 'info' via the --verbose flag. Additionally, obsolete MCP-related flags have been removed from the CLI parser.
cmd · high confidence
Improved model selection and missing-model warnings in the agent canvas
The LLM selector component now displays a clear warning icon and label when a canvas node references a model that is no longer available to the current owner, helping operators identify broken configurations. It also passes the owner tenant ID to model-fetching hooks, ensuring that model lists and validity checks are correctly scoped for shared canvases. Additionally, the selector UI has been updated to support filtering by model type (chat or vision) and includes a new search-enabled select component for easier model discovery.
web/src/components/llm-select · high confidence
Introduce Go-based PDFium renderer for higher-quality PDF page images
A new Go package (internal/deepdoc/parser/pdf/pdfium) provides a statically linked, CGO-based renderer for PDF pages using the PDFium library. This replaces the previous pdf\_oxide rendering path for use cases where image quality is critical for downstream OCR/DLA, while pdf\_oxide continues to handle text, character, and table extraction. The implementation includes page rendering at configurable DPI, page dimension retrieval (respecting rotation), and document outline (bookmark) extraction with UTF-16LE title decoding. To ensure stability, a process-wide mutex (pdfsync) serializes all access to the single PDFium instance shared across bindings, preventing crashes from concurrent calls. Tests cover rendering correctness, DPI scaling, rotation handling, and outline parsing, including Chinese titles.
internal/deepdoc/parser/pdf/pdfium · high confidence
Introduce centralized LESS design tokens and mixins
The web application now uses a structured LESS foundation with new \variable.less\ and \mixins.less\ files. This introduces a consistent set of design tokens for colors (such as \@blurBackground\ and \@purple\) and font weights, alongside reusable style mixins for table cells, text ellipsis, and node styling. A key behavioral addition is explicit RTL (Right-to-Left) support within the table mixins, ensuring that tables and their content align correctly when the document direction is set to RTL.
web/src/less · high confidence
Introduce new Next.js chat list page with rename, delete, and creation capabilities
The chat list interface has been replaced with a new implementation in the \next-chats\ directory. Users can now create new chats, rename existing ones via a dedicated modal, and delete chats through a confirmation dialog accessed from a dropdown menu. The list supports pagination, search filtering by owner, and displays an empty state with a prompt to create a new chat when no conversations exist.
web/src/pages/next-chats · high confidence
Introduce tool selection popover with duplicate prevention for Agent nodes
The Agent form now includes a dedicated tool-popover component that lets users select built-in tools (such as search, communication, and developer operators) and MCP servers via a tabbed command interface. A key behavioral change is the enforcement of single-instance tools: selecting an already-added tool removes it instead of appending a duplicate, preventing backend crashes caused by duplicate tool names. The popover also automatically manages the associated tool node on the canvas, adding it when tools are selected and removing it when none remain.
web/src/pages/agent/form/agent-form/tool-popover · high confidence
Introduces memory configuration form constants with mandatory 'Raw' memory type
A new constants file defines the structure for the memory creation form, including field configurations for name, memory type, embedding model, and LLM. It establishes an enum for memory types (Raw, Semantic, Episodic, Procedural) and configures the memory type selector as a multi-select where the 'Raw' option is locked and required. A custom validation rule ensures that 'Raw' is always included in the selected memory types, preventing users from removing it.
web/src/pages/memories/constants · high confidence
Message form restructured with shadcn-ui and webhook status support
The agent message form has been rebuilt using shadcn-ui components (Form, Switch, SelectWithSearch) and refactored into modular hooks for state management. This change introduces a status field display for webhook operators in streaming mode, hides the autoplay switch and file download type selector in webhook mode, and ensures form data synchronization to the canvas uses one-way data flow with proper string array conversion.
web/src/pages/agent/form/message-form · high confidence
Migrate web frontend to Vite and Go backend
The web frontend has been rebuilt using Vite, replacing the previous UmiJS setup, and is now configured to serve only the Go backend. This change introduces a new development server on port 9222, updates the default language configuration to English via environment variables, and includes a new AGENTS.md file to guide development conventions for the new stack. Also in this area: 10 commits (2 fixes) modifying web.
web, web/\\mocks\\_ · medium confidence_
NATS engine self-healing, migration, and cache infrastructure
The NATS engine now automatically recovers from consumer disappearance by recreating the durable consumer when a pull fails, preventing the queue from getting stuck on 'no responders' errors. On startup, the engine detects and migrates existing task streams to the current configuration (DiscardNew, larger limits) instead of failing. A new shared NatsKVStore helper provides a consistent interface for cache namespaces (OCR, embedding, checkpoint) with idempotent create/update semantics. Additionally, the engine now returns errors instead of panicking when initialized with a nil connection, and dedicated streams, consumers, and KV buckets are established for knowledge-compile events and datasource syncer tasks.
internal/engine/nats · high confidence
New Go-based chunking pipeline with configurable strategies and post-processing
The internal text chunking engine has been replaced with a new Go implementation in \internal/parser/chunk\. This pipeline introduces a typed configuration model (\ChunkOptions\) and a stage-based execution flow (preprocess, split, postprocess). Users can now select from multiple split strategies, including sentence, character, paragraph, and a new length-based strategy that supports configurable window sizes and overlap. The system also adds configurable post-processing capabilities, allowing users to merge chunks to a target size, filter by minimum length, and drop empty or duplicate chunks. Delimiter handling has been standardized to support both single-character and multi-character delimiters (using backtick syntax) with longest-first matching and deduplication.
