Skip to content
CAI
Software that uses CAICheck a score

HKUDS/DeepTutor

44.5

Weak · 15 November 2026

510.9k

lines of production code

Python

with TypeScript

5

measurements over time

CAI band scale
CAI trend line
CAI lens gauges

What this system is

Features

Add Invidious video browsing and timestamped mark management to the Watching view

Users can now browse and select videos from connected Invidious accounts directly within the Watching interface, with the browser preserving session state (view, query, scroll position) across visits. Additionally, a new marks panel allows users to view, filter, and manage timestamped video marks (key points, questions, reviews), accept AI-suggested marks, seek to specific times, and mark items as reviewed.

web/components/watching · high confidence

Add optional user authentication with login and registration pages

This change introduces the frontend UI for optional user authentication in the DeepTutor application. It adds a layout and dedicated pages for logging in and registering new accounts. The login page accepts either an email or a username, checks for existing sessions, and redirects first-time users to registration. The registration page validates password matches, informs the first user that they will receive admin privileges, and redirects to login upon success.

web/app/(auth) · high confidence

Add public device sign-in handoff page

A new handoff page has been added to support signing in on public devices. Users can now complete a session handoff by entering a pairing code manually or by automatically processing a code provided via URL parameters, which exchanges for a session ticket and redirects them to their intended destination.

web/app/handoff · high confidence

New components (AdminLink, ProfileLink, LogoutButton) provide sidebar navigation items that appear only when authentication is enabled. AdminLink shows a shield icon linking to /admin/users for users with admin privileges, ProfileLink displays the current user's avatar and username linking to /profile, and LogoutButton triggers the logout flow and redirects to /login. All links support collapsed/expanded states, use i18n for labels, and include accessible tooltips.

web/components/auth · high confidence

Added standalone asset package for DeepTutor Web

A new \\_\init\\_.py\ file has been added to the \deeptutor\_web\ package to support standalone release builds. This package is designed to hold Next.js standalone output artifacts, including \server.js\, \.next/static\, and \public\ directories, allowing the launcher to serve these assets directly when available, while falling back to the source \web/\ directory for development or source checkouts.

_deeptutor\web · high confidence

Admin controls for multi-user grants, book permissions, and learner profiles

Administrators can now manage multi-user access and learning configurations directly from the web UI. This change introduces dedicated editors for configuring user grants (including LLM models, MCP tools, and learning policies), setting granular book permissions (read/edit access per book), and editing learner profile details such as age, grade level, and curriculum. It also adds a Guardian Relationships editor to supervise learner accounts and manage guardian permissions, supported by new API endpoints for fetching and saving these resources.

web/features/multi-user · high confidence

Admin user management and runtime authentication

The web client now supports full user administration and runtime authentication. Admins can create, import, and delete users, assign roles (admin, teacher, student, user), and manage learner profiles (age, grade, curriculum). Authentication is resolved at runtime from the backend rather than build-time environment variables, enabling login, registration, and logout flows. The system also includes a user search filter and avatar management with deterministic fallbacks.

web/lib · high confidence

Centralized app-shell state and new interactive context providers

The application now manages core UI preferences—such as theme, interface language, sidebar state, and code-block display settings—through a new AppShellContext that persists these choices to localStorage and synchronizes them with backend settings. Additionally, new context providers have been introduced to support specific interactive features: QuizFollowupContext manages per-question follow-up chat sessions and their persistence, GeogebraTabContext enables opening GeoGebra visualizations in a side viewer, and ReadingContext handles the state of open reading materials and their annotations, ensuring this data survives navigation events.

web/context · high confidence

Co-writer workspace component implementation

The CoWriterWorkspace component has been introduced to manage the co-writer editing environment, providing users with document lifecycle management (loading, saving, autosave), title editing, and a rich toolbar for text formatting and actions like rewrite, shorten, and expand. It integrates model selection via the ModelSelector, supports knowledge base and web search tools for selection edits, and handles document export to DOCX.

web/features/co-writer/components · high confidence

Expose PDF attachment processing stages in chat UI

The chat interface now displays real-time processing stages for PDF attachments, including uploading, parsing, and retrieval phases. This change introduces a new selector that maps server-side stream events to user-visible status updates, allowing users to see exactly where a PDF is in the processing pipeline. Additionally, the UI now retains visibility of warnings and failures after the conversation turn settles, enabling users to inspect any issues that occurred during attachment handling.

web/features/chat/selectors · high confidence

Immersive reading workspace routes and layout

The reading section now uses a dedicated layout that provides Geogebra and quiz follow-up contexts, and new route pages for the library, materials, folders, sessions, and workspaces that render the corresponding immersive reading components.

web/app/(workspace)/learning/reading · high confidence

Initial project scaffolding and repository hygiene configuration

This change establishes the foundational structure for the DeepTutor project by adding essential repository configuration files. It introduces a \.dockerignore\ file to optimize Docker build contexts by excluding development artifacts, caches, and user data, and a \.gitignore\ file to prevent tracking of build outputs, virtual environments, and IDE settings. A \.pre-commit-config.yaml\ is added to enforce code quality through Ruff, Prettier, and security scanning via detect-secrets and Bandit. Additionally, it provides a \.env.example\ template for host-side port configuration, a \.gitattributes\ file to ensure consistent line endings for cross-platform compatibility, and an \AGENTS.md\ document detailing the agent-native architecture, CLI usage, and dependency layers for users and contributors.

(repo-wide) · high confidence

Introduce Agents utility console layout and page

Added a new layout and page component for the Agents utility section. The layout provides a full-height, scrollable container with a background color variable, while the page renders the AgentsHub component within a standard FeaturePage wrapper, establishing the visual structure for this new console area.

web/app/(utility)/agents · high confidence

Introduce Co-Writer workspace with document management and DOCX import

Users can now access the Co-Writer workspace to manage markdown drafts and projects. This new page allows users to create new documents from scratch or from a built-in sample template, import existing Word documents (.docx), and delete drafts. The interface displays a list of existing documents with relative timestamps and provides clear actions for starting new content or importing files.

web/app/(workspace)/co-writer · high confidence

Introduce chat workspace component and protocol mismatch notice

This change introduces the new ChatWorkspace component, which serves as the central container for the chat interface, integrating various sub-components like the composer, message list, and session viewer. Additionally, a new ProtocolMismatchNotice component is added to alert users when there is a version incompatibility between the client and server, providing a reload action to resolve the issue. Also in this area: 13 commits (10 fixes) fixing web; 1 commit (1 fix) fixing deeptutor/api, web/features/chat; 10 commits (4 fixes) modifying web.

web/features/chat/components · medium confidence · unverified

Introduce standalone DeepTutor CLI with agent-first capabilities and interactive chat

A new standalone \deeptutor-cli\ package provides a command-line interface for DeepTutor, featuring a \run\ command for single-turn agent execution of capabilities (such as \chat\, \deep\_solve\, \deep\_research\, and \visualize\) and a \chat\ command for interactive multi-turn REPL sessions. The CLI includes a \doctor\ command for setup diagnostics, an \init\ wizard for configuring LLM and embedding providers, and subcommands for managing knowledge bases (\kb\), sessions, notebooks, partners, plugins, and a three-layer memory subsystem. It also supports launching the full web/API stack via \start\ and \serve\ commands.

