Repository navigation
Releases
v0.4.0: Email/Password Authentication
What's New
Authentication
- Email/Password Login & Signup β Create an account or log in with your email and password
- JWT Tokens β Secure, stateless authentication with 30-day token expiry
- Password Hashing β bcrypt hashing for secure password storage
Backend Changes
- New
POST /api/auth/signupendpoint β register with email and password - New
POST /api/auth/loginendpoint β returns JWT token - All data routes now require authentication via
Authorization: Bearer <token>header - Added
user_idcolumn to sessions and study_spaces tables - All queries filtered by authenticated user β data is now per-user
Frontend Changes
- New login/signup page at
/loginwith form toggle - Auth context manages JWT token in localStorage
- All pages redirect to
/loginif unauthenticated - All API calls include the auth token in headers
- Logout button on dashboard
Migration
- Existing data automatically assigned to the seeded user account
- Database schema updated with
userstable anduser_idforeign keys
Default Account
- Email: nasa5346@gmail.com
- Password: 53461234@vi)
Tech Stack
- Backend: bcrypt for password hashing, PyJWT for token generation
- Frontend: React context for auth state management
Full Changelog: v0.3.18...v0.4.0
v0.3.18
v0.3.18 β SSE Streaming, Retry Logic, Formula Split
Backend
- SSE streaming chat β new
POST /{session_id}/chat/streamendpoint withchat_completion_stream()async generator - Retry with backoff β 3 retries with exponential backoff (1sβ2sβ4s) on 5xx, timeout, connection errors
- httpx streaming fix β replaced
stream().__aenter__()withsend(req, stream=True)(root cause of 'Failed to get response') - Lazy API key β
_get_api_key()reads env lazily, fixing load_dotenv race condition - Tool call refactors β extracted
_safe_int_id(),_parse_note_tool_call(),_parse_canvas_tool_call()(C901 from 25 to below threshold) - Lint clean β ruff passes with new
ruff.tomlsuppressing intentional patterns
System Prompts
- Formula node split β formula nodes now contain ONLY the expression (e.g.,
F = ma), separate note nodes explain variable definitions and usage - Mandatory connections β every formula MUST connect to its explanation note
- Updated Newton's Laws and Photosynthesis examples to show the pattern
Frontend
- Letter-by-letter streaming β
sendMessageStream()SSE reader with character-by-character rendering - Non-streaming fallback β automatic fallback to
sendMessage()on stream errors checkOk()helper β every API error now includes HTTP status (e.g.,send message (500))fetchWithTimeout()β eliminates 3x AbortController boilerplatefallbackToSync()β extracted fromhandleSend, kills 28-line duplication- 16 silent catches now log to console.error across 7 files
- Dead code removed β 5 unused exports deleted (ChainOfThought*, badgeVariants, BaseNodeFooter, generateTTS, listDeletedSessions, restoreSession)
Commits
0ed88af chore: bump version to 0.3.18
9763952 feat: frontend streaming, error handling, and dead code removal
e22c5c2 feat: SSE streaming chat, retry logic, and formula node split
c4f99d8 refactor: backend lint fixes, lazy imports, and ponytail comments
v0.3.15 β 73% Smaller System Prompt, 2-12s Faster
What's New
System Prompt Compressed 73% (18KB β 4.8KB)
The system prompt sent to the LLM on every request was bloated with 5 full canvas template JSON examples and duplicate rules. Now it fits in a fraction of the space β resulting in significantly faster responses:
| Request | Before | After | Improvement |
|---|---|---|---|
| Simple greeting | ~7.8s | ~5.7s | 2s faster |
| Canvas creation | ~29s | ~17s | 12s faster |
| Q&A explanation | ~22s | ~14.7s | 7s faster |
'My thinking' Header
The Chain of Thought collapsible header now reads My thinking in italic instead of 'Chain of Thought'.
Better Error Diagnostics
The chat panel's catch block now logs so the true error reason appears in the browser console instead of being silently swallowed.
v0.3.12 β Chain of Thought
What's New
Chain of Thought Reasoning
Every AI response now includes structured reasoning steps displayed in a collapsible panel. The AI thinks through its approach before responding β checking study notes, selecting templates, planning content β and shows you exactly what it's considering.
