langflow-ai/langflow
48.7
Weak · 14 November 2026
541.3k
lines of production code
Python
with TypeScript, JavaScript
5
measurements over time
What this system is
Features
Add Docker example for Langflow with PostgreSQL
Users can now run Langflow locally using the provided Docker setup in the \docker\_example\ directory. This includes a \Dockerfile\ and \docker-compose.yml\ that launch the Langflow application (using the \langflowai/langflow:latest\ image) alongside a PostgreSQL 16 database (pinned to the \trixie\ base to prevent collation mismatches). The configuration exposes port 7860 for the UI and requires setting the \LANGFLOW\_SUPERUSER\_PASSWORD\ environment variable. A pre-release variant is also available via \pre.Dockerfile\ and \pre.docker-compose.yml\, which uses the \langflowai/langflow:1.0-alpha\ image.
_docker\example · high confidence
Add GCP deployment scripts and documentation for Langflow
New shell scripts and documentation have been added to the scripts/gcp directory to enable deploying Langflow on Google Cloud Platform. The deploy\_langflow\_gcp.sh script provisions a standard Debian VM with Langflow and Nginx, while deploy\_langflow\_gcp\_spot.sh offers a similar setup using preemptible instances for cost savings. Accompanying documentation (GCP\_DEPLOYMENT.md and walkthrough tutorials) provides step-by-step instructions for setting up the necessary VPC, subnet, and firewall rules via Google Cloud Shell.
scripts/gcp · high confidence
Add Korean (ko) locale validation script
A new validation script (scripts/i18n/check\_ko.py) has been added to enforce quality and consistency for the Korean (ko) locale. This tool checks the frontend and backend Korean translation files against the English sources and a defined style guide, flagging issues such as missing keys, mismatched variables, glossary violations, and translationese to ensure the Korean UI text meets established standards.
scripts/i18n · high confidence
Add OpenAI Compatible provider bundle
A new standalone extension bundle, lfx-openai-compatible, registers an OpenAI Compatible model provider in Langflow's unified model picker. It allows users to connect any endpoint that speaks the OpenAI HTTP API shape (such as OpenRouter, Together AI, Groq, Fireworks AI, or self-hosted vLLM/TGI/LM Studio) by configuring a base URL and optional API key. The provider reuses the ChatOpenAI and OpenAIEmbeddings classes, automatically discovers available models from the endpoint's /v1/models route, and normalizes base URLs to ensure consistent /v1 API root usage for both chat and embedding requests.
src/bundles/openai-compatible · high confidence
Add PaddleOCR extension bundle for document parsing and OCR
Users can now install the standalone \lfx-paddle\ bundle to access the PaddleOCR component in Langflow. This component supports two modes: layout-aware document parsing into Markdown (using PP-StructureV3 or PaddleOCR-VL-1.6) and plain OCR text recognition (using PP-OCRv5 or PP-OCRv6). It communicates with the PaddleOCR AI Studio async Job HTTP API via \httpx\, eliminating the need for the local \paddleocr\ Python SDK. The component requires an AI Studio access token and is registered under the \paddle\ group in the component palette. Saved flows referencing the legacy \lfx.components.paddle.\*\ paths are automatically migrated to the new namespaced ID \ext:paddle:PaddleOCRComponent@official\.
src/bundles/paddle · high confidence
Add Serping API Search extension bundle
A new standalone Langflow extension bundle, \lfx-serpingapi\, is available that adds the \SerpingApiSearchComponent\. This component allows users to perform Google web searches via the Serping API, returning organic results as a table. It requires a user-supplied API key and supports optional inputs for country, language, location, max results (1-100), and page number. The component is registered under the ID \ext:serpingapi:SerpingApiSearchComponent@official\ and appears in the palette's \*\Bundles\\* section under \*\Serping API\\* after installation.
src/bundles/serpingapi · high confidence
Add Serply Search extension bundle
Users can now install the standalone \lfx-serply\ extension bundle to add the \SerplySearchComponent\ to Langflow. This component performs web searches via the Serply SERP API and returns organic Google results as a table. It requires a user-supplied API key, uses \httpx\ directly without vendor SDK dependencies, and exposes a specific tool named \serply\_search\ in tool mode to avoid naming conflicts with other search components.
src/bundles/serply · high confidence
Add migration rehearsal fixtures for testing data migrations
A new \scripts/migration\_rehearsal\ directory provides tooling to seed a small, adversarial Langflow instance for rehearsing whole-instance migrations (such as moving data between deployments or converting from SQLite to Postgres). The \seed\_instance.py\ script populates the database with specific fixture rows designed to catch silent data loss or corruption during migration, including issues with superuser teardown, role ID realignment, polymorphic share references, encrypted credential handling, and knowledge base vector stores. A generated manifest records the seeded data and expected failures to allow automated verification of migration integrity.
_scripts/migration\rehearsal · high confidence
Add vLLM as a standalone extension model provider
Langflow now supports vLLM as a first-class model provider via the new \lfx-vllm\ extension bundle. This provider appears in the unified model picker for both Language Models and Embedding Models, leveraging vLLM's OpenAI-compatible API to discover available models live from the server's \/v1/models\ endpoint. Users can configure the provider by setting the \VLLM\_API\_BASE\ environment variable (or via Settings) and optionally providing an API key. The bundle includes live discovery and credential validation, and respects the connector loopback policy for SSRF protection.
src/bundles/vllm · high confidence
Added IBM Globalization Pipeline (GP) integration scripts for backend i18n
The \scripts/gp\ directory now contains a complete toolchain for managing backend translations via IBM Globalization Pipeline. This includes \gp\_client.py\ for API authentication and communication, \extract\_backend\_strings.py\ to scan component classes and starter projects for translatable strings, \bake\_note\_keys.py\ to assign stable content-addressed keys to note nodes, and \upload.py\/\download.py\ to sync English source strings and target translations. A \check\_backend\_status.py\ script allows monitoring translation progress, and the suite is supported by unit tests for the client, extraction, and upload/download logic.