_deeptutor\cli · high confidence

Introduce structured group discussion modes and threaded sessions

This change introduces a new group discussion interface that replaces the previous single-transcript model with structured, threaded sessions. Users can now select from three distinct discussion modes—Parallel answers, One after another, and Cross-examination—which dictate how partner responses are generated and displayed. The interface supports multi-threaded discussions via a session picker, allowing users to create, switch between, and delete separate conversation threads. Key components include a GroupComposer for drafting messages with @mentions, GroupRound for visualizing parallel or sequential answers, and a GroupSidePanel for viewing private reasoning traces and shared context. The UI also features InvocationCards for peer-to-peer follow-up proposals and RoundSummaryAction for closing rounds with a consensus summary.

web/components/partners/group · high confidence

Introduce structured, versioned local storage layer

A new shared storage module has been added to the web application, providing a typed, versioned wrapper around browser local and session storage. This layer introduces schema validation, automatic data migration for version changes, and a subscription mechanism to sync state across tabs, ensuring that stored data remains consistent and recoverable even if formats evolve or storage races occur.

web/shared/storage · high confidence

Introduce three-layer memory subsystem with workbench and graph views

DeepTutor now organizes user memory across three distinct layers: L1 (a live workspace mirror across surfaces like chat and notebook), L2 (per-surface curated facts), and L3 (cross-surface synthesis such as user profile and recent summaries). This change adds a Memory Hub for high-level overview, a Memory Graph for visualizing relationships between memory nodes, and a Memory Workbench for managing L2/L3 documents with Update, Audit, and Dedup modes. A new MemoryArchivedBanner notifies users that v1 memory has been archived and v2 starts fresh, while the MemoryRunPanel handles background processing with real-time event streaming and reconnection logic.

web/components/memory · high confidence

Introduce unified agentic loop and chat pipeline

DeepTutor now uses a single-loop agent architecture for chat turns. A new \AgenticLoopPipeline\ drives the conversation, allowing the model to call tools and stream commentary in each round, ending only when it stops calling tools. This replaces the previous two-stage (exploring then responding) model with a more fluid, terminal-agent style interaction. The chat capability now binds to this loop, and shared utilities handle JSON parsing, tool composition, and workspace context injection.

deeptutor · high confidence

Introduces a four-theme design system with accessible CSS variables

The application now supports four distinct visual themes—Cream (default light), Dark, Snow (neutral light), and Glass (translucent dark)—defined by a comprehensive set of CSS custom properties in \globals.css\. This change establishes a unified design language using semantic tokens for backgrounds, surfaces, and interactive states, ensuring consistent styling across the UI. The \layout.tsx\ file integrates this system by applying the root CSS, configuring typography via Geist and Lora fonts, and injecting a \ThemeScript\ to handle theme persistence and SSR safety. Additionally, new TypeScript declarations in \file-system-access.d.ts\ provide ambient types for the File System Access API, enabling robust file handling capabilities within the chat import wizard.

web/app · high confidence

Introduction of Space utility page and layout

A new Space utility area has been added to the application, consisting of a layout component that wraps content in the SpaceMain component and a page component that renders the SpaceDashboard. This establishes the routing structure for the Space feature, allowing users to access the dashboard view at the /space route.

web/app/(utility)/space · high confidence

Introduction of the Guided Learning interface

A new Guided Learning feature is now available at the /learning route. This update introduces the layout and main dashboard page, which renders the LearningDashboard component to display module progress and allow users to manage their learning modules.

web/app/(workspace)/learning · high confidence

New Agents Hub with live local and remote agent connections

The Agents Hub page now displays a 'Connected agents' section that detects and lists supported local CLI agents (such as Claude Code, Codex, Grok CLI, Gemini, Kimi, and others) as well as remote Hermes gateways. Users can connect, disconnect, and manage these agents directly from the UI, with brand-specific glyphs and tooltips for each backend. This replaces the previous view with a unified interface for interacting with live agents in chat.

web/components/agents · high confidence

New Journal overview page for learning spaces

A new page at the /space/journal route has been added to provide a durable overview of the learning journal. This page renders the LearningJournalPanel component wrapped in a React Suspense boundary, allowing users to access their journal data within the space utility section.

web/app/(utility)/space/journal · high confidence

New MCP server management and catalog components

This change introduces a new set of React components in the web interface for managing Model Context Protocol (MCP) servers. It adds a deployment-wide admin registry for configuring global servers (including local stdio commands) and a per-user server list for managing individual accounts. The update includes a curated MCP catalog browser for discovering and installing servers, a unified server configuration form with connection testing, and a key-value editor that securely handles stored credentials by masking them and offering a replace action. These components replace the previous settings-based management approach with a dedicated, structured UI for server configuration and catalog browsing.

web/components/mcp · high confidence

New Math Animator Viewer component for displaying video, image, and code outputs

A new MathAnimatorViewer component has been added to the web interface, providing a dedicated UI for rendering results from the math animator. Users can now view generated video outputs, browse image artifacts in a grid with a fullscreen preview option, and inspect the underlying Manim code in a collapsible section. The component also displays metadata such as quality scores, retry counts, and processing timings, and includes a warning banner if a visual review check fails. Also in this area: 13 commits (6 fixes) modifying web.

web/components/math-animator · medium confidence · unverified

New Partners page for managing AI companions and groups

A new Partners page has been added to the workspace, providing a central list view for AI companions (partners) and the groups they form. Users can now see all their partners—including those shared with them—along with their running status and connected IM channels (such as Feishu, Telegram, and Slack). The page also displays existing partner groups and allows users to create new partners or groups directly from this interface, consolidating partner management and group convening into a single view.

web/app/(workspace)/partners · high confidence

New Space Questions page added

A new page at the space/questions route has been introduced, rendering the PracticePage component in 'library' mode. This page is wrapped in a Suspense boundary with a LearningSkeleton fallback, providing users with a dedicated interface to access the question library within the space context.

web/app/(utility)/space/questions · high confidence

New Space utility pages for CLI apps and MCP store

Added new pages at /space/cli-apps and /space/mcp that render the CliAppsSection and McpStoreSection components respectively, providing dedicated entry points for these utility features within the Space area.

web/app/(utility)/space/cli-apps, web/app/(utility)/space/mcp · high confidence

New Whisper dual-seat training components

Added three new React components to the Whisper module to support a dual-seat training workflow: WhisperComposer handles the input area with distinct placeholders and send controls for visitor and trainee roles; WhisperMessageList renders the conversation history with visual differentiation for whisper (private), debrief, and system messages; and WhisperRoomChip provides a copyable room ID for session coordination. All user-facing text in these components is now translatable via i18n.

web/components/whisper · high confidence

New accessible UI primitives: Button and ConfirmDialog components

The web components library now includes a new Button component and a ConfirmDialog component. The ConfirmDialog replaces native browser prompts with an accessible, keyboard-navigable modal that supports localization, busy states, and distinct visual tones for destructive actions, improving safety and usability for user management workflows.

web/components/ui · high confidence

New accessible design primitives and UI components

The web/shared/ui package now provides a suite of accessible, reusable React components including Button, Dialog, Field, IconButton, InlineAlert, Skeleton, StatusChip, EmptyState, and Tooltip. These components establish a consistent visual language and accessibility foundation, featuring proper ARIA attributes, keyboard navigation support (such as focus trapping in Dialog and accessible tooltips), and responsive styling via Tailwind CSS. The Tooltip component specifically handles complex interactions like nested controls and viewport positioning to ensure usability across different devices.

web/shared/ui · high confidence

New admin user management interface with batch operations and role controls

The admin section now includes a dedicated Users page that allows administrators to view, search, and manage user accounts. Key capabilities include creating new users with specific account presets, importing users in bulk, and deleting users individually or in batches. Administrators can also promote or demote user roles (admin vs. standard user) and manage learner-specific details such as guardian relationships and book permissions. The interface supports localization and includes confirmation dialogs for destructive actions to prevent accidental changes.

