Spaces:
Running
Running
Commit History
-
feat: enhance knowledge graph visualization and add new features
Komalpreet KaurUnverified signature- Updated package.json and package-lock.json to include "three-spritetext" for improved text rendering in the 3D graph. - Modified KnowledgeGraph component to integrate SpriteText for always-visible node labels and relationship tags. - Introduced new scripts for benchmarking memory retrieval latency and debugging direct interactions with the knowledge extraction system. - Added tests for belief revision, context extraction, plural reconciliation, and semantic aliasing to ensure robust knowledge handling. - Implemented a script to strip co-author information from commit messages.55f6497
-
fix: bundle images locally, fix sleep-cycle user attribution, cleanup
Komalpreet Kaur- Bundle brain/sleep images as optimized WebP (1.8MB+2.6MB PNGs from raw.githubusercontent.com -> 152KB+137KB local assets); fixes slow page load caused by fetching from GitHub on every visit - Fix sleep cycle and hippocampus consolidation storing memories and graph triples under "default_user" instead of the session owner - Remove decorative Entropy Score tile from Memory Explorer (value was hardcoded) - README: document local Neo4j in docker-compose, Aura username quirk (instance ID, not "neo4j"), cloud deployment notes, keepalive workflow - Remove obsolete docker-compose version attribute; gitignore chroma_db_docker/6ea3601
-
feat: implement responsive, theme-adaptive visitor analytics modal dialog box
Komalpreet Kaurccfe044
-
feat: use robust Pydantic structured output for 100% stable graph relationships extraction
Komalpreet KaurUnverified signaturec7fa78c -
fix: lower neocortex input word guard to 3 to support short statements
Komalpreet KaurUnverified signature7fb3f6a -
fix: bulletproof node validation - block ALL sentences and filler text from graph
Komalpreet KaurUnverified signaturef7dd68a -
fix: extract knowledge only from user messages, not Soma's verbose responses
Komalpreet KaurUnverified signatureee2f0d9 -
refactor: child-brain knowledge extraction - simple concept nodes only
Komalpreet KaurUnverified signature82bb846 -
fix: extract only real-world knowledge entities, not chat metadata
Komalpreet KaurUnverified signature4507045 -
fix: remove example contamination from neocortex extraction prompt
Komalpreet KaurUnverified signature2c33130 -
feat: implement memory consolidation service with sleep cycle, add frontend visualization components, and integrate backend database orchestration.
Komalpreet KaurUnverified signatureadfa6d0
-
feat: Overhaul Soma UI to premium Neural Gloss aesthetic with single-screen optimization
Komalpreet KaurUnverified signature6e94481
-
Refactor: Simplify auth flow and improve neural mesh update synchronization
Komalpreet KaurUnverified signature6d0aeb9 -
fix: allow natural conversation when no memories exist yet
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureThe system prompt was telling the LLM to only answer from stored memories, and to say "neural pathways do not contain information" otherwise. Since we wipe all data on login (clean slate), the bot could never respond to anything. Updated prompt to: - Respond naturally using general knowledge when no memories exist - Use stored memories to enrich responses when available - Acknowledge learning as the user talks - Warm, intelligent persona that grows with conversation Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>36fd791 -
fix: handle offline environments for embeddings model
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureHF Spaces and some deployment environments can't access the internet at runtime. Changes: - Pre-download embedding model during Docker build (when we have internet) - Add graceful error handling if model can't be loaded at runtime - Continue app startup even if embeddings fail - Only log warnings, don't crash This allows the app to work in completely offline environments (sensory memory becomes optional feature). Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>768bcd5 -
fix: lazy-load embeddings to prevent startup hangs
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureThe embedding model was being downloaded at app startup, causing: - HF Spaces restart loops when HF Hub is unreachable/slow - Startup delays and potential timeouts Changes: - Move embeddings to lazy initialization (get_embeddings()) - Only download model on first use, not at import time - Add error handling for network failures - Continue without embeddings if download fails This prevents app startup failures due to external HF Hub issues. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>a367b18
-
feat: replace sign up/sign in with single 'Enter' flow
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureOne form, one button. Enter username + password: - New username: account auto-created - Existing username: logs in (verifies password) - Either way: memory is wiped for a clean slate Removed the login/register toggle from the frontend and merged the two backend endpoints into a single /auth/enter route. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>d6a10f8 -