Canvas Improvements
- Flowcharts are now mandatory for every canvas creation
- Canvas tool calls are emitted reliably (no more descriptions without actual canvas creation)
- Canvas creation follows a strict order: reasoning β short text β tool call β confirmation
Reliability Fixes
- Fixed frontend fetch timeout causing 'Failed to get response' errors on slow canvas/note creation requests (120s AbortController)
- Reasoning steps are persisted in the database and survive page reloads
- System prompt restructured so output format rules and tool call requirements are at the top
Technical
- Added AI Elements ChainOfThought component (@ai-elements/chain-of-thought)
- Added reasoning column to messages table (auto-migrated)
- Backend parser distinguishes reasoning blocks (JSON arrays) from tool calls (JSON objects)
v0.3.7 β Canvas: Visual Study Board
Canvas: Visual Study Board
AI now generates visual canvases β colored, connected concept cards β alongside notes and chat.
What's New
- Canvas Tool β Ask the AI to "create a canvas", "visualize this", or "draw a flowchart" and it generates a visual board with colored element cards
- 6 Element Types β Definitions (blue), Formulas (green), Flowcharts (orange), Notes (purple), Examples (pink), Headings (gray)
- Connected Elements β Flowchart arrows connect related concepts automatically
- Notes & Canvas Tabs β Side panel now switches between Notes and Canvas views
- Lazy Loading β Messages load 7 at a time with scroll-to-load-more; notes and canvases load content on demand for faster session startup
- Canvas CRUD β Create, rename, and delete canvases from the UI
Performance
- Session loading optimized: only last 7 messages fetched initially, older messages loaded on scroll
- Notes and canvases list endpoints return metadata only; full content fetched when a tab is selected
- Reduces initial session payload by ~60-80% for sessions with long chat history
Bug Fixes
- Fixed voice mode not reflecting canvas changes in the side panel
- Fixed scroll position jumping when loading older messages
Backend
- New
canvasestable with JSON elements storage - New
/api/canvasesCRUD endpoints - New
/api/sessions/{id}/messagespaginated endpoint - Canvas tool call parsing in chat and voice endpoints
- CORS updated for
localhost:3008
Dependencies
- Added
@xyflow/react@^12.11.1for React Flow canvas rendering
v0.2.13 β Study Spaces, Voice Improvements & Performance
What's New
Study Spaces
- Create folders to organize your sessions
- Custom 3D folder component with pop-out files showing session names
- Right-click/long-press to rename or delete spaces
- Sessions in spaces are excluded from the dashboard root
Voice Mode Improvements
- Rewritten system prompts using OpenAI voice agent best practices
- AI identifies as Void X1 by VoidLabs
- Better teaching structure: Preamble β Direct Answer β Breakdown β Example β Check-in
- Notes only created when explicitly asked (not automatically)
- Note content formatted with markdown in both voice and chat modes
- Hold Spacebar to speak on desktop, Hold to speak on mobile
Performance Fixes
- Fixed httpx connection leak (reuse single client)
- Chat history limited to last 20 messages
- Cursor trail throttled to ~20fps
- Image compression before storage (4MB to ~100KB)
- Images sent as multimodal content to LLM (no extraction step)
- Session auto-title generation after 5 messages
Dashboard Changes
- Instant session creation (no upload required)
- Upload images anytime during chat
- Removed emoji cursor trail animation
- Bottom blur fades out at scroll bottom
Bug Fixes
- Fixed SQLite subquery error in voice endpoint
- Fixed TTS URL too long (play cached audio directly)
- Fixed image thumbnails cropping (object-contain)
- Fixed voice mode not playing audio
Full Changelog: v0.1.15...v0.2.13
v0.2.0
What's New
Session Management
- Rename sessions: Click the 3-dot menu on any session card and select Rename to edit inline
- Delete sessions: Click the 3-dot menu and select Delete to remove a session (with confirmation)
- Long-press support: On touch devices, long-press a session card to open the actions menu
- New backend endpoints:
PATCH /api/sessions/:id(rename) andDELETE /api/sessions/:id(delete)
Cursor Icon Trail
- Session cards on the dashboard now have a cursor icon trail effect
- Dark lucide icons (Book, Brain, Lightbulb, GraduationCap, Flask, Sparkles, Atom, Palette) follow the cursor with fade, scale, and rotation animations
- Works on desktop hover; trail items cycle through 8 unique icons
Voice Mode
- Subtitles now auto-scroll as words appear during playback
Bug Fixes
- Fixed cursor trail rendering at wrong position by removing overflow-hidden clipping
- Stabilized trail item references to prevent constant re-rendering
v0.1.15
Release Notes - v0.1.15
This release introduces real-time voice mode subtitles, dynamic word-by-word fade-in animations, text formatting cleanup, server-side caching, and persistent question/answer display.