web/app/(admin) · high confidence

New annotation system with persistent highlights, citations, and AI actions

The reading workspace now features a comprehensive annotation layer that allows users to highlight, underline, and add notes to text selections, with marks persisting across sessions. A new annotation list panel organizes these marks by locator and includes a dedicated tab for citations, which are now stored alongside other annotations. The annotation popover has been redesigned to support color-coded highlights, underlines, notes, and citations, and integrates AI actions (such as 'Ask about this') directly into the toolbar. The system uses percentage-based positioning for annotations to ensure they remain accurate during zoom and resizing, and includes visual feedback like focused rings and flash animations for navigation.

web/components/reading · high confidence

New build and development tooling scripts for the web frontend

The web frontend now includes a suite of new build and development scripts in the \web/scripts\ directory. \build.mjs\ orchestrates the Next.js build, handling PDF.js asset copying, TypeScript configuration isolation, and standalone bundle packaging, while skipping forced Webpack on Vercel to prevent build failures. \build-brand-icons.mts\ generates a typed, minimal set of brand icons from the \simple-icons\ package to reduce bundle size. \dev.mjs\ fixes memory issues in the development server by correctly setting the V8 heap ceiling for the Next.js worker process. Additional scripts include \generate-contracts.mjs\ for typed backend contracts, \i18n\_audit.mjs\ and \i18n\_parity.mjs\ for translation coverage checks, \route\_budgets.mjs\ for performance budget enforcement, and \typecheck.mjs\ for isolated TypeScript validation.

web/scripts · high confidence

New chat history import for Claude Code, Codex, and ChatGPT

Users can now import conversation history from local agent folders and exported files. The \web/lib/chat-import\ module adds adapters for Claude Code (scanning \\~/.claude/projects\), Codex (scanning \\~/.codex/sessions\ and filtering out harness-injected context turns), and ChatGPT (parsing exported JSON conversations). Imported sessions are organized into named agents with scoped sync (by project for Claude Code, by calendar day for Codex), and attribution logic maps backend sessions back to these agents for display in the Space management page and chat reference picker.

web/lib/chat-import · high confidence

New chat home composer with agent selection, resource management, and attachment processing

The chat home interface now features a comprehensive composer toolbar that allows users to select connected agents (such as Claude Code or Codex) to route turns through subagent capabilities, manage context resources like knowledge bases and books, and monitor attachment processing stages (upload, parsing, retrieval) with detailed status indicators. The composer includes a new AskUserOptions component for handling interactive multi-question cards, a CapabilityConfigCard for configuring tools like Quiz and Visualize, and a ChatMarkdownNoteTab for saving inline notes. These changes replace the previous unified chat context path with a more modular, resource-aware input surface.

web/components/chat/home · high confidence

New chat message presentation and token usage tracking components

The chat interface now uses a new \ChatMessageList\ component to render messages, introducing a \UsageFooter\ that displays token consumption and performance metrics (such as cache hit rate, TTFT, and tokens per second) for individual turns or the entire session. This change also adds a \usage-summary\ module to aggregate and calculate these metrics from stream events, replacing previous presentation logic with a more detailed, accessible, and performant message rendering system.

web/features/chat/messages · high confidence

New chat resource pickers for books, history, agents, personas, and more

The chat composer now includes dedicated picker dialogs for selecting various reference sources: Book chapters, Reading materials, Question Bank entries, Chat History sessions, My Agents (imported conversations), Personas, and Memory artifacts. These new components (BookReferencePicker, ReadingReferencePicker, QuestionBankPicker, HistorySessionPicker, MyAgentsPicker, PersonaPicker, MemoryPicker) provide a unified interface for attaching context to chat turns, featuring search, filtering, and preview capabilities. The ChatSpaceMenu has been updated to expose these new resource types alongside existing attachment and knowledge options.

web/components/chat · high confidence

New client-side file previewers for documents, spreadsheets, and code

The chat preview panel now includes dedicated components for rendering various file types directly in the browser. Users can view DOCX files (with layout-aware scaling), XLSX spreadsheets (with sheet tabs and row/column limits), Markdown, plain text, and code files (with syntax highlighting). Office files (PPTX, legacy Office) fall back to extracted text when PDF conversion is unavailable, while images, SVGs, and videos use native browser rendering. A new FallbackPreview handles unsupported or legacy files with a download option. All previews use authenticated API calls and enforce size limits to prevent performance issues.

web/components/chat/preview/previewers · high confidence

New client-side i18n infrastructure with support for Chinese, French, German, and Ukrainian

The web interface now uses a new client-side internationalization system (I18nClientBridge and I18nProvider) that initializes i18next and dynamically loads language bundles. This change adds full interface support for Chinese (zh), French (fr), German (de), and Ukrainian (uk) alongside the existing English (en) default, with language switching handled via the AppShell context and the HTML lang attribute kept in sync for accessibility. Also in this area: 13 commits (6 fixes) modifying web.

web/i18n · medium confidence · unverified

New co-writer document page route

A new page component has been added at the co-writer/\[docId\] route to render the collaborative writing workspace. This page extracts the document ID from the URL parameters and passes it to the CoWriterWorkspace component, enabling users to access specific documents within the co-writer feature.

web/app/(workspace)/co-writer/\[docId\] · high confidence

New dual-seat Whisper page with real-time chat

A new Whisper page has been added to the workspace, enabling a dual-seat interaction model where both 'visitor' and 'trainee' roles can participate in the same session. The page uses the UnifiedTurnClient for real-time streaming and filters messages based on the current seat, allowing users to see context-specific conversations. It includes features like crisis detection, room management, and translation support via react-i18next.

web/app/(workspace)/whisper · high confidence

New file preview drawer for chat attachments

A new slide-in drawer component has been added to the chat interface, allowing users to preview attached files without leaving the conversation. The system supports a wide range of file types—including PDFs, images, videos, SVGs, Markdown, code, Office documents (Word, Excel, PowerPoint), and plain text—by dynamically loading the appropriate renderer only when needed to maintain performance. Users can also open supported documents (such as PDFs, EPUBs, and Office files) directly in the immersive reading workspace for a distraction-free experience.

web/components/chat/preview · high confidence

New follow-up chat capability for quiz questions

Users can now engage in a dedicated chat conversation about a specific quiz question, its answer, and the AI's judgment. This feature introduces a new 'Follow-up Chat' tab within the Quiz Viewer, allowing learners to ask clarifying questions or request deeper explanations without leaving the quiz context. The implementation includes a new \FollowupChatComposer\ component that mirrors the main chat interface, a \QuizFollowupTabBody\ for displaying the threaded conversation, and a \QuizConfigPanel\ for configuring quiz parameters. The system supports persisting chat history across sessions and handles various question types, including coding and selection-tutor scenarios.

web/components/quiz · high confidence

New immersive reading workspace with media support and AI companion

The reading workspace has been replaced with a new immersive interface that integrates a chat-based AI companion directly alongside the reading material. This update introduces a unified composer for asking questions, grounding them in selected passages, and managing linked conversation sessions. It adds a dedicated media reading stage supporting YouTube and Bilibili video players with synchronized transcript search and navigation, alongside a read-aloud toggle and page-specific action buttons (such as quizzes). The layout features a resizable companion panel, a consolidated workspace menu, and persistent state for bookmarks and annotations, creating a cohesive environment for interactive reading and study. Also in this area: 13 commits (10 fixes) fixing web; 25 commits (11 fixes) modifying web.