internal/parser/chunk · high confidence
New agent canvas implementation with shadcn-ui and React Flow
The agent canvas has been rebuilt using the React Flow library and the shadcn-ui component library, replacing the previous Ant Design-based implementation. This change introduces a new context provider for managing dropdown states and updates the visual styling of iteration and loop nodes to ensure proper layering and interaction. The canvas now supports a wider variety of node types, including data operations, variable aggregators, and loop controls, providing a more robust visual programming interface for building agents.
web/src/pages/agent/canvas · high confidence
New chat app settings interface with validation and error handling
The chat configuration UI has been replaced with a new settings panel located in \web/src/pages/next-chats/chat/app-settings\. This new interface introduces a \ChatBasicSetting\ component for managing the chat's name, icon, description, and LLM model, alongside a \ChatPromptEngine\ component for advanced prompt configurations including web search providers, metadata filters, and reranking settings. A key behavioral improvement is the addition of \useRevealSubmitErrors\, which automatically expands collapsed settings sections and scrolls to the first validation error when a user attempts to save invalid data. The settings are governed by a new Zod-based schema (\use-chat-setting-schema\) that enforces specific validation rules, such as requiring the \{knowledge}\ placeholder in system prompts when datasets are selected and preventing empty responses when no datasets are attached.
web/src/pages/next-chats/chat/app-settings · high confidence
New chat hooks for conversation management, streaming, and file uploads
The chat page now uses a new set of hooks in \web/src/pages/next-chats/hooks\ to manage conversations, streaming, and file uploads. \use-chat-url\ consolidates URL parameter handling and ensures a conversation is created before sending a message or uploading files. \use-click-card\ handles conversation switching with improved abort controller logic to prevent losing in-flight questions. \use-send-chat-message\ and \use-send-single-message\ manage the chat stream store, handle message sending, regeneration, and file uploads. \use-message-references\ optimizes reference resolution to prevent performance issues during streaming. \use-build-form-refs\, \use-button-disabled\, \use-create-chat\, \use-create-conversation\, \use-rename-chat\, \use-select-conversation-list\, \use-send-shared-message\, and \use-upload-file\ provide supporting functionality for form management, button states, chat creation/renaming, conversation list selection, shared chat messaging, and file uploads.
web/src/pages/next-chats/hooks · high confidence
New configuration schema for model metadata and search index mappings
The \conf\ directory now includes a comprehensive set of new configuration files that define the system's data structures and model capabilities. \all\_models.json\ and \llm\_factories.json\ provide a centralized catalog of supported AI models, providers, and their specific attributes (such as context length, output limits, and tool support), replacing previous ad-hoc model lists. Additionally, new mapping files (\mapping.json\, \infinity\_mapping.json\, \doc\_meta\_infinity\_mapping.json\, \message\_infinity\_mapping.json\, \skill\_infinity\_mapping.json\, \os\_mapping.json\, \doc\_meta\_es\_mapping.json\, \skill\_es\_mapping.json\) explicitly define the schema for Elasticsearch, OpenSearch, and Infinity vector databases, ensuring consistent indexing for documents, messages, and skills. \service\_conf.yaml\ has been updated to reflect these structural changes, including new sections for NATS, ClickHouse, and OpenTelemetry, while \model.md\ serves as the reference documentation for the new model configuration fields.
conf · high confidence
New data source components and UI refinements
The data source settings page now includes new React components for configuring Box and Gmail/Google Drive connections, featuring web-based OAuth flows and credential parsing. A new 'Added Source Card' displays connected sources with settings and delete actions, while a reusable 'Delete Source Modal' handles confirmation for both deletion and unlinking. These changes also incorporate UI style fixes for the S3 page and personal center, alongside general functional bug fixes within the data source module.
web/src/pages/user-setting/data-source/component · high confidence
New file management page with folder navigation and bulk operations
The file management interface has been replaced with a new implementation in web/src/pages/files. This update introduces a hierarchical file browser with breadcrumb navigation and folder creation, allowing users to organize files into directories. It adds bulk operation capabilities (move, link to knowledge base, delete) via a selection bar, and includes a dedicated dialog for linking files to datasets with optimistic name resolution. The new table supports row selection, sorting, and pagination, while specific UI behaviors have changed, such as hiding action buttons for the 'skills' folder and ensuring selection state is cleared on page changes.
web/src/pages/files · high confidence
New index-document processing task for ingestion
The ingestion pipeline now uses a dedicated \indexdoc\ task to normalize, validate, and prepare chunks before they are written to the search index. This change introduces a strict pre-index wire contract where the \text\ field must be a string, and it enforces the removal of internal bookkeeping fields (such as \layout\, \tk\_nums\, and \\_pdf\_positions\) that are not part of the persistent schema. The task also handles the conversion of raw position data into indexable \position\_int\ arrays for PDF and spreadsheet layouts, materializes parent-child chunk relationships with deterministic ordering, and merges typed metadata while preserving value types. Additionally, it ensures that \kb\_id\ is not stamped during ingestion, leaving that responsibility to the search engine write boundary.
internal/ingestion/task/indexdoc · high confidence
New interactive edge component with hover-to-delete and execution highlighting
The agent canvas now uses a new \ButtonEdge\ component to render connections between nodes. This change introduces a delete button that appears only when the cursor hovers over the connection line or the edge is selected, allowing users to remove edges more precisely. Additionally, edges are visually highlighted (using the accent color) when they are part of the active execution path during workflow runs, and placeholder target nodes are also highlighted to indicate valid connection points.