scripts/gp · high confidence
Azure and Ollama bundles graduate to standalone extension packages
The Azure and Ollama model providers are now shipped as standalone Langflow Extension Bundles (\lfx-azure\ and \lfx-ollama\) with proper \extension.json\ manifests, replacing their previous manifest-less \lfx-bundles\ provider status. This structural change ensures existing saved flows retain their canonical component IDs while providing a clearer installation path via \pip install lfx-azure\ or \lfx-ollama\. The Ollama bundle also includes a fix for the model picker: it now gracefully skips individual models that fail capability checks (e.g., cloud models returning 410 errors) instead of emptying the entire dropdown, and it prevents raw HTTP error text from leaking into the UI by correctly catching \httpx.HTTPStatusError\.
src/bundles/ollama · high confidence
DuckDuckGo Search component extracted into standalone extension bundle
The DuckDuckGo Search component is now distributed as a standalone Langflow Extension Bundle (\lfx-duckduckgo\), replacing the previous monolithic \lfx.components\ structure. Users can install the bundle via \pip install lfx-duckduckgo\ to register the \DuckDuckGoSearchComponent\ under the namespaced ID \ext:duckduckgo:DuckDuckGoSearchComponent@official\. Existing flows referencing the legacy class name or import path are automatically migrated to the new canonical ID upon loading, ensuring continuity of saved workflows.
src/bundles/arxiv, src/bundles/duckduckgo · high confidence
Initial repository scaffolding and configuration
The repository is initialized with essential configuration files, including ESLint and Storybook settings for the frontend, Git attributes for consistent line endings, a Whitesource security scan configuration, and a Contributor Covenant Code of Conduct. This establishes the baseline development environment and governance standards for the project.
(repo-wide) · high confidence
Introduce DataStax Cassandra components with SSRF protection
This change adds three new components to the DataStax extension: a Cassandra Vector Store, a Cassandra Graph Vector Store, and a Cassandra Chat Memory component. To address security concerns, the components now validate database contact points and discovered peer addresses against SSRF attacks before establishing connections, ensuring that only safe hostnames are used for driver connections.
_src/bundles/datastax/src/lfx\datastax/components/cassandra · high confidence
Introduce LFX as a standalone Langflow Executor CLI
This change introduces the LFX (Langflow Executor) project as a standalone CLI tool for running and serving Langflow workflows. It includes a new CLI entry point with command groups for setup, authoring, upgrades, extensions, and remote execution, alongside Dockerfiles for production and development environments. The package also ships with a static component index and a versioned hash history to support deterministic component resolution, and provides a pluggable service configuration example to wire LFX to Langflow's full-featured service implementations.
src/lfx · high confidence
Introduce Microsoft 365 and Teams integration bundle
This change adds the \lfx-microsoft\ bundle, providing a set of components for Microsoft 365 and Teams. Users can now interact with Outlook (send mail, search mail, create/list calendar events), SharePoint/OneDrive (list and fetch files), and Microsoft Teams (post to channels and chats). The bundle also includes source triggers for calendar, file, and mail events. It relies on a new \MicrosoftGraphComponent\ base class for credential management and scope validation, and includes a \capabilities.v1.json\ manifest defining the required OAuth scopes and permissions for each action.
src/bundles/microsoft/src · high confidence
Introduce NextPlaid multi-vector vector store bundle
This change adds the \lfx-nextplaid\ bundle, introducing two new Langflow components: \NextPlaidVectorStoreComponent\ for storing documents using NextPlaid's multi-vector (ColBERT/ColPali-style) embeddings, and \VllmMultivectorEmbeddingsComponent\ for generating those token-matrix embeddings via vLLM. The bundle is registered as a standalone extension with version 0.1.6, enabling users to leverage late-interaction retrieval models within Langflow workflows. Also in this area: 3 commits (3 fixes) fixing src/bundles.
_src/bundles/nextplaid/src/lfx\nextplaid · medium confidence · unverified
Introduce Slack integration bundle with user and bot actions and event triggers
This change adds the \lfx-slack\ bundle, providing a set of Slack Web API actions and event triggers for Langflow. The bundle supports two executing identities—connected user and app bot—each with its own authorization profile and required scopes. It includes components for searching messages, reading threads, sending messages, creating canvases, posting as a bot, adding reactions, and listing channel members. Additionally, it introduces trigger components for Slack messages and reactions, which can operate via the Events API or Socket Mode depending on the connection type. The implementation includes a custom error normalizer to handle Slack's HTTP 200 responses with error payloads, ensuring proper user-facing error messages for issues like authentication failures, rate limits, and invalid requests. The bundle also enforces strict identity checks to prevent actions from running with mismatched tokens.
src/bundles/slack · high confidence
Introduce langflow-stepflow package for translating and executing Langflow workflows
A new \langflow-stepflow\ package is added to provide a translation and execution backend for Langflow workflows. It includes a \LangflowConverter\ that translates Langflow node graphs into Stepflow steps, handling dependency analysis, schema mapping, and routing of core and custom-code components. A dedicated Stepflow worker server is provided to execute these translated components, supporting both known core components and custom code blobs, along with utilities for applying Langflow-style tweaks to translated workflows.
src/langflow-stepflow · high confidence
Introduce lfx-toolguard extension with PoliciesComponent for tool protection
Users can now install the standalone \lfx-toolguard\ extension (available via \lfx\[toolguard\]\ or \langflow-base\[toolguard\]\) to access the \PoliciesComponent\. This component enables building tool protection code from textual business policies using the ALTK ToolGuard runtime. It supports two modes: generating new guard code from policies and applying existing guards to tools, with the generated code executed asynchronously to enforce policy validation before tool invocation. Also in this area: 3 commits (3 fixes) fixing src/bundles.
src/bundles/toolguard · medium confidence · unverified
Introduce managed Google Workspace components and triggers
This change adds the initial set of Google Workspace integration components to the Langflow UI. Users can now connect their Google accounts via managed connections to perform actions such as sending emails (Gmail), listing and fetching files (Drive), and managing calendar events. The update also introduces source triggers that allow flows to react to changes in Gmail, Drive, and Calendar. A shared client adapter ensures consistent error handling, authentication, and telemetry across these new components.