web/components/reading/workspace · medium confidence · unverified

New knowledge base creation and management interface

The Knowledge Center now features a comprehensive set of new UI components for managing knowledge bases. Users can create new bases or link existing ones via a new modal that supports multiple providers including LightRAG, WeKnora, Tencent IMA, Obsidian, MarginNote 4, and Kiwix ZIM archives. The interface includes an embedding model selector to pin specific models to a base, a file drop zone for uploads, and a document list with folder management capabilities. Additionally, a dedicated preview pane allows users to view supported file types directly within the knowledge base view.

web/components/knowledge · high confidence

New learning space components for conversation history, agents, and skills

This change introduces a suite of new UI components in the learning space: ArchivedConversations and ChatHistorySection for browsing and managing conversation history (including filtering, searching, and restoring archived chats); MyAgentsSection and ImportWizard for importing and managing imported chat agents; ScopeEditorModal and ScopePicker for defining and editing the scope of imported agents; McpStoreSection for managing MCP service connections; PersonasSection for creating and editing personas; and EduHubImportModal for importing skills from the EduHub catalog.

web/components/space · high confidence

New mastery study workflows and visual learning board

The learning space now features a complete v2 mastery workflow, replacing the previous static outline editor with a two-step creation wizard (goal and materials) where the study route is designed interactively with the tutor. A new visual learning board displays modules as columns and knowledge points as cards with mastery status and review timing. The study screen supports three distinct modes (outline, study, review) with a mode switcher, and includes a coverage notice to show which files were included in the generated outline. Additional components include a learner profile card, a level-up celebration animation, and a dedicated mastery composer. Also in this area: 13 commits (10 fixes) fixing web; 13 commits (6 fixes) modifying web.

web/components/space/learning · medium confidence · unverified

New memory subsystem pages and resolver routing

This change introduces the frontend pages for the new three-layer memory subsystem. It adds a Memory Hub landing page, dedicated workbench pages for L1 (with deep-link support for surfaces like chat, notebook, quiz, kb, book, partner, and cowriter), L2 (scoped by surface), and L3 (scoped by slot: recent, profile, scope). It also adds a resolver page that handles citation redirects by fetching entry details from the backend and navigating to the appropriate L2 or L3 workbench with a focus target.

web/app/(utility)/memory · high confidence

New operational scripts for architecture checks, Docker port mapping, and optional dependency installation

The scripts directory now includes several new tools to improve development hygiene and deployment reliability. A pre-commit hook and associated scripts (check\_architecture, check\_repo\_hygiene, check\_branch\_policy, check\_workspace\_hygiene) enforce dependency boundaries, prevent direct commits to main, and reject tracked generated files. A new docker\_compose.py wrapper reads port settings from JSON configuration to ensure consistent Docker environment variable injection. Additionally, install\_extras.py ensures optional dependencies are installed on every container start, and prepare\_web\_package.py handles the packaging of the Next.js frontend build.

scripts · high confidence

New page routes for Knowledge Bases, Notebooks, and Chat with secure return-url handling

The application now includes dedicated page components for the Knowledge Bases, Notebooks, and Chat sections. The Knowledge Bases page wraps the KnowledgePage component in a Suspense boundary to handle loading states. The Notebooks section introduces a route that resolves course scope from URL parameters and passes it to the NotebookConsole, ensuring notebooks created within a course context are correctly scoped. The Chat section provides entry points for both the main chat interface and specific chat sessions, rendering the ChatWorkspace component. Additionally, a new shared authentication utility normalizes return URLs to prevent open-redirect vulnerabilities by validating origins and sanitizing paths, while also supporting the inheritance of URL fragments from login redirects.

web/app/(utility)/knowledge-bases, web/app/(workspace)/chat, web/shared/auth · high confidence

New partner configuration and management UI components

The partners workspace now includes a suite of new UI components for managing partner identity, channels, and history. Users can now customize partner avatars with emojis, colors, or uploaded images via the FaceEditor, and select knowledge bases, skills, and notebooks to equip partners using the AssetPicker. Channel management is enhanced with a schema-driven configuration panel, live runtime status indicators, and a QR-based onboarding flow for supported channels. Additionally, a new PartnerArchives view allows users to browse, resume, and delete past partner conversations.

web/components/partners · high confidence

New partner creation wizard

Users can now create new partners through a five-step guided interface (Identity, Soul, Mind, Library, Review) that allows configuring the partner's name, description, avatar, language, personality source, selected LLM model, and available tools before launching them into the chat. Also in this area: 13 commits (6 fixes) modifying web.

web/app/(workspace)/partners/new · medium confidence · unverified

New personal learning progress view in settings

A new 'Learning progress' page has been added under the settings section, allowing users to review their own reading and learning activity. The view displays a summary of total materials and recent learning actions, followed by detailed sections for reading progress (showing the furthest percentage reached for each material with a progress bar) and a list of recent learning actions including the material, action type, extension, and locator.

web/app/(utility)/settings · high confidence

New practice workflow with question import, spaced-repetition sessions, and activity analytics

The practice area now supports importing questions via XLSX, CSV, TSV, or JSON files into either the Question Bank or Mistakes collection, with a preview step to validate data before committing. Users can launch daily review sessions that pull from a spaced-repetition queue, allowing them to answer questions, receive immediate feedback, and rate their confidence (Again, Hard, Good, Easy) to schedule future reviews. A new Practice Insights panel displays daily activity charts for new questions, new mistakes, and review submissions over 7, 30, or 90-day periods, with options to switch between line and bar chart styles. The interface also distinguishes between a general review home (across all workspaces) and a scoped library view (per course), ensuring that imported questions and review activities are correctly attributed to their respective collections.

web/components/learning/practice · high confidence

New public device sign-in page for secure session handoff

A new 'Public device sign-in' page has been added to the application, allowing authenticated users to generate a secure, time-limited handoff link for signing in on a public device. Users enter the public HTTPS origin, and the system creates a unique handoff code and URL that can be copied and shared. The generated link expires after a set number of seconds, providing a secure method for transferring session context between devices.

web/app/(utility)/session-handoff · high confidence

New research configuration and outline editing components

Added \ResearchConfigPanel\ and \ResearchOutlineEditor\ components to the web interface. The config panel allows users to select research modes (Study Notes, Report, Comparison, Learning Path) and depths (Quick, Standard, Deep, Manual), with manual mode exposing sliders for sub-topics and iterations. The outline editor enables users to review, edit, add, and remove sub-topics before confirming the research plan, displaying real-time status indicators for researching, complete, or failed states.

web/components/research · high confidence

New runtime health dashboard and multi-worker coordination settings

This release introduces a new Runtime Health Card component that displays the operational status of the turn runtime (healthy, degraded, recovering, or unavailable) along with detailed metrics such as worker count, coordination mode, and leader health. It also adds a Turn Coordination Settings interface, allowing operators to configure backend worker counts, switch between in-process and Redis coordination, and manage Redis secrets and advanced lease/recovery intervals. These changes provide visibility into multi-worker runtime health and enable safe configuration of turn coordination backends directly from the web UI.

web/features/runtime-status · high confidence

New settings components for chat timeout, CodeBuddy/Codex auth, connections, and model capabilities

The settings interface now includes dedicated sections for configuring chat response timeouts, managing CodeBuddy and Codex OAuth authentication, editing credential connections, and overriding model capabilities. Users can set a per-user chat idle timeout (10–300 seconds) that takes effect immediately without a restart, authenticate and manage sessions for CodeBuddy and Codex providers via OAuth cards, and create reusable credential connections that link to multiple services (LLM, task, embedding, search, TTS, STT, image, video) to avoid pasting keys across multiple pages. Additionally, model capability fields allow overriding tool calling, vision, JSON output, and reasoning support per model, while embedding dimension fields support auto-detection and custom values, and memory usage monitoring provides live system memory readouts for backend, frontend, and sandbox processes.