What's New
π€ Real-Time Subtitles & Dynamic Word-by-Word Timings
-
Dynamic Timing Calculation: Subtitle words are dynamically highlighted in perfect sync with the audio. Rather than relying on hardcoded speed averages or failing on streaming
Infinitydurations, timings are calculated on the backend using the exact audio sample counts from the Kokoro pipeline:
$$\text{Duration} = \frac{\text{sample count}}{24000}$$
The duration is linearly distributed across the words in the corresponding text segment to produce exact start/end timestamps. -
Micro-Animations: Renders words as individual spans, transitioning spoken words to
opacity-100 translate-y-0 scale-100 font-mediumand keeping future words hidden atopacity-0 translate-y-2 scale-95. This creates a smooth slide-up and fade-in typing effect as the assistant speaks. - Transparent Layout: Subtitles are displayed in a borderless, shadowless, and fully transparent container for a cleaner floating aesthetic.
π§ Dedicated Voice Chat & Prompt Formatting
- Strict Formatting Prompts: Created a dedicated
POST /voice-chatendpoint that uses a customized, voice-mode system prompt instructing the AI not to generate markdown bolding (*), hashes, dashes, separators, or emojis. - Text Cleaning Pipeline: Added a Python regex cleaner on the backend to filter out any stray formatting symbols or emojis, ensuring completely clean spoken output.
- Persistent Text Rendering: Spoken prompts (questions) and AI answers remain visible on the screen after the AI finishes speaking (when the avatar returns to
idle), instead of disappearing instantly. - New Session Reset: Old questions and answers are cleared strictly when the user holds the halo to start a new speaking query.
β‘ Server-Side Audio Caching
- Cached WAV Playback: The backend caches the generated WAV audio file locally during the
/voice-chatcall. When the browser makes the subsequent GET/ttsrequest, the server immediately streams the cached file, resulting in zero generation latency at the moment of playback. - Git Ignore: Ignored the
backend/cache/directory in.gitignoreto prevent caching artifacts from cluttering commits.
v0.1.13
Release Notes - v0.1.13
This release introduces significant enhancements to the session creation flow, background image processing in chats, low-latency real-time voice streaming, and general stability improvements.
What's New
π Features & Enhancements
- Bypass Initial Material Upload: Added the option to skip/bypass uploading study materials initially and start with an empty session workspace. A premium empty-state placeholder in the notes sidebar guides users on how to automatically generate notes by uploading files inside the chat later.
- Background Image Processing & Appending: Image uploading in the chat panel now runs asynchronously in the background. The notes sidebar automatically displays a loading indicator during extraction and note synthesis, merging new information with existing notes once complete. This prevents proxy timeouts (
ECONNRESET/ socket hangup) and guarantees a smooth, uninterrupted user experience. - Low-Latency Voice Mode: Replaced the previous audio chunks system with a real-time voice streaming engine using a single
GET /ttsendpoint, greatly improving audio start times and keeping the AI voice agent highly responsive. - Better Voice-Mode Animations: Keeps the AI avatar's thinking state animation active while the voice stream buffers, transitioning smoothly to speaking animation once the audio is ready.
- Enhanced Chat & Voice Prompting: Relaxed constraints to allow the AI to handle casual chats, off-topic questions, and general discussions instead of strictly restricting its knowledge to generated notes.
- Session Progress & Control: Re-architected session creation to use FastAPI background tasks. Added error handling states and user prompts to confirm stopping if a user accidentally dismisses the progress dialog.
π Bug Fixes
- Removed file upload size caps and fixed handling for multi-file study material uploads.
- Moved proxy logic from edge middleware to server-level
next.configrewrites to prevent socket hangups on long-running proxy calls. - Resolved boundary parsing errors on empty multipart uploads by appending dummy form fields.
v0.1.5 β Voice Output Fix
π Voice Output Fix
What's Fixed
- Voice mode now produces audio output β the core issue was that
audio.play()was silently failing due to browser autoplay policy. After the user holds-to-speak and the TTS response comes back from the server, the browser no longer considers the original press a "user gesture", blocking audio playback with no error.
What Changed
Frontend β Voice Mode (voice-mode.tsx)
- AudioContext unlock on gesture β a silent audio buffer is played the moment the user presses down, pre-unlocking the browser's audio pipeline for subsequent playback
- Awaited
audio.play()β promise rejections are now properly caught instead of being silently swallowed - Web Audio API fallback β if
HTMLAudioElement.play()still fails, audio is decoded and played through the already-unlockedAudioContext - Proper cleanup β
AudioContextis closed on component unmount
Backend
- Minor configuration updates to CORS and TTS service
Infrastructure
- Next.js 16 proxy routing configured for API pass-through to FastAPI backend
- Upload dialog and chat panel UX improvements
- TypeScript config updates
Full Changelog: 3d3945b...v0.1.5