_src/bundles/google/src/lfx\google/components · high confidence
Introduce new DataStax Astra DB components and lazy-loading module structure
This change adds a new set of Langflow components for interacting with DataStax Astra DB and Hyper-Converged Database (HCD), including Astra DB Chat Memory, CQL tools, a direct Data API interface, Graph Vector Store, standard Vector Store, Vectorize configuration, and Graph RAG traversal. To support the ongoing bundle refactoring, a new \\_\init\\_.py\ module is introduced that uses lazy imports to re-export these components, ensuring that existing flows referencing the old module paths continue to resolve correctly.
_src/bundles/datastax/src/lfx\datastax/components/datastax · high confidence
Introduce standalone Google and Gemini component bundle
Users can now access Google and Gemini components as a standalone Langflow Extension Bundle (lfx-google v0.3.5). This new bundle exposes components for BigQuery, Gmail, Google Drive, Google Search, and Google Generative AI, along with OAuth token handling, allowing for modular integration of these services.
_src/bundles/google/src/lfx\google · high confidence
Introduce standalone Valkey extension bundle with Vector Store and Chat Memory components
Langflow now offers a standalone \lfx-valkey\ extension bundle that provides Valkey integration via two new components: \ValkeyVectorStoreComponent\ for vector search and \ValkeyIndexChatMemory\ for chat history. The bundle registers these components under the \ext:valkey\ namespace and includes a migration table to automatically update saved flows from the legacy \lfx.components.valkey.\*\ paths. The Vector Store component now validates connector hostnames to prevent SSRF attacks and patches the underlying \langchain-aws\ library to correctly handle \RequestError\ exceptions when checking for index existence. The Chat Memory component ensures secure URL construction by properly encoding credentials and validating against SSRF threats.
src/bundles/valkey · high confidence
Introduce the Langflow Python SDK (v0.1.0)
This change adds the initial release of the Langflow Python SDK, providing both synchronous and asynchronous clients to interact with the Langflow REST API. Users can now manage flows and projects (list, create, update, upsert, delete) and execute flows via run endpoints. The SDK includes environment configuration support via a TOML file (with lookup from explicit paths, environment variables, or default locations) to manage multiple instances (e.g., local, staging, production) without hardcoding credentials. It also features git-friendly flow serialization utilities that strip volatile metadata, clear secrets, and sort keys for stable diffs, as well as a BackgroundJob class for non-blocking flow execution in async contexts. A pytest plugin is included for integration testing flows against live Langflow instances.
src/sdk · high confidence
Introduction of the standalone lfx-datastax extension bundle
The lfx-datastax package is now available as a standalone Langflow Extension Bundle (version 0.1.9), allowing users to install and register DataStax components independently. This bundle exposes a set of AstraDB and Cassandra integration components—including tools, vector stores, chat memory, and graph RAG capabilities—under the namespace ext:datastax, making them discoverable by Langflow's loader via the provided extension.json manifest.
_src/bundles/datastax/src/lfx\datastax · high confidence
New AI agent skills for backend code review and component refactoring
The AI coding agent now includes two new skills to assist with code quality and maintainability. The \backend-code-review\ skill provides structured guidance for reviewing backend Python code, covering security (SQL injection, SSRF), performance (N+1 queries, blocking I/O), architecture (layering, dependency direction), and testing standards. The \component-refactoring\ skill guides the refactoring of high-complexity React frontend components, offering patterns for extracting custom hooks and sub-components to reduce complexity and improve testability.
.agents · high confidence
New CI and deployment tooling for bundle releases, authorization, and production readiness
Added several new scripts and modules to support the Langflow bundle separation and production deployment workflows. The \bundle\_release\_plan.py\ script manages the release lifecycle for bundle packages, handling version planning, PyPI content-aware retries, and artifact verification. A new \check\_authz\_endpoint\_matrix.py\ script enforces authorization contracts by validating that API routes are correctly classified in the OSS authorization matrix. Production deployment safety is improved with \preflight.py\, which runs fail-loud infrastructure checks (database, storage, encryption) before booting in production mode, and \preload.py\, which implements a Gunicorn master preload hook to optimize memory usage via copy-on-write. Additionally, \check\_router\_trust.py\ acts as a CI guard to forbid unsafe extension installation routes, and \port\_bundle.py\ automates the migration of in-tree components into standalone extension bundles.
langflow · high confidence
New CI linting rules to prevent environment variable leaks and exception text exposure
Two new static-analysis scripts have been added to the CI linting suite to enforce security best practices in component and flow code. The first script, \check\_component\_env\_writes.py\, prevents components from writing to \os.environ\ (including methods like \update\, \setdefault\, and calls to \load\_dotenv\), which could cause credential leakage between concurrent requests in long-lived serving processes. The second script, \check\_logged\_exception\_text.py\, ensures that caught exceptions are not interpolated into log messages, preventing sensitive data (such as API keys or prompts) from being exported to external logging systems via the message field rather than the local traceback.
scripts/lint · high confidence
New CI validation and release tooling
The CI pipeline now includes a suite of new scripts and configuration files to enforce release integrity and integration contracts. This adds an authorization endpoint matrix to define route-to-authorization contracts, a pytest plugin for backend test metrics and timeout diagnostics, and a Dockerfile for extending base images with provider bundles. It also introduces scripts to validate capability manifests against discovery-gate matrices, check authentication flows, verify core dependency pins in release wheels, and ensure retired dependencies like Chroma are not reintroduced.
scripts/ci · high confidence
New Cursor rules and settings for documentation development
Added a new \.cursor/rules/docs\_development.mdc\ file that provides AI coding assistants with specific guidelines for maintaining Langflow's Docusaurus-based documentation, including environment setup (Node.js v22, npm, Docusaurus v3), directory structure, MDX writing conventions (frontmatter, imports, admonitions), and testing procedures. Additionally, created a \.cursor/settings.json\ file to configure the Biome linter path for the frontend project.