web/components/settings · high confidence

New shared UI components for assistant responses, pickers, and file handling

The \web/components/common\ directory now includes a suite of new shared components that standardize and enhance the user interface. Assistant responses are rendered via \AssistantResponse\, which supports smooth streaming, model thinking blocks, and citation linkification. Picker overlays (e.g., History, Books) now use \PickerShell\ and \PickerHeader\ for consistent dialog behavior, focus management, and animations. File interactions are improved with \InlineFileCard\ for clickable generated file links in chat, and \BrandIcon\ for standardized service logos. Additional utilities include \Modal\ for accessible dialogs, \McpToolGroups\ for provider-based tool selection, \ProcessLogs\ for terminal-style output, and \ProviderIcon\ for vendor-specific branding.

web/components/common · high confidence

New user profile page with avatar customization

A new profile page has been added to the web application, allowing users to view their profile information and manage their avatar. Users can now upload custom images, which are automatically cropped to a square and downscaled to 256x256 pixels for efficient storage, or select from a set of predefined icons and colors. The page handles authentication status, redirecting unauthenticated users to the login page, and provides options to remove the avatar image or sign out. Also in this area: 22 commits (9 fixes) modifying web.

web/app/(utility)/profile · medium confidence · unverified

New visualization capability with Chart.js, SVG, and HTML rendering

Users can now generate and view visualizations directly in the chat. This adds a new rendering pipeline supporting Chart.js for charts, inline SVGs with theme-aware styling (light/dark mode), and HTML content rendered in sandboxes. The \VisualizeConfigPanel\ allows users to select render modes (Chart.js, SVG, Mermaid, HTML, Animation, Storyboard) and manage bundled visualizers, while the \VisualizationViewer\ handles the actual rendering, including fullscreen overlays and iframe height adjustments.

web/components/visualize · high confidence

New visualizer components and UI infrastructure for sessions and themes

This update introduces several new React components to the web interface. Users can now view interactive mathematical visualizations via the new Geogebra component, which loads the GeoGebra applet and executes commands. Diagrams are now supported through a new Mermaid component that renders charts and dynamically updates colors to match the active application theme. The sidebar experience is enhanced with a new SessionList component that groups conversations by time, supports inline renaming, and displays unread indicators. Additionally, theme persistence is now handled by a ThemeScript component that applies the saved theme before React hydration to prevent flash-of-wrong-theme, and user identity is represented by a new UserAvatar component that supports custom icons, colors, and image uploads with admin-specific styling.

web/components · high confidence

Opt-in cross-browser partner conversation continuity

The Partner detail page now supports keeping a partner conversation synchronized across multiple browser tabs or windows. Users can enable this feature to maintain a single active session context, allowing changes made in one tab (such as switching conversations) to be reflected in others. The implementation includes automatic polling to detect session updates from other clients and handles edge cases like busy states and visibility changes to ensure smooth operation.

web/app/(workspace)/partners/\[partnerId\] · high confidence

Unified knowledge engine configuration and source management

The knowledge feature now provides a consolidated interface for managing retrieval engines and data sources. Users can configure detailed settings for local and hosted engines—including LightRAG, GraphRAG, LlamaIndex, and Tencent IMA—via a new EngineDetail component that exposes model selection, retrieval modes, and preflight checks. The system also introduces robust source management, allowing users to connect and sync linked folders, web sources, and GitHub repositories, alongside support for Kiwix ZIM archives and IMA knowledge bases.

web/features/knowledge · high confidence

Web app scaffolding and mobile-responsive layout

The web directory has been initialized with a Next.js application structure, including configuration for TypeScript, ESLint, Prettier, Tailwind CSS, and testing (Playwright, Vitest). A new AppShell component introduces a mobile-responsive layout where the sidebar becomes an overlay drawer on smaller screens, and a public handoff page enables secure session transfers via pairing codes. Additionally, a temporary avatar preview page has been added to visualize session states.

web · high confidence

Architecture

Introduce ChatRuntimeProvider for v2 runtime scoping

The chat feature now exposes a ChatRuntimeProvider component that establishes a runtime context for the chat UI. This provider wraps the ChatStateAdapterProvider and enforces that only a single runtime instance exists per route subtree, preventing nested providers. This change supports the migration to the v2 runtime provider architecture by centralizing the chat state initialization and ensuring proper scoping of the chat session state.

web/features/chat · high confidence

Behavioural changes

1 commit (0 fixes) modifying web/app/(utility)/personalization, web/app/(workspace)/calendar

A change to existing behaviour in web/app/(utility)/personalization, web/app/(workspace)/calendar — 1 commit, 2 files: “feat: preserve TypeScript engine migration and workspace updates”.

web/app/(utility)/personalization, web/app/(workspace)/calendar, web/components/video-lesson · low confidence · unverified

10 commits (4 fixes) modifying web

A change to existing behaviour in web — 10 commits (4 fixs), 58 files: “feat: add conversational creation drafts and refresh local dev UI”; “fix(web): decode the partner id from the route”; “fix: repair CJK emphasis boundaries for Japanese and Korean replies too”; and 7 more. Also in this area: Added smoke tests for ChatWorkspace and StandaloneComposer. Also in this area: 13 commits (10 fixes) fixing web.

(repo-wide) · medium confidence · unverified

18 commits (8 fixes) modifying docs-for-dev

A change to existing behaviour in docs-for-dev — 18 commits (8 fixs), 32 files: “fix: preserve dotted identifiers in BM25 retrieval”; “fix(integration): complete cancellation, documentation and browser regressions”; “docs: consolidate developer and agent handbook”; and 15 more.

docs-for-dev · medium confidence · unverified

3 commits (0 fixes) modifying web/app/(workspace)/learning/mastery/\[pathId\]/sessions

A change to existing behaviour in web/app/(workspace)/learning/mastery/\[pathId\]/sessions — 3 commits, 3 files: “feat: update notes, chat interactions and workspace UI on preview”; “feat(reading): complete Watching convergence with safe legacy migration (\#1378)”; “release: v1.6.9” (refs \#1378).

web/app/(workspace)/learning/mastery/\[pathId\]/sessions · medium confidence · unverified

4 commits (1 fix) modifying apps/desktop

A change to existing behaviour in apps/desktop — 4 commits (1 fix), 41 files: “fix(desktop): support Unicode paths during Windows first launch”; “feat(desktop): add Open DeepTutor for macOS under apps/desktop”; “feat(desktop): add Intel Mac and Windows editions with platform builds”; and 1 more. Also in this area: 1 commit (1 fix) fixing apps/desktop. Also in this area: 1 commit (1 fix) fixing apps/desktop; 4 commits (2 fixes) modifying apps/desktop.