web/src/pages/agent/canvas/edge · high confidence
New navigation and table selection logic hooks
This change introduces a suite of new React hooks in the logic-hooks directory to standardize navigation and table interactions. The \useNavigatePage\ hook provides a centralized API for routing to datasets, agents, chats, and search results, while \useClientPagination\ and \useRowSelection\ manage client-side table state, including a new \useClearSelectionOnPageChange\ hook that ensures row selections are cleared when the user switches pages or changes the page size. Additionally, \useBuildOperatorOptions\ and \useBuildOptions\ streamline the creation of dropdown menus for agent logic and model types.
web/src/hooks/logic-hooks · high confidence
PDF parser types are re-exported from the shared doctype package
The PDF parser now uses a new \pdftype\ package that re-exports shared types, constants, and functions (such as \ParseResult\, \LayoutTypeText\, \IsCJK\, and \CollectFigures\) via Go type aliases from the central \doctype\ package. This change allows existing PDF parser code to continue working without modification while centralizing type definitions, and includes unit tests to verify the correctness of the re-exported utilities like \IsCJK\ and \CollectFigures\.
internal/deepdoc/parser/pdf/type · high confidence
Parser form restructured with global vision enhancement and per-file-type configuration
The parser configuration form has been refactored to support granular, per-file-type settings (PDF, Spreadsheet, PPT, Word, Audio, Email, Image, Text/Markdown, HTML) while consolidating vision capabilities into a single global toggle. Users can now enable 'Vision Enhancement' once to apply an image-to-text model across all vision-capable file types (PDF, Spreadsheet, Doc, Docx, Text/Markdown, Video), with the system automatically prefilling the tenant's default model. Audio files retain their own separate ASR model selection. The form also introduces new configuration options including TCADP parser settings (table result type, markdown image response type) for PDFs and spreadsheets, dynamic page range selection for PDFs, and header/footer removal options for Word, HTML, and Text files. The UI now allows adding and removing file-type setups dynamically, with validation ensuring at least one file type is configured and email parsers include required field selections.
web/src/pages/agent/form/parser-form · high confidence
Port agentic RAG to Go with eino ADK and new retrieval tools
The agentic RAG engine has been rewritten in Go and integrated with the eino ADK framework, introducing a new ReAct-based agent loop that supports configurable templates (e.g., smart-reasoning), per-chat prompts, and tool selection. This change adds new retrieval capabilities including BM25 lexical search and regex-based grep tools, while improving citation handling through a per-turn evidence registry that ensures accurate \[ID:N\] markers. The agent now features robust retry policies for LLM calls, metadata filter preservation, and failover mechanisms, with the maximum iteration count now configurable via environment variables or explicit request parameters.
_internal/agentic\rag · high confidence
Port dataset navigation tree routing to Go
The navigation-tree routing logic has been ported from Python to Go within the \internal/service/nav\ package. This introduces the \ChunkAggRouter\, which aggregates raw-chunk retrieval hits to score and return documents, and the \NavServiceRouter\, which routes queries through the compiled navigation index. Both routers implement the \NavTreeRouter\ contract used by the agentic harness, ensuring that retrieval failures are propagated as errors rather than empty results, and that only document leaves (not clusters) are routed. The change includes the \NavService\ interface and its singleton management, along with comprehensive unit tests verifying the routing, aggregation, and error-handling behaviors.
internal/service/nav · high confidence
Port knowledge graph retrieval pipeline to Go
The knowledge graph retrieval pipeline has been implemented in Go within the \internal/service/graph\ package, replacing the previous Python logic. This change introduces a new \Pipeline\ struct and supporting functions in \pipeline.go\, \retrieval.go\, \search.go\, and \scoring.go\ to handle entity and relation search, N-hop path analysis, and community report retrieval. A key behavioral fix included in this port is the use of a max-wins strategy for PageRank scores during N-hop path analysis (in \AnalyzeNHopPaths\), ensuring that when multiple paths share an edge, the highest weight is retained rather than the last processed value. The implementation includes comprehensive unit tests in \scoring\_test.go\ and \search\_test.go\ to verify the correctness of the scoring, search request building, and chunk parsing logic.
internal/service/graph · high confidence
Prompt editor rewritten with Lexical for improved variable handling and formatting
The prompt editor component has been rebuilt using the Lexical editor framework, replacing the previous implementation. This change introduces a dedicated VariableNode for structured variable insertion, a new variable picker plugin with typeahead menu support, and custom plugins for handling line breaks and paste operations to ensure consistent formatting. It also includes logic to prevent premature merging of variable paths during typing and adds a toolbar with a variable insertion trigger and a merge-path toggle.
web/src/pages/agent/form/components/prompt-editor · high confidence
Rebuilt the classification operator form with shadcn-ui and react-hook-form
The classification operator's configuration panel has been rewritten to use shadcn-ui components and react-hook-form for state management. This change introduces a dynamic form where users can define multiple classification categories, each with a name, description, and a collapsible list of example inputs. The form now validates that category names are unique and required, and it synchronizes changes to the canvas in real-time. Additionally, the form includes fields for selecting the large language model and configuring the message history window size.
web/src/pages/agent/form/categorize-form · high confidence
Redesigned message-item component with new interaction controls
The message-item component has been refactored to introduce a new toolbar with action buttons for assistant and user messages. Assistant messages now feature a read-aloud button (using the openai-speech-stream-player library), a copy button, and like/dislike feedback buttons, while user messages include copy, regenerate, and delete options. The component also supports displaying thinking/retrieval sections with distinct styling in dark mode, and integrates reference document and image lists below assistant messages.