.cursor · high confidence
New Dockerfile and Nginx configuration for the frontend image
The frontend Docker image now uses a dedicated Dockerfile (build\_and\_push\_frontend.Dockerfile) that builds the frontend assets using Node.js and serves them via Nginx. The Nginx configuration (default.conf.template) enables gzip compression, sets a 1-day cache expiration for static assets, and proxies /api, /health\_check, and /health requests to the backend. The start-nginx.sh script allows runtime customization of the frontend port, backend URL, and maximum upload file size via environment variables.
docker/frontend · high confidence
New Dockerfile suite and build infrastructure for Langflow
This change introduces a comprehensive set of new Docker build files and configuration scripts for the Langflow project. It includes production-ready images (\build\_and\_push.Dockerfile\, \build\_and\_push\_backend.Dockerfile\, \build\_and\_push\_ep.Dockerfile\) that utilize \uv\ for Python dependency management and Node.js 24 for frontend builds, based on the UBI 10 Python 3.14 minimal image. It also adds development environment configurations (\dev.Dockerfile\, \dev.docker-compose.yml\, \dev.start.sh\) to streamline local development with hot-reloading and PostgreSQL integration. Additionally, it provides specific deployment targets for CDK (\cdk.Dockerfile\, \cdk-docker-compose.yml\) and Render (\render.Dockerfile\, \render.pre-release.Dockerfile\), along with a hardened npm installation script (\install\_hardened\_npm.sh\) to address security vulnerabilities in npm dependencies.
docker · high confidence
New EmpirioLabs extension bundle for text and image generation
A new standalone Langflow extension bundle (lfx-empiriolabs) is available, providing two components: EmpirioLabs AI for text generation and EmpirioLabs AI Image Generation. These components connect to the EmpirioLabs OpenAI-compatible API (https://api.empiriolabs.ai/v1) to support language models and image generation. The bundle includes a migration path to automatically update saved flows from the legacy lfx.components.empiriolabs.\* import paths to the new namespaced IDs (ext:empiriolabs:EmpirioLabsModelComponent@official and ext:empiriolabs:EmpirioLabsImageGenerationComponent@official).
src/bundles/empiriolabs · high confidence
New Grafana Alloy and Loki reference stack for structured log ingestion
A new reference deployment in deploy/observability/grafana-loki provides an end-to-end stack for ingesting Langflow's structured JSON logs into Grafana Loki and visualizing them. The stack uses Grafana Alloy (v1.18) to scrape log files, replacing the legacy Promtail agent which has reached end of life. It includes a pre-provisioned Grafana dashboard (v11.3) that displays panels for PII redaction verification, error tracebacks, and log rates by logger. Users must configure Langflow to output JSON logs (LANGFLOW\_LOG\_ENV=container) and point the Alloy container to the log directory to enable this observability feature.
deploy/observability · high confidence
New IBM component bundle (lfx-ibm)
Users can now install the standalone \lfx-ibm\ extension bundle to access IBM-specific components in Langflow. This bundle introduces five new components: a DB2 Vector Store, WatsonxAI LLM and Embeddings, and WatsonxData integrations for Presto and MCP. The package is registered as a distribution unit that Langflow's loader discovers via \extension.json\, making these capabilities available as namespaced extensions (e.g., \ext:ibm:DB2VectorStoreComponent@official\).
_src/bundles/ibm/src/lfx\ibm · high confidence
New IBM components: Db2 Vector Store, watsonx.ai, watsonx.data, and embeddings
The IBM bundle now includes five new components: a Db2 Vector Store for ingesting and searching vector data with SSL and bulk-insert options; an IBM watsonx.ai model component for text generation; an IBM watsonx.ai embeddings component for generating embeddings; an IBM watsonx.data Presto component to run SQL queries against the Presto engine and return results as a DataFrame; and an IBM watsonx.data MCP component that exposes watsonx.data tools to an Agent via the remote MCP server. All components include input validation and SSRF protections to secure connections and credential handling.
_src/bundles/ibm/src/lfx\ibm/components · high confidence
New NextPlaid vector store and vLLM multivector embeddings components
Users can now index and retrieve data using the NextPlaid vector store, which supports multi-vector (ColBERT/ColPali-style) embeddings for both text and images. This change introduces the NextPlaid component for managing the vector store index and the vLLM Multivector Embeddings component to generate token-level embeddings via a vLLM server. The implementation includes SSRF protections for the vLLM API connections and provides inputs for configuring quantization, timeouts, and batch sizes to optimize retrieval quality.
_src/bundles/nextplaid/src/lfx\nextplaid/components · high confidence
New Windows build and run automation scripts
Added Windows batch (.bat) and PowerShell (.ps1) scripts to automate the Langflow build and startup process on Windows. The \build\_and\_run\ scripts handle installing frontend dependencies, building the frontend, copying the build artifacts to the backend directory, and launching the application via \uv\, with support for optional \.env\ file configuration. Additionally, \build\_frontend\_only\ scripts are provided to perform just the frontend build step, allowing users to build the UI separately before running the backend.
scripts/windows · high confidence
New environment setup and validation scripts
Added check\_env.sh to detect the active Python interpreter (supporting virtualenv, conda, pipenv, and pyenv) and validate that the installed Python version meets the required range, and added setup\_env.sh to automatically create a .env file if it does not already exist.
scripts/setup · high confidence
New operational and maintenance scripts for LFX and Langflow
This change introduces a suite of new scripts in the \scripts/\ directory to support development, release, and maintenance workflows. Key additions include \build\_component\_index.py\ for generating a static, deterministic component index to speed up startup; \migrate\_secret\_key.py\ for rotating encryption keys and re-encrypting sensitive database fields; \release-lfx.sh\ for managing LFX version bumps and releases; and \factory\_restart\_space.py\ for Hugging Face Space management. Additionally, new tooling includes \check\_changes\_filter.py\ for CI coverage validation, \check\_deprecated\_imports.py\ for enforcing import standards, \generate\_migration.py\ for creating Alembic migrations, and \test-api-examples-local.sh\ for local API example testing.