(repo-wide) · medium confidence · unverified

5 commits (0 fixes) modifying web/app/(workspace)/learning/mastery

A change to existing behaviour in web/app/(workspace)/learning/mastery — 5 commits, 2 files: “feat: update notes, chat interactions and workspace UI on preview”; “feat: preserve TypeScript engine migration and workspace updates”; “feat: choose the model before a MasteryPath outline starts”; and 2 more (refs \#2407).

web/app/(workspace)/learning/mastery · medium confidence · unverified

5 commits (2 fixes) modifying web/components/cli-apps

A change to existing behaviour in web/components/cli-apps — 5 commits (2 fixs), 1 file: “release: v1.5.8”; “fix(i18n): format dates in the interface language”; “release: v1.5.7”; and 2 more (refs \#1526).

web/components/cli-apps · medium confidence · unverified

Centralized API client with structured error handling and retry logic

The web application now uses a new shared API client (\web/shared/api/client.ts\) that centralizes HTTP requests, authentication redirects, and error normalization. Users benefit from consistent error reporting, including correlation IDs for support, and improved retry behavior where the system now respects explicit \retryable\ flags from the server response while defaulting to retry for network errors, timeouts (408), rate limits (429), and server errors (5xx). The client also handles JSON, void, and blob responses uniformly, and integrates with the authentication system to redirect to login on 401 errors when appropriate.

web/shared/api · high confidence

Chat model introduces v2 turn protocol types and explicit reply language support

The chat model layer now defines a new v2 turn protocol with UI-facing type aliases that normalize wire schema fields (such as content, metadata, and source) for surface consumption, alongside unversioned command types for start, subscribe, resume, cancel, regenerate, and reply actions. Additionally, the start-turn input model now includes a \replyLanguageOverride\ field, allowing users to explicitly set the language for a specific turn's response, which persists per conversation as indicated by the associated fix commit.

web/features/chat/model · high confidence

Co-writer drafts now persist across storage-full errors

The co-writer's draft storage logic has been refactored to handle cases where local storage is full. The new \drafts.ts\ module introduces a versioned storage format (v2) and includes migration logic to move legacy drafts. Crucially, the \saveDraft\ function now catches storage errors and returns \null\ instead of throwing, allowing the application to retain draft data gracefully even when the underlying storage mechanism fails due to capacity limits.

web/features/co-writer/storage · high confidence

Enforce strict validation for turn protocol commands and events

The web client now strictly validates turn protocol interactions against version 2.0 boundaries. Command builders in \turn-command.ts\ enforce non-empty IDs, non-negative sequence numbers, and required fields (e.g., text or answers for replies), while \turn-event.ts\ parses incoming server events by verifying protocol version, event types, and field structures, rejecting invalid or unsupported frames with diagnostic details. This ensures that malformed commands are caught before transmission and that the client reliably handles or reports protocol errors from the server.

web/contracts/parse · high confidence

Introduce multi-worker turn runtime client with unified transport layer

The chat transport layer has been replaced with a new multi-worker turn runtime client. This update introduces a dedicated TurnRuntimeClient that manages WebSocket connections, handles reconnection with exponential backoff and jitter, and ensures reliable command delivery with acknowledgment timeouts. A UnifiedTurnClient provides a simplified surface API that translates between the chat model and the new turn protocol, including robust handling of protocol errors and stream events. The implementation also includes a fallback mechanism for generating unique command IDs when crypto.randomUUID is unavailable, ensuring compatibility across different browser environments.

web/features/chat/transport · high confidence

New web hooks for chat, knowledge, and UI state management

This change introduces a suite of new React hooks in the web application to centralize and improve specific user-facing behaviors. \useChatAutoScroll\ replaces the previous scrolling logic with a jitter-free, layout-effect-based approach that correctly handles streaming, late-mounting viewers, and user interruptions. \useAuthStatus\ shifts authentication state resolution to runtime, allowing the frontend to adapt to backend auth settings without rebuilds. \useKnowledgeBases\ and \useKnowledgeHistory\ provide a unified interface for managing knowledge base operations, progress, and local history. Additionally, \useCardSubmission\ handles the three-state lifecycle of question card answers, \useLingerExpand\ improves toolbar selector usability with delayed collapsing, and \useDevice\ standardizes responsive layout branching across the application.

web/hooks · high confidence

Question Bank rebuilt with tagging, filing, and advanced filtering

The Question Bank interface has been completely rebuilt to support organizing questions into custom categories. Users can now create, rename, and delete tags, and file questions into them directly from the card view or a multi-selection bar. The new layout includes a scope rail to filter by status (All, Wrong, Needs Review, Bookmarked, Untagged) and a toolbar with search, sort, and filters for source, material, and score trend.

web/components/space/question-bank · high confidence

Redesigned chat trace presentation with phase-aware activity indicators

The chat trace view has been refactored to provide a clearer, more semantic representation of the AI's internal process. A new \ActivityOrb\ component maps specific streaming modes (such as reasoning, planning, exploring, and tool usage) to distinct visual states and animation speeds, giving users immediate feedback on what the agent is currently doing. The \TracePresentation\ component now organizes these events into a structured flow, including dedicated panels for research stages (understand, decompose, evidence, result) and detailed call traces. Additionally, a new \memory.ts\ module introduces logic to compact and bound trace events for previews, ensuring performance while preserving critical context like user questions and tool results.

web/features/chat/trace · high confidence

Refactored chat controller logic for improved input validation and session handling

The chat controllers have been restructured to improve reliability and clarity. A new \buildStartTurnInput\ controller centralizes the assembly of turn requests, adding strict validation for runtime-only configuration keys and ensuring that explicit reply language overrides are correctly persisted per conversation. Attachment handling is now managed by a dedicated \pending-attachments\ controller that enforces file type, size, and quota limits before submission. Session management is also refined: a new \useChatRouteSession\ hook standardizes session ID extraction from route parameters, and a \waitForReplyLanguageSave\ utility ensures that composer drafts are restored if language persistence fails, preventing data loss during submission.

web/features/chat/controllers · high confidence

Refactored co-writer into modular hooks and a pure state model

The co-writer feature has been restructured to separate concerns, introducing dedicated React hooks for document lifecycle management, selection editing, split-pane resizing with local storage persistence, and synchronized scrolling. A new pure state model in the editor-state module now handles undo/redo history, text replacement, and scroll marker interpolation, providing a cleaner foundation for the co-writer's interactive editing capabilities.

web/features/co-writer/hooks, web/features/co-writer/model · high confidence

Restructured settings and learning routes for Next.js App Router

The application's routing structure has been updated to align with the Next.js App Router conventions. The settings area now uses a dedicated layout that wraps content with capability access providers and a guided tour overlay, while the entry point handles client-side redirection for legacy bookmarks. In the learning workspace, the book reader and practice pages have been reorganized into new route segments, with the practice page wrapped in a Suspense boundary to display a loading skeleton during data fetching.

web/app/(settings), web/app/(workspace)/learning/books/\[bookId\], web/app/(workspace)/learning/practice · high confidence

Settings page restructured into feature-owned sections with granular access control

The Settings interface has been refactored to organize options into distinct, feature-owned sections (such as Usage Activity, Appearance, Archived Chats, Attachments, Capabilities, and Data Migration) rather than a single flat list. This change introduces a new navigation and access-control system that dynamically shows or hides settings based on the user's role: admin-only pages are hidden from standard users, while specific sections like Learner Profile and Guardian settings are only visible to accounts with those respective presets. The new layout also includes a dedicated Usage Activity calendar, improved scrolling behavior within the settings pane, and a master-detail view for wide configuration pages. Also in this area: 3 commits (2 fixes) fixing web/features/settings; Settings store refactored into feature-owned context providers; Settings page restructured into feature-owned sections with new usage activity visualization; Restructured settings navigation with role-based access and isolated scrolling.

web/features/settings · low confidence · unverified

The sidebar has been restructured to support a customizable, drag-and-drop navigation layout where users can reorder and collapse workspace features into a 'More' menu. Session rows now display distinct status avatars (running, unread, failed, idle) with refined animations and colors, replacing the previous per-session icon identity. A new version badge in the footer indicates app update status, and the navigation entries now support capability-based locking for features like LLM access.