web/src/components/message-item · high confidence
Redesigned model instance cards with batch save and inline rename
The model provider instance cards in the user settings have been refactored to replace the previous auto-save behavior with a batch 'Save-all' workflow driven by a top-level page button. The UI now distinguishes between draft and saved states: draft cards display the instance name input at the top without a save button, while saved cards use a collapsible layout that supports inline renaming via double-click. Additionally, the verify button is now conditionally hidden for draft cards and OpenAI-API-Compatible providers, and the Aimlapi provider includes a dedicated key retrieval button within the draft form.
web/src/pages/user-setting/setting-model/instance-card/components · high confidence
Redesigned model provider settings with sidebar navigation and batch saving
The model provider configuration page has been rebuilt with a new sidebar-driven layout that lists providers on the left and manages instances on the right. This update introduces a batch 'Save All' workflow, allowing users to validate and persist multiple provider instances in a single operation, replacing the previous per-card auto-save behavior. The interface now supports managing multiple instance drafts simultaneously and includes specific logic to prevent duplicate instance names. Additionally, the view now features a dedicated system settings section for default model configuration and provides inline support for AWS Bedrock region selection.
web/src/pages/user-setting/setting-model · high confidence
Redesigned team management interface with search, sorting, and role-based actions
The Team Settings page has been rebuilt to provide a modern, interactive experience for managing team members and joined workspaces. Users can now search and sort both their own team members and the teams they have joined by update date. The interface displays user roles using color-coded badges (Member, Owner, Invite) and provides specific actions based on status: inviting new members via a modal, deleting existing members with a confirmation dialog, and accepting or refusing team invitations. For regular members of joined teams, a 'Quit' option is available.
web/src/pages/user-setting/setting-team · high confidence
Refactored agent chat interface with session-scoped streaming and improved input handling
The agent chat experience has been restructured into a new \AgentChatBox\ component and a \ChatSheet\ wrapper, replacing the previous implementation. This change introduces session-scoped streaming gating, ensuring that streamed responses are correctly isolated to their active session and preventing answer leakage when switching between conversations. The chat input now supports vertical resizing, with the message list anchored to maintain visibility during height changes. Additionally, the system now surfaces backend memory errors as user-facing toasts and handles file uploads with progress tracking within the chat context.
web/src/pages/agent/chat · high confidence
Refined graph entity search and template selection logic
The graph representation view now handles entity searches more precisely: pressing Enter on an entity name highlights that specific node and refetches the relevant subgraph, rather than rendering the entire graph, while unmatched text falls back to a server-side keyword search. Additionally, switching the document-level graph selector now correctly updates the total node count, and the underlying template selection logic ensures a valid template is always displayed even when search results return a different bucket than the currently selected one.
web/src/pages/chunk/representation/hooks · high confidence
Restructured User Settings page with dedicated sidebar and local LLM support
The User Settings page has been restructured into a dedicated layout component featuring a sidebar navigation and a main content area, replacing the previous implementation. This change introduces a centralized constants file that defines the sidebar icon mapping for settings sections (Profile, Password, Model, System, Team, Logout, API, MCP) and establishes a list of supported local LLM factories (including Ollama, Xinference, LocalAI, LM Studio, and others). Additionally, a utility function has been added to identify local LLM factories, and the API key submission interface now explicitly supports a group\_id field.
web/src/pages/user-setting · high confidence
Structure graph component refactoring and stability fixes
The structure-graph components have been refactored to improve stability and data handling. The adapter layer now explicitly filters out self-referencing relations to prevent the mind map from freezing due to infinite recursion, and it correctly passes claim evidence and descriptions to the tree detail panel. Additionally, the mind map rendering now defers heavy data transformations to avoid blocking the render phase, and the timeline graph layout has been corrected to prevent node overlap by properly accounting for node sizes.
web/src/components/structure-graph · high confidence
Structured table parsing with merged-cell support and correct footer ordering
The table parser now converts HTML tables into a structured grid that preserves merged-cell geometry (colspan/rowspan) instead of discarding it, ensuring that cells spanning multiple rows or columns are correctly represented and that no data is silently dropped. Additionally, table footer rows are now consistently placed after the body rows, regardless of their order in the source markup, and the new implementation includes comprehensive tests to verify these behaviors.
internal/parser/tableutil · high confidence
Tokenizer now loads model assets from disk and enforces exact token counts
The internal tokenizer now loads BPE tables and SentencePiece models directly from the local filesystem (honoring MODEL\_ASSETS\_DIR, TIKTOKEN\_CACHE\_DIR, and DATA\_GYM\_CACHE\_DIR) instead of relying on network downloads, ensuring token counting works reliably in offline or restricted environments. It implements exact, per-model token counters (cl100k, XLM-R, BERT, Qwen, Llama) that match the model's own tokenizer files, replacing the previous generic cl100k approximation to prevent embedding API rejections caused by inaccurate token counts. Also in this area: Improved MinerU 4.0 support and more robust tool choice and rerank handling; Improved agent error readability and parallel resume reliability; Admin heartbeat API returns structured data and user creation no longer auto-provisions LLM configs.
internal/tokenizer · high confidence
Updated public assets: icon font and logo
The public directory now includes an updated icon font library (iconfont.js) and a revised logo file (logo.svg). These asset changes support the UI improvements mentioned in the commit history, such as the addition of new file type icons (HTML, JSON, video) and general icon replacements, ensuring the frontend has the necessary resources to render the updated visual elements.
web/public · medium confidence · unverified
Webhook trace status and terminal error messaging
The webhook debugging interface now accurately reflects the terminal status of agent traces. It displays specific error messages for failures such as node errors, cancellations, or user-wait states, and correctly distinguishes between successful completions and unsuccessful finishes, ensuring users see precise feedback on webhook execution outcomes.
