How sup3rus3r/omnex works
Omnex is a self-hosted AI memory system: users ingest local files, which are classified, processed and enriched into searchable memory, then recall them through natural-language queries. Its API also exposes memory to agents and federates instances; the UI supports identity management, timeline, setup and voice features. The README documents the broader capabilities; sampled source confirms selected API, ingestion and UI wiring, so unsampled subsystem links are omitted where ownership is uncertain.
Main entry points: main.py FastAPI application · ingest.py Ingestion endpoints · query.py Query endpoints · query_engine.py Query engine
Components 25
User experience
- Omnex interface
interface/app/page.tsx - Ingestion controls
interface/components/IngestionPanel/IngestionPanel.tsx - Ingestion dashboard
interface/components/IngestionDashboard/IngestionDashboard.tsx - Recall query
interface/components/QueryBar/QueryBar.tsx - Result presentation
interface/components/ResultGrid/ResultGrid.tsx - Identity manager
interface/components/IdentityManager/IdentityManager.tsx
API and integrations
- FastAPI application
api/main.py - Ingestion endpoints
api/routes/ingest.py - Query endpoints
api/routes/query.py - Query engine
api/query_engine.py - Agent and MCP API
api/routes/mcp.py - Federation API
api/routes/federation.py - Timeline API
api/routes/timeline.py - Voice API
api/routes/tts.py
Memory pipeline
- File detection and routing
ingestion/router.py - Media processors
ingestion/processors - Multimodal embeddings
embeddings - Semantic tagging
ingestion/semantic_tagger.py
Memory state
- Metadata and sessions
storage/mongo.py - Vector index
storage/leann_store.py - Binary content store
storage/binary_store.py
Other components
- Human user
- AI agent
- Configured LLM
- Peer Omnex instance
Connections 28
- Human user → Omnex interface: uses
- Omnex interface → Ingestion controls: opens
- Omnex interface → Recall query: opens
- Ingestion controls → Ingestion endpoints: uploads or triggers
- Ingestion dashboard → Ingestion endpoints: starts and polls
- Ingestion endpoints → File detection and routing: routes files
- File detection and routing → Media processors: selects processor
- Media processors → Multimodal embeddings: embeds content
- Media processors → Semantic tagging: enriches chunks
- Ingestion endpoints → Metadata and sessions: tracks ingestion
- Ingestion endpoints → Vector index: deletes vectors
- Ingestion endpoints → Binary content store: deletes blobs
- Recall query → Query endpoints: submits query
- Query endpoints → Query engine: runs search
- Query endpoints → Metadata and sessions: loads and saves sessions
- Query engine → Vector index: searches index
- Query engine → Metadata and sessions: expands source
- Query endpoints → Result presentation: returns results
- Identity manager → FastAPI application: requests identity data
- FastAPI application → Query endpoints: mounts routes
- FastAPI application → Ingestion endpoints: mounts routes
- AI agent → Agent and MCP API: reads and writes memory
- FastAPI application → Agent and MCP API: mounts routes
- FastAPI application → Federation API: mounts routes
- Federation API → Peer Omnex instance: federates
- FastAPI application → Timeline API: mounts routes
- FastAPI application → Voice API: mounts routes
- Query engine → Configured LLM: generates answers