web/components/sidebar · high confidence

Unified activity vocabulary for consistent work-in-progress display

Replaced the previous inconsistent status indicators (thirteen hand-drawn glyphs and varied spinners) with a single, shared activity vocabulary used across chat, book compilation, co-writer runs, and sidebar sessions. This introduces a standardized two-level layout: a compact first level showing one line per action with a unified status dot (replacing semantic icons with a single atom of the thought-orb's dot cloud), and a second level for details that folds away once work settles. The new components include \ActivityHeader\ (orb + label + duration), \ActivityStack\ (aligned row grid), \ActivityRow\ (expandable action line), \ActivityFold\ (animated disclosure), \ActivityDetailGrid\ (label/value pairs with plumbing argument filtering), and \ActivityMark\ (for list rows without a header orb). All surfaces now share the same animation curves, color semantics (blue for running, amber for awaiting, red for error), and interaction patterns, ensuring visual consistency and reducing cognitive load when tracking ongoing work.

web/components/activity · high confidence

Unified notebook management with cross-notebook record selection

The notebook interface has been rebuilt to support a centralized library view and cross-notebook record selection. Users can now manage notebooks directly from the console, including creating, renaming, and deleting notebooks, as well as editing, deleting, moving, or copying individual records between notebooks. A new modal allows saving content to a notebook with the option to select specific messages from a chat transcript. Additionally, a new picker enables selecting records from multiple notebooks to ground subsequent requests, improving workflow continuity across different notebook collections.

web/components/notebook · high confidence

Workspace layout restructured with new context providers and runtime boundaries

The workspace layout has been reorganized to wrap the application shell with specific context providers for chat, reading, and watching capabilities, along with a new WorkspaceRuntimeBoundary and Suspense wrapper to preserve state during navigation. A new utility layout has also been introduced to handle sidebar and access control for utility sections, while the workspace root page now explicitly redirects to /chat.

web/app/(workspace) · high confidence

Fixes

1 commit (1 fix) fixing web/app/(workspace)/learning/new

A fix in web/app/(workspace)/learning/new — 1 commit (1 fix), 1 file: “fix”.

web/app/(workspace)/learning/new · low confidence · unverified

13 commits (10 fixes) fixing web

A fix in web — 13 commits (10 fixs), 63 files: “style: trim trailing blank lines in issue regressions”; “fix(chat): restore account-scoped side chat tabs”; “fix(visualize): inherit the application color scheme in HTML widgets”; and 10 more. Also in this area: Added tests for VisualizationViewer render branches.

web/tests/visualize · medium confidence · unverified

15 commits (8 fixes) fixing web/app/(workspace)/learning/books/components

A fix in web/app/(workspace)/learning/books/components — 15 commits (8 fixs), 13 files: “fix: render and resume book page chat questions”; “fix(web): migrate learning and reading button title hints (\#1526)”; “feat: unify page chrome and refine DeepTutor navigation”; and 12 more (refs \#1526, \#1648).

web/app/(workspace)/learning/books/components · medium confidence · unverified

6 commits (4 fixes) fixing web/app/(workspace)/learning/books/components/blocks

A fix in web/app/(workspace)/learning/books/components/blocks — 6 commits (4 fixs), 16 files: “fix(web): migrate learning and reading button title hints (\#1526)”; “fix(web): expose learning navigation and progress hints (\#1526)”; “feat: preserve TypeScript engine migration and workspace updates”; and 3 more (refs \#1526).

web/app/(workspace)/learning/books/components/blocks · medium confidence · unverified

Test coverage

1 commit adding/updating tests in tests/api/utils; 1 commit adding/updating tests in tests/events; 1 commit adding/updating tests in tests/fixtures/memory\_evaluation; 1 commit adding/updating tests in tests/fixtures/web; 1 commit adding/updating tests in tests/services/imagegen, tests/services/videogen; 10 commits adding/updating tests in tests/services/web\_source; 12 commits adding/updating tests in tests/plugins; 14 commits adding/updating tests in tests/services/sandbox; 178 commits adding/updating tests in tests/services/llm; 2 commits adding/updating tests in tests/fixtures/lightrag\_bridge; 2 commits adding/updating tests in tests/services/fetch; 2 commits adding/updating tests in tests/services/learning\_journal; 2 commits adding/updating tests in tests/tools/builtin; 34 commits adding/updating tests in tests/services/workspace; 38 commits adding/updating tests in tests/services/memory; 4 commits adding/updating tests in tests/i18n; 5 commits adding/updating tests in tests/co\_writer; 54 commits adding/updating tests in tests/cli; 6 commits adding/updating tests in tests/fixtures; 64 commits adding/updating tests in tests/tools; 65 commits adding/updating tests in tests/runtime; 7 commits adding/updating tests in tests/services/cli\_apps; 9 commits adding/updating tests in tests/api/routers; 9 commits adding/updating tests in tests/services/mcp; Added comprehensive test coverage for the Book module; Added comprehensive test coverage for the chat agent pipeline; Added regression tests for the vision solver agent; Added rendered component test harness; Added smoke and grading interaction tests for QuizViewer; Added test coverage for deployment and utility scripts; Added test coverage for new and existing loop capabilities; Added test coverage for partner groups, channel linking, and channel management; Added test coverage for services module; Added test coverage for session service components; Added test coverage for skill service, hub, and CLI login flows; Added test coverage for the Deep Research pipeline; Added test coverage for the cron service and repository; Added test coverage for utility modules; Added test suite for RAG pipeline services; Added test suite for multi-user account management and access controls; Added tests for Codex authentication and model catalog services; Added tests for LLM model selection and reasoning effort handling; Added tests for LLM probe token budget configuration; Added tests for NotebookSummarizeAgent extra\_headers handling; Added tests for PersonaService CRUD, migration, and preset seeding; Added tests for RAG retrieval evaluation components; Added tests for agentic loop reasoning replay, tool argument validation, and streaming tool-call handling; Added tests for config service behavior; Added tests for data volume permission checks; Added tests for embedding service client and adapters; Added tests for header-based chapter reconstruction; Added tests for knowledge base manager and document handling; Added tests for language directive and prompt assembly logic; Added tests for markdown table parsing edge cases in partner channels; Added tests for math-animator agent reliability and error handling; Added tests for research service security boundaries and scientific primitives; Added tests for search provider configuration and web search runtime behavior; Added tests for the MarginNote 4 capability integration; Added tests for the document parsing service and its engines; Added tests for the immersive reading domain; Added tests for the logging subsystem; Added tests for video learning capabilities and Invidious account integration; Added tests for visualize agent robustness and error diagnosis; Added unit tests for the Question agent pipeline and language handling; Expanded end-to-end test coverage for reading, UI, and multi-worker scenarios; Expanded test coverage for core agentic and runtime components; Expanded test coverage for tool registry, session isolation, and release workflows; Expanded test coverage for web application components and logic; New API test suite for authentication, book management, and co-writer features.

Dependencies

DeepTutor v1.6.5: New CLI distribution, Python 3.14 support, and major frontend upgrades

This release introduces a new \deeptutor-cli\ package for a lightweight, CLI-only installation and updates the main \deeptutor\ package to support Python 3.14 (with specific version caps for compiled dependencies like FAISS and PyStemmer that lack 3.14 wheels). The Python dependency stack now includes \json-repair\, \oauth-cli-kit\, and \pageindex\ as core dependencies, while adding optional extras for \codebuddy\ (Agent SDK), \video-learning\, and \partners\ (IM SDKs). On the frontend, the web application has been upgraded to Next.js 16 and React 19, with corresponding updates to \framer-motion\, \lucide-react\, \react-markdown\, and \tailwind-merge\, alongside a refreshed \package-lock.json\.

(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 44 → 44 (+0.6)
  • Rubric changed (rubric-2026.11.1 → rubric-2026.11.4) — scores are not directly comparable.