scripts · high confidence
New provider bundles for AgentQL, AI/ML API, ALTK, Apify, and AssemblyAI
Langflow now includes several new provider bundles in the lfx-bundles metapackage, adding components for web data extraction (AgentQL), LLM and embedding generation via AI/ML API, an advanced agent with pre-tool validation and post-tool processing (ALTK), data extraction via Apify Actors, and a full suite of transcription and subtitle tools via AssemblyAI. These components are registered as official provider bundles and are discovered at runtime, expanding the available integrations for users building flows.
src/bundles/lfx-bundles/src · high confidence
New route-aware accessibility scanning infrastructure
A new Python-based accessibility scanner (\scripts/a11y/a11y\_scan.py\) and a canonical route manifest (\scripts/a11y/a11y\_routes.json\) have been introduced to standardize how the application is audited for accessibility. The scanner uses the IBM ACE engine to evaluate specific frontend routes against defined policies, ensuring that the results are consistent with the existing Playwright test suite. The manifest explicitly categorizes routes into static, dynamic, gated, and excluded groups, allowing the scanning process to target specific surfaces like the Flows list, Settings pages, and Flow editor without scanning irrelevant redirects or aliases. Additionally, support for state files (e.g., \settings-messages.json\) enables the scanner to test complex UI states, such as scrolled grids, ensuring that accessibility issues in interactive components are detected.
scripts/a11y · high confidence
Security
Azure OpenAI components enforce credential egress policy
The Azure OpenAI chat and embeddings components now enforce a credential egress policy that prevents server-provisioned API keys or Azure AD tokens from being sent to untrusted endpoints. Before connecting to Azure, the components validate that the target endpoint is either explicitly configured by the operator or listed in the allowed hosts, and they apply SSRF protection to block requests to private or metadata addresses. This ensures that credentials are only transmitted to authorized Azure resources, mitigating risks from misconfigured endpoints or SSRF vulnerabilities.
src/bundles/azure · high confidence
OpenAI bundle introduces SSRF protection for custom API endpoints
The new \lfx-openai\ bundle (version 0.1.5) adds security controls to the OpenAI Embeddings and Chat Model components. When users configure a custom \OpenAI API Base\ URL, the system now validates that the endpoint is allowed and applies a connector SSRF policy to prevent server-side request forgery attacks. This ensures that operator credentials are only sent to sanctioned hosts, protecting against malicious tenant-chosen endpoints.
src/bundles/openai/src · high confidence
Behavioural changes
2692 commits (1290 fixes) modifying src/backend/base
A change to existing behaviour in src/backend/base — 2692 commits (1290 fixs), 1045 files: “Add asyncio loop to LangflowApplication initialization”; “feat: add new node Input type for data in table format (\#2635)”; “fix: url component output types and id generator component (\#3740)”; and 2608 more (refs \#2635, \#3740, \#14919, \#5811, \#2269, …).
src/backend/base · medium confidence · unverified
34 commits (16 fixes) modifying scripts/migrate
A change to existing behaviour in scripts/migrate — 34 commits (16 fixs), 4 files: “feat(integrations): add lfx connection resolution contract (\#14919)”; “feat(components): add MrScraper integration (\#14516)”; “fix(deps): require cryptography 50 (\#14514)”; and 31 more (refs \#14919, \#14516, \#14514, \#13869, \#15333, …).
scripts/migrate · medium confidence · unverified
6 commits (3 fixes) modifying src/bundles/confluent
A change to existing behaviour in src/bundles/confluent — 6 commits (3 fixs), 16 files: “chore: move 1.13 bundles onto their own minor line (\#15656)”; “fix: patch release 1.12.10 image vulnerabilities (\#16201)”; “fix: harden release 1.13.2 image dependencies (\#16196)”; and 3 more (refs \#15656, \#16201, \#16196, \#14612, \#14652, …).
src/bundles/confluent · medium confidence · unverified
8 commits (4 fixes) modifying src/bundles/oracle
A change to existing behaviour in src/bundles/oracle — 8 commits (4 fixs), 24 files: “fix(oracle): confine loader local file access (\#15973)”; “fix(release): keep bundle versions ahead of PyPI”; “fix(security): validate component connection targets (\#16043)”; and 5 more (refs \#15973, \#16043, \#16119, \#15656, \#13563, …).
src/bundles/oracle · medium confidence · unverified
Centralized cookie management and new utility modules for security, validation, and formatting
The frontend utility layer has been reorganized with several new modules. A singleton CookieManager has been introduced to centralize cookie handling, automatically applying secure and SameSite flags based on the connection protocol. Security is enhanced by a new utility that sanitizes MCP server configurations before export, stripping sensitive headers, environment variables, and tokens. File uploads now use a dedicated validation module for stricter extension and MIME-type checks. Additional helpers include functions for parsing RFC 5987 filenames, formatting token counts, handling markdown table cleanup in chat messages, and managing local storage paths. The codebase also adds utilities for extracting API error messages, appending error suggestions, and generating scoped duplicate flow names. Also in this area: 1 commit (1 fix) fixing langflow, src/frontend; 3 commits (3 fixes) fixing src/frontend; Add test coverage for AG-UI bridge and API controllers; Added i18n validation tests for locale bundles; Expanded E2E test coverage for core workspace and accessibility features; and 170 more.
src/frontend · medium confidence · unverified
Dynamic version detection for multiple package variants
The backend now determines its version at runtime by checking for the presence of several possible package names (langflow, langflow-base, langflow-nightly, and langflow-base-nightly). This ensures the correct version string is reported regardless of which specific variant is installed, and includes a helper to identify pre-release versions according to PEP 440.