web/src/pages/agent/webhook-sheet · high confidence
Fixes
1 commit (1 fix) fixing internal/permission/testsupport
A fix in internal/permission/testsupport — 1 commit (1 fix), 1 file: “fix: isolate community test stubs from EE builds (\#21672)” (refs \#21672).
internal/permission/testsupport · low confidence · unverified
1 commit (1 fix) fixing internal/service/youcom
A fix in internal/service/youcom — 1 commit (1 fix), 2 files: “Bugfix: add youcomsearch agent tool (\#21080)” (refs \#21080).
internal/service/youcom · medium confidence · unverified
11 commits (6 fixes) fixing web/src/pages/agent/components
A fix in web/src/pages/agent/components — 11 commits (6 fixs), 4 files: “Fix: The pipeline name is too long and overflows the release dialog. (\#21957)”; “Feat: Remove the execution restrictions imposed when an agent fails the checklist validation. (\#21232)”; “Fix: The alias for a historical version of an agent is too long, pushing the title out of view. (\#21653)”; and 8 more (refs \#21957, \#21232, \#21653, \#13542, \#13447, …).
web/src/pages/agent/components · medium confidence · unverified
12 commits (7 fixes) fixing internal/binding/cpp
A fix in internal/binding/cpp — 12 commits (7 fixs), 30 files: “Port PR14140 and PR16881 to GO (\#17102)”; “Go: format code (\#16813)”; “Add Slovak & Czech language dataset support (\#19013)”; and 9 more (refs \#17102, \#16813, \#19013, \#16602, \#21583, …).
internal/binding/cpp · medium confidence · unverified
16 commits (10 fixes) fixing internal/agent/component/io
A fix in internal/agent/component/io — 16 commits (10 fixs), 21 files: “feat(doc-generator): render markdown and expose html decorations (\#22040)”; “fix(agent): preserve Markdown headings and lists in DOCX (\#22078)”; “fix(agent): stamp PDF watermark on every page of DocsGenerator exports (\#19915)”; and 13 more (refs \#22040, \#22078, \#19915, \#15952, \#16858, …).
internal/agent/component/io · medium confidence · unverified
27 commits (14 fixes) fixing internal/channels
A fix in internal/channels — 27 commits (14 fixs), 28 files: “fix: secure channel egress against SSRF (\#21486)”; “feat\[Go\]: add dingtalk chat bot for chat channel (\#17565)”; “feat(go-channels): register qqbot channel (\#17598)”; and 24 more (refs \#21486, \#17565, \#17598, \#22031, \#20026, …).
internal/channels · medium confidence · unverified
3 commits (2 fixes) fixing internal/agentic\_rag/decisionhead
A fix in internal/agentic\_rag/decisionhead — 3 commits (2 fixs), 11 files: “fix(agentic\_rag): make the decisionhead package compile without cgo (\#21932)”; “fix(ingestion): contain process-fatal errors and bound parser work (\#21480)”; “refactor(agentic\_rag): fold internal/rag into internal/agentic\_rag (\#21464)” (refs \#21932, \#21480, \#21464).
_internal/agentic\rag/decisionhead · medium confidence · unverified
4 commits (3 fixes) fixing internal/channels/whatsapp-gateway
A fix in internal/channels/whatsapp-gateway — 4 commits (3 fixs), 6 files: “fix(channels): support WhatsApp gateways through Docker (\#22032)”; “refactor: remove python implemention of whatsapp channel (\#20432)”; “fix: harden OAuth, gateway auth, and provider parsing (\#22072)”; and 1 more (refs \#22032, \#20432, \#22072, \#21539).
internal/channels/whatsapp-gateway · medium confidence · unverified
57 commits (31 fixes) fixing web/src/pages/agent/utils
A fix in web/src/pages/agent/utils — 57 commits (31 fixs), 14 files: “Feat: Add a data compilation layer. (\#16777)”; “Feat: Delete the agent and tool nodes downstream of the agent node \#3221 (\#8450)”; “Fix: Limit the pipeline canvas to a maximum of one extractor operator. (\#20711)”; and 54 more (refs \#16777, \#3221, \#8450, \#20711, \#21487, …).
web/src/pages/agent/utils · medium confidence · unverified
8 commits (5 fixes) fixing web/src/components/parser-tree-select
A fix in web/src/components/parser-tree-select — 8 commits (5 fixs), 4 files: “Feat: Use a tree select component to display different versions of the pipeline. (\#21628)”; “Filter agent version, allow filter with alias (\#21943)”; “Feat: There should be a notification at the file level when a pipeline is deleted. (\#21765)”; and 5 more (refs \#21628, \#21943, \#21765, \#21695, \#21653, …).
web/src/components/parser-tree-select · medium confidence · unverified
9 commits (5 fixes) fixing web/src/pages/agent/explore/hooks
A fix in web/src/pages/agent/explore/hooks — 9 commits (5 fixs), 3 files: “Fix list chat/agent session paging (\#21604)”; “Feat: Adjust the page layout and move the navigation bar to the left side. (\#22330)”; “Fix: Clicking on the empty dialog box on the agent exploration page will result in an error. (\#14198)”; and 6 more (refs \#21604, \#22330, \#14198, \#19761, \#21212, …).