fix: strict user isolation — remove all user_id IS NULL leaks
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureEvery Neo4j query had 'OR user_id IS NULL' which leaked unscoped entities to all users. Removed from: graph endpoint, graph stats, neocortex retrieval, and brain vitals. Also added memory reset on register (not just login) so even re-registering gets a clean slate. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>988181f -
feat: clean slate on every sign-in — wipe all prior user data
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureOn login, all user data is cleared across every memory layer: - SQLite/Postgres: messages and neural sparks deleted - ChromaDB: all sensory memory vectors deleted - Neo4j: all knowledge graph entities and relationships deleted Each session starts completely fresh. Users build their neural mesh from scratch through conversation within that session. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>c308372 -
feat: auto-build neural mesh from chat conversations
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureAfter each chat exchange, the user's message and Soma's response are now automatically ingested into ChromaDB (sensory memory) and extracted as knowledge triples into Neo4j (semantic memory). This runs in the background via asyncio so it doesn't slow down the streaming response. New users start with blank memory and their neural mesh grows organically as they chat — no manual consolidation required. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>c29d27b -
fix: always define DB_PATH so SQLite fallback works
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signatureDB_PATH was only set when USE_POSTGRES was False at import time. When DATABASE_URL is configured but unreachable, the runtime fallback to SQLite failed with NameError: DB_PATH not defined. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>f04fc03 -
fix: gracefully fall back to SQLite when Postgres is unreachable
Komalpreet Kaur +1: Claude Haiku 4.5Unverified signaturePostgres is now checked at startup with a 3-second timeout. If the connection fails, the app automatically falls back to SQLite instead of crashing. This allows Soma to start successfully even when Supabase or other Postgres instances are temporarily unavailable. - Added _test_postgres_connection() to check connectivity - Made _db_backend dynamic so it can be switched at runtime - Updated all SQL queries to use correct placeholders for each backend - Added fallback warning message to logs Fixes the startup error: "Network is unreachable" when DATABASE_URL is set but Postgres is not accessible. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>9607bf7
-
feat: add Supabase Postgres support with SQLite fallback
Komalpreet Kaur +1: Claude Sonnet 4.6Unverified signature- session.py auto-detects DATABASE_URL env var: uses Postgres (Supabase) if set, SQLite for local dev - Add psycopg2-binary to requirements - Add DATABASE_URL to config settings - Fix Dockerfile port back to 7860 for HF Spaces Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>1683d23 -
fix: normalize usernames and ensure case-insensitive authentication
Komalpreet KaurUnverified signature00e3427 -
refactor: optimize auth flow and improve API error handling
Komalpreet KaurUnverified signature65125eb
-
feat: Implement secure Authentication system with JWT and AuthScreen UI
Komalpreet KaurUnverified signatureded8028
-
feat: Finalize Onboarding UI and improve orchestrator memory grounding
Komalpreet KaurUnverified signatured4cde5c
-
fix: Use IS NULL for Neo4j legacy compatibility across all graph queries
Komalpreet KaurUnverified signature7633793 -
fix: Graph rendering with legacy data + Brain-organic color palette overhaul
Komalpreet KaurUnverified signature9b52b76 -
feat: Add generative visual dream sequence gallery
Komalpreet KaurUnverified signature48b770b -
style: implement neural aesthetic UI revamp
Komalpreet KaurUnverified signature863d9b5 -
feat: Add Multi-User Persona memory compartmentalization
Komalpreet KaurUnverified signature6551744
-
fix: restore missing retrieve function header in orchestrator
Komalpreet KaurUnverified signaturec560599 -
Evolution Complete: Soma is now an Autonomous Cognitive Architecture.
Komalpreet KaurUnverified signature78ccba5 -
Phase 3 Complete: Background dreaming and Neural Spark log implemented.
Komalpreet KaurUnverified signature0a09b4b -
feat: implement SQLite session management and background neural spark generation service
Komalpreet KaurUnverified signaturec334e4e -
Phase 2 Complete: Real-time Neural Mesh highlighting implemented.
Komalpreet KaurUnverified signature7e9698c -
Phase 1 Complete: Implemented Internal Reflection node and Monologue UI.
Komalpreet KaurUnverified signaturee95f953 -
Final Evolution: Secure Neural Build (Full Scrub & Sync)
Komalpreet KaurUnverified signaturea9df765
-
feat: Establish initial FastAPI project with core configuration, API endpoints, and a basic Langgraph orchestrator.
Komalpreet Kaurb0a98f0