src/backend/langflow · high confidence
Exa Search component modernized to exa-py SDK in standalone bundle
The Exa Search component has been moved from the legacy \lfx-bundles\ metapackage into a new standalone \lfx-exa\ bundle and updated to use the \exa-py\ SDK. This change introduces a new \exa\_api\_key\ input field while maintaining backward compatibility by still accepting the deprecated \metaphor\_api\_key\ for existing saved flows. The component now exposes two tools—\search\ and \get\_contents\—supporting advanced filtering options such as search type, categories, domain allow/deny lists, and date ranges, with token-efficient highlights enabled by default.
src/bundles/exa · high confidence
Firecrawl integration upgraded to v2 SDK and extracted into a standalone bundle
The Firecrawl components (Scrape, Crawl, Map, and Search) have been extracted from the core into a standalone extension bundle (\lfx-firecrawl\) and upgraded to use the \firecrawl-py\ v2 SDK. This change updates the underlying API calls to match v2 requirements, including mapping legacy v1 parameters (such as \maxDepth\ to \max\_discovery\_depth\ and \allowBackwardLinks\ to \crawl\_entire\_domain\) and handling new v2 response types. Saved flows are automatically migrated to the new namespaced component IDs via the migration table, and the bundle is registered automatically upon installation.
src/bundles/firecrawl · high confidence
Restructured DataStax bundle into internal base and components packages
The DataStax bundle has been reorganized to improve modularity. A new \base\ package now houses shared infrastructure, specifically the \AstraDBBaseComponent\ mixin that all components in this bundle inherit from. This base logic was previously located at \lfx.base.datastax\ but has been moved into the bundle itself since it is specific to DataStax and only used by its components. The \components\ package has also been established as a distinct module within the bundle.
_src/bundles/datastax/src/lfx\_datastax/base, src/bundles/datastax/src/lfx\datastax/components · high confidence
Fixes
10 commits (7 fixes) fixing src/bundles/docling
A fix in src/bundles/docling — 10 commits (7 fixs), 11 files: “fix(release): keep bundle versions ahead of PyPI”; “fix: Move Docling components into bundle (\#13442)”; “fix(deps): dependency upgrades to 1.12.6 (\#15654)”; and 7 more (refs \#13442, \#15654, \#15656, \#15019, \#14196, …).
src/bundles/docling · medium confidence · unverified
Amazon Bedrock Converse component now gates Top K by provider
The Amazon Bedrock Converse component now only sends the Top K parameter to providers that explicitly support it (currently Anthropic), preventing API errors on models like Meta, Amazon Titan, AI21, Cohere, and Mistral that reject the field. Geography prefixes in model IDs are correctly stripped before provider resolution to ensure Top K is applied to regional Anthropic profiles. The S3 Bucket Uploader component also includes updated file input handling and strategy dispatch logic, with tests verifying correct behavior for both 'Store Data' and 'Store Original File' strategies. Also in this area: 2 commits (2 fixes) fixing src/bundles.
src/bundles/amazon · medium confidence · unverified
Anthropic bundle introduces SSRF protection and thinking-block compatibility
The new standalone \lfx-anthropic\ bundle adds security and compatibility improvements for Anthropic model usage. It enforces SSRF (Server-Side Request Forgery) protection on the customizable \base\_url\ field by validating endpoints against internal/cloud-metadata ranges and installing DNS-pinned HTTP clients, ensuring tenant-editable URLs cannot be exploited. Additionally, it patches the underlying \ChatAnthropic\ client to backfill missing \thinking\ fields in request payloads, resolving compatibility issues with Anthropic's thinking blocks without requiring subclassing that could break Pydantic defaults.
src/bundles/anthropic · high confidence
Test coverage
3 commits adding/updating tests in src/backend/tests/integration/triggers; Added Locust-based load testing suite for Langflow and LFX APIs; Added empty test package for language model integration tests; Added end-to-end deployment tests for Watsonx Orchestrate; Added integration test suite for backend components and APIs; Added integration tests for AstraDB vector store component; Added integration tests for MCP component and session management; Added integration tests for ParseJSONDataComponent; Added integration tests for PromptComponent; Added integration tests for backward compatibility and basic prompting flows; Added integration tests for the CSV output parser; Added integration tests for the CometAPI component; Added integration tests for the Parse JSON component; Added missing \_\init\\_.py files in test directories; Added performance benchmarks for API key validation, flow authorization, and server initialization; Added property-based tests for serialization logic; Added test coverage for the DataStax bundle; Added test coverage for the IBM bundle components; Added test fixtures for chat, loop, and API graph workflows; Added test suite for Google Workspace integration bundle; Added test suite for the Microsoft bundle; Added tests for OpenAI model component SSRF protection and error handling; Added tests for database migration validation rules; Added tests for task service revoke and temp flow cleanup; Added tests for trace/span enum serialization and cascade delete behavior; Added unit and end-to-end tests for the agentic assistant; Added unit and integration tests for flow version CRUD and deployment attachment handling; Added unit tests for API exception handling; Added unit tests for API key CRUD operations; Added unit tests for API utilities, build execution, and security policies; Added unit tests for API v2 file management, MCP server configuration, and workflow resume routes; Added unit tests for Agentics and ALTK bundle components; Added unit tests for Alembic database migrations; Added unit tests for CLI admin, directory, integrity, and migration-preflight commands; Added unit tests for Calculator, Python REPL, SearXNG, and SerpAPI tools; Added unit tests for Celery application and configuration; Added unit tests for ChatInput and TextInput components; Added unit tests for ChatOutput and TextOutput components; Added unit tests for Code