web/src/pages/agent/explore/hooks · medium confidence · unverified
Dataset deletion now properly removes storage buckets and files
Deleting a dataset now correctly cleans up associated storage resources, including the underlying MinIO bucket and all document objects. Previously, deletion might leave orphaned storage data behind; this change ensures that both the database records and the physical storage are fully removed when a dataset is deleted, preventing storage leaks and ensuring complete data removal. Also in this area: 5 commits (1 fix) modifying internal/service; 17 commits (8 fixes) modifying internal; Permission checks now enforce canvas context for dataset access; Dataset service adopts centralized permission checks and exposes connector sync logs; New Go-based Agentic RAG chat mode with configurable templates and per-dialog failover; and 2 more.
internal/service/dataset · medium confidence · unverified
Document preview component refactored with new previewers and stability fixes
The document preview component has been refactored to use dedicated previewers for various file types, including audio, CSV, DOCX, EPUB, Excel, images, PDFs, PPTs, and TXT files. Key improvements include handling non-UTF-8 encodings in CSV/TXT previews, fixing overlapping CJK text lines in DOCX previews by injecting default line spacing, and preventing infinite update loops in DOCX previews by disabling internal page virtualization. Additionally, the Excel previewer now sanitizes worksheet names that could crash the previewer and strips WPS DISPIMG formulas to prevent rendering issues. The CSV viewer also features stable horizontal scrolling for ragged rows and virtual scrolling for performance. These changes enhance the reliability and user experience of document previews across multiple file formats.
web/src/components/document-preview · high confidence
Fix bin directory inclusion in Docker images
The bin directory is now correctly included in Docker images by adding a .gitkeep file, resolving the issue where the directory was previously omitted during the build process.
bin · high confidence
Fix metadata value persistence and stale closure bugs
Resolves issues where metadata changes were not correctly saved or were lost due to stale state closures. The new hooks (\use-manage-modal.ts\, \use-manage-values-modal.ts\) implement proper state management for metadata operations, ensuring that inline value edits and modal submissions persist to the backend and that validation errors are surfaced correctly.
web/src/pages/dataset/components/metedata/hooks · high confidence
Fix: Document Generator form configurations can now be saved
The Document Generator form in the agent builder has been refactored to ensure that configuration changes (such as header/footer text, watermarks, and page numbers) are correctly persisted. This change introduces a new form structure using react-hook-form and Zod validation, along with a hook to sync form state with the graph store, resolving the issue where these settings could not be saved previously.
web/src/pages/agent/form/doc-generator-form · high confidence
Fixes agent retrieval bridge to preserve metadata and prevent infinite recursion
The new \retrievalbridge\ package resolves two critical issues in the agent's retrieval path. First, the \RuntimeAdapter\ now forwards the complete \RetrievalRequest\ (including \RerankID\, \ExcludeCompiled\, and \OnlyOriginalText\) to the underlying NLP service, preventing the silent loss of reranking and metadata filters that previously caused agents to fall back to local KNN scoring. Second, the \Enhancer\ correctly handles empty auto metadata filters by preserving the caller's document scope instead of returning a no-match sentinel, ensuring retrieval does not return zero results when a filter cannot narrow the search. Additionally, the \RuntimeAdapter\ captures the retrieval service at construction time to prevent a stack overflow caused by the adapter resolving itself from the shared registry.
internal/agent/retrievalbridge · high confidence
Fixes for global variable sheet form handling and object editing
The global variable sheet now correctly handles saving complex variable types, specifically ensuring that array-of-objects and object variables are properly parsed and persisted without data loss. This is achieved by introducing new hooks (\use-form.tsx\ and \use-object-fields.tsx\) that manage form submission logic and provide a dedicated JSON editor for object fields, including validation for key formats and empty parameters. Users can now reliably add and edit session variables of these complex types without them being overwritten or failing to save.
web/src/pages/agent/gobal-variable-sheet/hooks · high confidence
MCP server management UI and connection handling
The MCP server management interface in User Settings has been rebuilt to improve reliability and usability. The new edit dialog preserves authorization tokens and custom headers when editing imported servers, preventing accidental credential loss. Bulk operations now correctly reset selection state when switching modes or navigating pages, and the list view automatically navigates to the previous page when the last item on a page is deleted. Importing MCP configurations now processes all uploaded JSON files instead of stopping at the first one, and exporting generates correctly named files based on the selected server.
web/src/pages/user-setting/mcp · high confidence
New chat box components with improved state isolation and hydration
This change introduces two new components in the chat box area: SingleChatBox and MultipleChatBox. SingleChatBox now safely hydrates conversation messages from the server, ensuring that stale messages from a previous conversation are not leaked when switching to a new one. MultipleChatBox enables independent model configuration per chat card, preventing the configuration applied in one card from overwriting the settings in sibling cards.
web/src/pages/next-chats/chat/chat-box · high confidence
Profile page validation and timezone display fixes
The profile settings page now enforces strict validation on the nickname field, limiting it to 100 characters and allowing only letters, numbers, spaces, and specific punctuation. Additionally, the timezone selection has been fixed to properly display and update the user's timezone, resolving previous issues where the timezone was not shown or could not be saved correctly.
web/src/pages/user-setting/profile · high confidence
Surface dataset navigation tree load failures
When the dataset navigation tree fails to load, the application now displays an error message to the user instead of silently failing or showing a blank state. This ensures users are immediately aware of connectivity or backend issues when browsing their dataset structure. Also in this area: 10 commits (2 fixes) modifying web; Google Drive sync flexibility and MinerU 4.0 support.
