All projects/Products/2025 –/In development
Enkd OS: an AI operating system for a person and a company
Event-driven agents over a knowledge graph, with a Telegram hub and a web dashboard as the interfaces. Local-first, with a human approval gate on every action that touches money or a calendar.
What I built
- —Event-driven core on Redis Pub/Sub and Streams: capabilities publish and subscribe, they never call each other directly.
- —Every tool is its own MCP server, one per domain, so adding a capability means adding a server rather than widening one.
- —Knowledge graph on Memgraph, with Cognee for entity extraction and Graphify for codebase mapping, as shared memory.
- —LiteLLM in front of every model call with fallback across providers and local models; a Telegram Approve / Edit / Reject gate on every financial or calendar write.
Stack
Python · FastMCP · FastAPI · Postgres + pgvector · Redis · Memgraph · Cognee · LiteLLM · Next.js · Docker
Outcomes
- —Architecture fixed and in build: event bus, MCP per domain, dual knowledge graph.
- —Designed to cut monthly assistant API spend by roughly an order of magnitude by moving to local models.
- —Targets an ARM64 build under 4 GB RAM on a Raspberry Pi 5.