Agents bundle imports and lazy loading; Added unit tests for CreateData and UpdateData components; Added unit tests for EmbeddingsWithModels and OllamaEmbeddingsComponent; Added unit tests for Figranium bundle components; Added unit tests for Groq model discovery and integration; Added unit tests for Gunicorn preload and server pre-fork safety; Added unit tests for IO schema and table column logic; Added unit tests for LLM operations components; Added unit tests for Langflow Assistant service behaviors; Added unit tests for MCP client utilities and configuration; Added unit tests for MCP server discovery, credential handling, and URL configuration; Added unit tests for MCP server security, assistant tooling, and support utilities; Added unit tests for MrScraper components; Added unit tests for RunFlow component logic; Added unit tests for TelemetryWriterService; Added unit tests for Watsonx Orchestrate deployment service; Added unit tests for agentic assistant helpers; Added unit tests for agentic flow components and prompts; Added unit tests for agentic utility modules; Added unit tests for authorization service internals; Added unit tests for backend components and services; Added unit tests for backend helper modules; Added unit tests for cache service security and teardown; Added unit tests for catalog policy database model; Added unit tests for component importability and SSRF guard conformance; Added unit tests for custom component loading logic; Added unit tests for database model utility functions; Added unit tests for database service reliability and authorization; Added unit tests for deployment artifact builder; Added unit tests for deployment database models and guards; Added unit tests for deployment provider accounts, flow version deployment attachments, and transaction models; Added unit tests for durable background execution; Added unit tests for files, knowledge, and memory components; Added unit tests for flow control components; Added unit tests for flow loading and remote flow fetching; Added unit tests for folder guard pre-check logic; Added unit tests for graph branch traversal and Smart Router isolation; Added unit tests for graph cache restoration and execution path equivalence; Added unit tests for initial setup and starter project integrity; Added unit tests for input validation and instantiation; Added unit tests for interface initialization parameter loading; Added unit tests for knowledge base utility functions; Added unit tests for local and S3 storage services; Added unit tests for model utilities and provider variable resolution; Added unit tests for new and unified processing components; Added unit tests for new language model components; Added unit tests for prompt validation rules; Added unit tests for schema cross-module compatibility and timestamp handling; Added unit tests for secret key migration and migration rehearsal seeding scripts; Added unit tests for starter project graphs; Added unit tests for starter project templates; Added unit tests for telemetry service boundaries and metrics; Added unit tests for template utility functions; Added unit tests for template utility functions; Added unit tests for the Agentic API assistant and file endpoints; Added unit tests for the EventManager component; Added unit tests for the GitLoaderComponent; Added unit tests for the JobService concurrency, liveness, and retention logic; Added unit tests for the NextPlaid bundle and vLLM multivector SSRF protections; Added unit tests for the authentication service layer; Added unit tests for the backend tools subsystem; Added unit tests for the background execution service; Added unit tests for the database-backed catalog policy service; Added unit tests for the experimental flow runner; Added unit tests for the policies component; Added unit tests for the transaction service module; Added unit tests for the trigger service; Added unit tests for tracing service components; Added unit tests for v1 API endpoints; Added unit tests for variable service and Kubernetes secret resolution; Added unit tests for vector store components; Expanded test coverage for lfx-bundles security, policy, and component contracts; Integration tests for S3 storage service and client reuse; New integration test suite for backend components and services; New integration test suite for backend components and services; New test infrastructure and stress-testing harness for backend reliability; Unit tests for GDPR data subject erase and request workflows; Unit tests for Knowledge Base ingestion sources and backend registry; Unit tests for connection service security, validation, and resolution logic.
Dependencies
1788 commits updating dependencies (40 manifests)
A dependency / build maintenance change in (dependencies) — 1788 commits (495 fixs), 40 files: “🔖 chore(pyproject.toml): bump up version to 0.0.81 The version number in the pyproject.toml file has been updated from 0.0.80 to 0.0.81. This is a chore commit as it does not introduce any new features or fix any bugs, but it is necessary to keep track of the version number of the package”; “chore: dup langchain-qdrant to langchain-mongodb”; “chore(pyproject.toml): update package version from 0.0.66 to 0.0.67”; and 1686 more (refs \#9478, \#11827, \#15757, \#14490, \#11111, …).
(dependencies) · high confidence · unverified
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 49 → 49 (-0.3)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.4) — scores are not directly comparable.
Lenses
- Code Health 68 → 66 (-2.5)
- Architecture 64 → 61 (-3.4)
- Maturity 85 → 96 (+11.2)
- Readiness 36 → 36 (-0.0)
- Security 60 → 60 (-0.2)
- Domain Modelling 100 → 100 (+0.0)
- Accessibility 56 → 53 (-2.5)
- Performance 60 → 70 (+10.0)
Resolved (305)
- ADR has no Consequences section (design/dedicated-integrations/decisions/broad-content-read.md)
- AuthService._authenticate_with_token (cognitive 16) (src/backend/base/langflow/services/auth/service.py)
- Boundary-crossing change coupling: custom_component.py ↔ base_component.py (src/backend/base/langflow/custom/custom_component/custom_component.py)
- Change coupling: azure_openai.py ↔ baidu_qianfan_chat.py (src/bundles/azure/src/lfx_azure/components/azure/azure_openai.py)
- Change coupling: mistral.py ↔ openai_chat_model.py (src/bundles/lfx-bundles/src/lfx_bundles/mistral/mistral.py)