(repo-wide) · medium confidence · unverified
Test coverage
1 commit adding/updating tests in internal/docguard; 1 commit adding/updating tests in sdk/python/tests; 10 commits adding/updating tests in test/testcases/test\_sdk\_api/test\_agent\_management; 11 commits adding/updating tests in test/testcases/test\_sdk\_api/test\_memory\_management; 13 commits adding/updating tests in test/testcases/test\_sdk\_api/test\_message\management; 2 commits adding/updating tests in internal/harness/graph/managed; 2 commits adding/updating tests in web/src/components/markdown-content/\\tests\\; 21 commits adding/updating tests in web/src/hooks/\\tests\\_; Added PDF parser regression and parity test suite; Added Playwright end-to-end tests for dataset, agent, chat, and search workflows; Added RAGFlow HTTP Benchmark CLI and test suite; Added SDK API tests for chat assistant management; Added SDK chunk management tests for datasets; Added SDK dataset management test suite; Added SDK tests for dataset file management operations; Added cross-domain registry smoke test; Added end-to-end tests for PDF page range filtering via pdfium; Added integration tests for the /api/v1/components endpoint; Added integration tests for wiki incremental build and deletion; Added parity tests for orphan-row cleanup in PDF table parsing; Added single-core PDF throughput and peak-RSS benchmark; Added test file for Stop Parse Documents API; Added test infrastructure for HTTP and SDK test cases; Added test infrastructure for Python SDK integration tests; Added test utility modules for file generation and helper functions; Added tests for CAJAL agent template, PDF version validation, and file integrity hooks; Added tests for ES to OceanBase migration tool; Added tests for MultiSelect component behavior; Added tests for agent canvas utilities and DSL serialization; Added tests for chunker delimiter utility functions; Added tests for pipeline file log detail query key generation; Added tests for the useManageValues metadata hook; Added tests for webhook header field resolution in the Begin form; Added unit tests for Collapse, Knowledge Base Item, and Switch Form Field components; Added unit tests for web utility functions; Chunker regression tests and structural layout contract; Improved ingestion reliability and metadata handling; Initial Playwright E2E test suite and infrastructure; Migrated API integration tests from Python to Go; New Playwright end-to-end test suite for authentication flows; New Python-intermediate parity harness for PDF parsing; New SDK API test suite with shared fixtures and Go-proxy skip logic; New test utilities for ingestion components; Parser alignment tests and structured output types.
Dependencies
New dependency manifests for admin client, sandbox, and Go core
The project now includes explicit dependency definitions for several previously unmanaged or implicit areas. The admin CLI (\ragflow-cli\) is defined in \admin/client/pyproject.toml\ with Python 3.13 support and dependencies like \requests\ and \beartype\. The agent sandbox executor manager and base images now have dedicated \requirements.txt\ and \package.json\ files, pinning Python packages (FastAPI, Uvicorn) and Node.js packages (Axios). Additionally, the Go backend's dependencies are now explicitly tracked in a new \go.mod\ file, listing core libraries for cloud storage, database drivers, and the Infinity SDK.
(dependencies) · high confidence
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
How this codebase got here
Score
- CAI 43 → 40 (-2.7)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.3) — scores are not directly comparable.
Lenses
- Code Health 58 → 67 (+9.0)
- Architecture 58 → 56 (-2.0)
- Maturity 88 → 88 (+0.4)
- Readiness 32 → 27 (-4.8)
- Security 54 → 51 (-2.6)
- Event-Driven 91 → 91 (+0.0)
- Event Sourcing 61 → 49 (-12.5)
- Accessibility 43 → 45 (+2.7)
- Performance 90 → 97 (+7.0)
Resolved (896)
- (anonymous) (cognitive 22) (internal/channels/whatsapp-gateway/index.js)
- (anonymous) (cyclomatic 24) (internal/channels/whatsapp-gateway/index.js)
- ADR has no Consequences section (internal/agent/audio/tts_design.md)
- AgentHandler.ListAgents (cyclomatic 18) (internal/handler/agent.go)
- BotHandler.ChatbotCompletion (cognitive 16) (internal/handler/bot.go)
- Change coupling: llm.ts ↔ api.ts (web/src/interfaces/request/llm.ts)
- Change coupling: manage-values-modal.tsx ↔ knowledge-service.ts (web/src/pages/dataset/components/metedata/manage-values-modal.tsx)
- Change coupling: operator-icon.tsx ↔ use-add-node.ts (web/src/components/operator-icon.tsx)
- Change-coupling hub: accordion-operators.tsx → operator-icon.tsx, form-config-map.tsx, use-add-node.ts, tool-timeline-item.tsx (web/src/pages/agent/canvas/node/dropdown/accordion-operators.tsx)
- Change-coupling hub: en.ts → confirm-delete-dialog.tsx, operator-icon.tsx, tooltip.tsx, chat.ts, accordion-operators.tsx, form-config-map.tsx, workflow-timeline.tsx, canvas-checklist.ts, manage-modal.tsx, manage-values-modal.tsx, index.tsx, hooks.tsx, available-models.tsx, index.tsx (web/src/locales/en.ts)
- Change-coupling hub: form-config-map.tsx → operator-icon.tsx, use-add-node.ts, tool-timeline-item.tsx (web/src/pages/agent/form-sheet/form-config-map.tsx)
- Change-coupling hub: index.tsx → next-step-dropdown.tsx, index.tsx, index.tsx (web/src/pages/agent/constant/index.tsx)