Lenses

  • Code Health 57 → 55 (-1.3)
  • Architecture 57 → 57 (+0.2)
  • Maturity 74 → 74 (+0.1)
  • Readiness 34 → 34 (+0.0)
  • Security 71 → 71 (-0.1)
  • Accessibility 45 → 48 (+3.1)
  • Performance 60 → 60 (+0.0)

Resolved (127)

  • Change coupling: protocol.py ↔ session-api.ts (deeptutor/services/session/protocol.py)
  • ChatMessageList.PlayAudioButton (cognitive 30) (web/features/chat/messages/ChatMessageList.tsx)
  • ChatMessageList.PlayAudioButton (cyclomatic 31) (web/features/chat/messages/ChatMessageList.tsx)
  • Critical vulnerability: [GHSA redacted] (web/package-lock.json)
  • DiscordHistoryExporter._channel_record (cognitive 18) (scripts/export_discord_history.py)
  • DiscordHistoryExporter._channel_record (cyclomatic 20) (scripts/export_discord_history.py)
  • DiscordHistoryExporter._list_joined_private_archived_threads (cognitive 17) (scripts/export_discord_history.py)
  • DiscordHistoryExporter._list_public_archived_threads (cognitive 20) (scripts/export_discord_history.py)
  • DiscordHistoryExporter._request_json (cognitive 28) (scripts/export_discord_history.py)
  • DiscordHistoryExporter.discover_threads (cognitive 24) (scripts/export_discord_history.py)
  • DiscordHistoryExporter.export (cognitive 19) (scripts/export_discord_history.py)
  • DiscordHistoryExporter.export (cyclomatic 17) (scripts/export_discord_history.py)
  • DiscordHistoryExporter.fetch_messages (cognitive 24) (scripts/export_discord_history.py)
  • Duplicated block (10 lines × 2) (deeptutor/capabilities/marginnote4/store.py)
  • Duplicated block (10 lines × 2) (deeptutor/co_writer/storage.py)
  • Duplicated block (101 lines × 2) (deeptutor/capabilities/mastery/tools.py)
  • Duplicated block (11 lines × 2) (deeptutor/services/config/runtime_settings.py)
  • Duplicated block (11 lines × 2) (deeptutor/services/imagegen/adapters/dashscope.py)
  • Duplicated block (11 lines × 2) (deeptutor/services/mastery_hints.py)
  • Duplicated block (12 lines × 2) (deeptutor/services/config/provider_runtime.py)
  • …and 107 more

New (462)

  • AnthropicProvider._assistant_blocks (cyclomatic 17) (deeptutor/services/llm/provider_core/anthropic_provider.py)
  • AnthropicProvider._parse_response (cognitive 16) (deeptutor/services/llm/provider_core/anthropic_provider.py)
  • AnthropicProvider._parse_response (cyclomatic 17) (deeptutor/services/llm/provider_core/anthropic_provider.py)
  • BookEditReview.BookEditReview (cognitive 16) (web/app/(workspace)/learning/books/components/BookEditReview.tsx)
  • BookEditReview.BookEditReview (cyclomatic 17) (web/app/(workspace)/learning/books/components/BookEditReview.tsx)
  • BookEngine._compile_page_now (cyclomatic 17) (deeptutor/book/engine.py)
  • BookEngine.confirm_spine (cognitive 22) (deeptutor/book/engine.py)
  • BookEngine.confirm_spine (cyclomatic 17) (deeptutor/book/engine.py)
  • BookEngine.delete_chapter (cognitive 22) (deeptutor/book/engine.py)
  • BookEngine.delete_chapter (cyclomatic 16) (deeptutor/book/engine.py)
  • BookEngine.rename_chapter (cognitive 16) (deeptutor/book/engine.py)
  • BookStructureEditor.BookStructureEditor (cognitive 31) (web/app/(workspace)/learning/books/components/BookStructureEditor.tsx)
  • BookStructureEditor.BookStructureEditor (cyclomatic 30) (web/app/(workspace)/learning/books/components/BookStructureEditor.tsx)
  • Change coupling: ChatMessageList.tsx ↔ failed-submissions.ts (web/features/chat/messages/ChatMessageList.tsx)
  • Change coupling: ChatWorkspace.tsx ↔ failed-submissions.ts (web/features/chat/components/ChatWorkspace.tsx)
  • Change coupling: init.ts ↔ datetime.ts (web/i18n/init.ts)
  • Change coupling: interface_settings.py ↔ datetime.ts (deeptutor/services/settings/interface_settings.py)
  • Change-coupling hub: launch_settings.py → init.ts, languages.ts, datetime.ts (deeptutor/services/config/launch_settings.py)
  • ClassTooLong: PdfDocumentView (web/components/reading/PdfDocumentView.tsx)
  • CodeBuddyAuthService.start_login (cognitive 20) (deeptutor/services/codebuddy_auth.py)
  • …and 442 more

Changes since last survey

  • 300 commits — 71 feature/other, 229 fixes

By area

  • (repo) — 112 commits
  • deeptutor/services — 52 commits
  • deeptutor/api — 16 commits
  • web/locales — 16 commits
  • tests/services — 11 commits
  • deeptutor/capabilities — 10 commits
  • deeptutor/runtime — 9 commits
  • deeptutor/learning — 8 commits
  • deeptutor/agents — 7 commits
  • apps/desktop — 5 commits
  • deeptutor/utils — 5 commits
  • web/features — 5 commits
  • web/app — 4 commits
  • web/tests — 4 commits
  • web/components — 3 commits
  • web/lib — 3 commits
  • web/public — 3 commits
  • (root) — 2 commits
  • deeptutor/multi_user — 2 commits
  • deeptutor/reading — 2 commits

Notable commits

  • fix: Merge PR #2489: fix: preserve book chat scroll position while streaming
  • fix: Merge PR #2490: fix: render and resume book page chat questions
  • fix: Merge PR #2491: fix(llm): recognize published GPT-6 context capacities
  • fix: Merge PR #2492: fix(mcp): refresh persisted OAuth grants after restart
  • fix: Merge PR #2493: fix(runtime): separate indexless streamed tool calls
  • fix: Merge PR #2502: fix: refuse mastery visual cards whose pixels were not inspected
  • fix: Merge PR #2503: fix(llm): keep pooled clients alive for active requests
  • fix: Merge PR #2504: fix(runtime): recognize framework Python orphan backends on macOS
  • fix: Merge PR #2505: fix(parsing): recheck groups that disappear during cleanup inspection
  • fix: Merge PR #2506: fix(runtime): pin backend imports to the launcher installation
  • fix: Merge PR #2509: fix(runtime): replay submissions during concurrent admission
  • fix: Merge PR #2510: fix(mastery): stop tutoring-memory retries from growing every turn
  • fix: Merge PR #2512: fix(embedding): restore indexed responses to input order
  • fix: Merge PR #2513: fix: pick the JSON fence when a code snippet fence comes first
  • fix: Merge PR #2517: fix(question): retry tool summarizer without stream_options when provider rejects it
  • fix: Merge PR #2519: fix(parsing): report a Docling option-setup fallback instead of swallowing it
  • fix: Merge PR #2520: fix: stop the dev extra from downgrading packaging
  • fix: Merge PR #2526: fix: resolve workspace KB scope past dead registrations; keep the progress stream off the activity lease
  • fix: Merge PR #2527: fix: keep first-chat tokenizer acquisition local and nonwaiting
  • fix: Merge PR #2530: fix(agents): a math animator visual review with no verdict says so (#1202)
  • …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

HKUDS/DeepTutor 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 15 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 4c1e000d8563c5ac3b231baaaf149dc4dc4d2406 — 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-2026.11.73+preprod-52de90b74216.