- Change coupling: mongodb_atlas.py ↔ qdrant.py (src/bundles/lfx-bundles/src/lfx_bundles/mongodb/mongodb_atlas.py)
- Change coupling: qdrant.py ↔ vectara.py (src/bundles/lfx-bundles/src/lfx_bundles/qdrant/qdrant.py)
- Change-coupling hub: cohere_models.py → amazon_bedrock_model.py, anthropic.py, azure_openai.py, huggingface.py, mistral.py, vertexai.py, openai_chat_model.py (src/bundles/cohere/src/lfx_cohere/components/cohere/cohere_models.py)
- Change-coupling hub: huggingface.py → amazon_bedrock_model.py, anthropic.py, azure_openai.py, mistral.py, vertexai.py (src/bundles/lfx-bundles/src/lfx_bundles/huggingface/huggingface.py)
- Change-coupling hub: pgvector.py → mongodb_atlas.py, qdrant.py, vectara.py (src/bundles/lfx-bundles/src/lfx_bundles/pgvector/pgvector.py)
- Change-coupling hub: supabase.py → astradb_vectorstore.py, mongodb_atlas.py, pgvector.py, qdrant.py, vectara.py (src/bundles/lfx-bundles/src/lfx_bundles/supabase/supabase.py)
- Change-coupling hub: vertexai.py → amazon_bedrock_model.py, azure_openai.py, baidu_qianfan_chat.py, mistral.py (src/bundles/lfx-bundles/src/lfx_bundles/vertexai/vertexai.py)
- Checkable ADR rule enforced only by prose: Desktop OAuth registrations: Langflow-owned public clients by default (design/dedicated-integrations/decisions/desktop-oauth-ownership.md)
- Critical CVE: [GHSA redacted] (uv.lock)
- Critical CVE: [GHSA redacted] (uv.lock)
- Critical CVE: [GHSA redacted] (src/frontend/package-lock.json)
- Critical vulnerability: [GHSA redacted] (uv.lock)
- Critical vulnerability: [GHSA redacted] (src/frontend/package-lock.json)
- Documentation: other issue (docs/docs/API-Reference/README.md)
- Duplicated block (10 lines × 2) (src/lfx/src/lfx/services/mcp_composer/service.py)
- …and 285 more
New (596)
- AuthService.validate_tool_call (cognitive 23) (src/backend/base/langflow/services/auth/service.py)
- AuthService.validate_tool_call (cyclomatic 24) (src/backend/base/langflow/services/auth/service.py)
- Change coupling: amazon_bedrock_model.py ↔ vertexai.py (src/bundles/amazon/src/lfx_amazon/components/amazon/amazon_bedrock_model.py)
- Change coupling: azure_openai.py ↔ vertexai.py (src/bundles/azure/src/lfx_azure/components/azure/azure_openai.py)
- Change coupling: mongodb_atlas.py ↔ pgvector.py (src/bundles/lfx-bundles/src/lfx_bundles/mongodb/mongodb_atlas.py)
- Change coupling: pgvector.py ↔ qdrant.py (src/bundles/lfx-bundles/src/lfx_bundles/pgvector/pgvector.py)
- Change-coupling hub: cohere_models.py → anthropic.py, azure_openai.py, huggingface.py, mistral.py, vertexai.py (src/bundles/cohere/src/lfx_cohere/components/cohere/cohere_models.py)
- Change-coupling hub: huggingface.py → amazon_bedrock_model.py, mistral.py, vertexai.py (src/bundles/lfx-bundles/src/lfx_bundles/huggingface/huggingface.py)
- Change-coupling hub: supabase.py → astradb_vectorstore.py, mongodb_atlas.py, pgvector.py, qdrant.py (src/bundles/lfx-bundles/src/lfx_bundles/supabase/supabase.py)
- Duplicated block (10 lines × 2) (src/backend/base/langflow/serialization/serialization.py)
- Duplicated block (13 lines × 2) (src/lfx/src/lfx/utils/ssrf_requests.py)
- Duplicated block (19 lines × 2) (src/backend/base/langflow/serialization/serialization.py)
- Duplicated block (5 lines × 9) (src/backend/base/langflow/alembic/versions/369268b9af8b_add_job_id_to_vertex_build_create_job_.py)
- Edited copies of one member (3 members, 11+ corresponding lines) (scripts/gp/gp_client.py)
- Edited copies of one member (3 members, 11+ corresponding lines) (src/backend/base/langflow/api/v1/triggers.py)
- Edited copies of one member (3 members, 13+ corresponding lines) (src/bundles/microsoft/src/lfx_microsoft/components/microsoft/outlook_calendar_create.py)
- Edited copies of one member (3 members, 14+ corresponding lines) (scripts/ci/update_lf_base_dependency.py)
- Edited copies of one member (3 members, 15+ corresponding lines) (src/lfx/src/lfx/base/knowledge_bases/backends/destination_policy.py)
- Edited copies of one member (3 members, 16+ corresponding lines) (src/backend/base/langflow/services/auth/mcp_encryption.py)
- Edited copies of one member (3 members, 18+ corresponding lines) (src/backend/base/langflow/api/v1/memories.py)
- …and 576 more
Changes since last survey
- 114 commits — 27 feature/other, 87 fixes
By area
- src/backend — 44 commits
- src/lfx — 23 commits
- (root) — 11 commits
- (repo) — 9 commits
- src/frontend — 7 commits
- src/bundles — 6 commits
- scripts/ci — 5 commits
- .github/workflows — 4 commits
- docs/docs — 3 commits
- docker/install_hardened_npm.sh — 1 commit
- docs/package-lock.json — 1 commit
Notable commits
- fix: chore: merge release-1.13.2 Gunicorn fix into main
- fix: fix(agentic): check implicit calls in component definitions (#15943)
- fix: fix(agents): keep parallel tool results on their own run (#16233)
- fix: fix(amazon): address Bedrock Converse QA failures (#16326)
- fix: fix(api): recover milestones a slow event stream dropped (#16234)
- fix: fix(api): rename a flow that loses a concurrent create for its name (#16302)
- fix: fix(auth): restrict external resolver credentials (#15980)
- fix: fix(authz): refuse unavailable configured plugins (#16286)
- fix: fix(components): send If-Else message on default route (#16275)
- fix: fix(deps): dependency upgrades for 1.13.1 (#16131)
- fix: fix(deps): dependency upgrades for 1.13.2 (#16169)
- fix: fix(deps): update FastMCP to 3.4.10
- fix: fix(deps): use patched Zod in the docs toolchain (#16299)
- fix: fix(frontend): create one flow on a repeated New Flow click (#16232)
- fix: fix(graph): give list inputs the same value order in every process (#16285)
- fix: fix(lfx): apply Split Text Text Key to lists of Data records (#16164)
- fix: fix(lfx): keep non-string text values in Message (#16218)
- fix: fix(lfx): keep top-level JSON arrays in JSON Cleaner (#16206)
- fix: fix(lfx): report input length and format in Batch Run metadata (#16152)
- fix: fix(logging): redact credentials from query strings (#16042)
- …and 94 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
langflow-ai/langflow 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 14 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 260cbf347cc5563cc48c79094c175ca31532ea2b — the exact code this score is about.
- Scored under rubric-2026.11.4 — 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-2026.11.65+preprod-36d4cd83012f.