- Change-coupling hub: tool-command.tsx → index.tsx, constant.tsx, use-add-node.ts, use-agent-tool-initial-values.ts, tool-timeline-item.tsx (web/src/pages/agent/form/agent-form/tool-popover/tool-command.tsx)
- Change-coupling hub: zh.ts → tooltip.tsx, chat.ts, accordion-operators.tsx, form-config-map.tsx, manage-values-modal.tsx, hooks.tsx, available-models.tsx (web/src/locales/zh.ts)
- ChatHandler.ListChats (cognitive 18) (internal/handler/chat.go)
- ChatSessionHandler.ListChatSessions (cognitive 20) (internal/handler/chat_session.go)
- ChunkService.StopParsing (cognitive 17) (internal/service/chunk/chunk.go)
- ClassTooLong: ModelFactory (internal/service/model_factory.go)
- ClassTooLong: TenantService (internal/service/tenant.go)
- Consumer.recoverDocTypes (cognitive 22) (internal/ingestion/knowledge_compile/consumer.go)
- …and 876 more
New (2017)
- (anonymous) (cognitive 23) (internal/channels/whatsapp-gateway/index.js)
- (anonymous) (cyclomatic 25) (internal/channels/whatsapp-gateway/index.js)
- ADR still proposed, but the code already adopted it (internal/agent/audio/tts_design.md)
- AgentHandler.UploadAgentFile (cyclomatic 17) (internal/handler/agent_upload.go)
- AgentHandler.persistAgentToolInvocations (cognitive 22) (internal/handler/agent.go)
- AgentHandler.persistAgentToolInvocations (cyclomatic 16) (internal/handler/agent.go)
- AgentService.CreateAgent (cognitive 16) (internal/service/agent.go)
- AgentService.CreateAgent (cyclomatic 16) (internal/service/agent.go)
- AgentService.PublishAgent (cyclomatic 18) (internal/service/agent.go)
- AgentService.mergeAgentsAndGroups (cognitive 16) (internal/service/agent.go)
- AgentWebhookCredentialService.Resolve (cognitive 24) (internal/service/agent_webhook_credential.go)
- AgentWebhookCredentialService.Resolve (cyclomatic 26) (internal/service/agent_webhook_credential.go)
- AimlapiModel.Embed (cognitive 18) (internal/entity/models/aimlapi.go)
- AimlapiModel.Embed (cyclomatic 16) (internal/entity/models/aimlapi.go)
- AimlapiModel.ListModels (cognitive 16) (internal/entity/models/aimlapi.go)
- AliyunModel.uploadAudio (cognitive 28) (internal/entity/models/aliyun.go)
- AliyunModel.uploadAudio (cyclomatic 29) (internal/entity/models/aliyun.go)
- AuthHandler.AuthMiddleware (cognitive 17) (internal/handler/auth.go)
- AuthHandler.BetaAuthMiddleware (cognitive 17) (internal/handler/auth.go)
- CLI.AdminListUsersConditionCommand (cognitive 17) (internal/cli/admin_command.go)
- …and 1997 more
Changes since last survey
- 300 commits — 142 feature/other, 158 fixes
By area
- web/src — 62 commits
- internal/service — 39 commits
- internal/ingestion — 31 commits
- internal/agent — 15 commits
- (root) — 14 commits
- internal/engine — 14 commits
- internal/parser — 13 commits
- internal/entity — 12 commits
- internal/handler — 12 commits
- internal/channels — 10 commits
- internal/dao — 10 commits
- internal/deepdoc — 8 commits
- docs/guides — 7 commits
- internal/admin — 6 commits
- docs/administrator — 4 commits
- internal/common — 4 commits
- internal/harness — 4 commits
- internal/observability — 4 commits
- .github/workflows — 3 commits
- internal/agentic_rag — 3 commits
Notable commits
- fix: Fix build.sh --test error (#22285)
- fix: Fix doc icons (#22299)
- fix: Fix doc icons (#22300)
- fix: Fix doc icons (#22303)
- fix: Fix docs icons (#22298)
- fix: Fix(chatchannel):Strip the DingTalk sessionWebhook query credential from transport errors (#22077)
- fix: Fix: Add a text description to the cancel-parsing button. (#21966)
- fix: Fix: Adjust layout of DatasetFilter component to conditionally render FilterPopover and SearchInput (#21924)
- fix: Fix: Agent – Incomplete display of blueprint examples when creating a wiki template in Compiler (#21919)
- fix: Fix: Align the agent operator title to the left. (#22051)
- fix: Fix: Chat embedded in a website: automatically scrolls down when a question is answered. (#21940)
- fix: Fix: Dataset file configuration: switching from "pipeline" to "built-in" also triggers an exclamation mark warning. (#22010)
- fix: Fix: Delete the agent log entry (#21910)
- fix: Fix: Deleting a pipeline bound to a dataset should not trigger an error when navigating to the file list page of a different dataset. (#21985)
- fix: Fix: File Management - Move - Folder name is too long, causing visual clipping (clipping through UI elements) (#21962)
- fix: Fix: If the title of an agent operator is too long, it overflows beyond the boundaries of the sheet. (#22013)
- fix: Fix: Increase the right-hand margin on the agent share page. (#22016)
- fix: Fix: Increase the spacing between the labels and the dropdown menus for the PDF and table parsers. (#21882)
- fix: Fix: Long messages in floating embedded pages overflow the message box. (#21951)
- fix: Fix: The agent operator name is too long, causing the icon to be compressed. (#22042)
- …and 280 more
Architecture
- Unchanged — 0 containers · 1 contexts · 0 edges
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
Survey your own repository
infiniflow/ragflow was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.
About this page
- The score is its most recent published measurement, taken on 13 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
- Measured at commit c451c648a4dbc22db89f5246fb117d961977aae5 — the exact code this score is about.
- Scored under rubric-2026.11.3 — the same rubric and the same method as every other entry in this index.
- Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.61+preprod-f3449e6eb431.