What is agentOrc?
agentOrc is a lightweight TypeScript SDK that gives multiple AI agents a shared, persistent semantic memory layer. Memory is stored locally with SQLite and searched with sqlite-vec — no servers, no hosted vector database, and no framework lock-in.
You initialize once, then call a small API: remember(), recall(), compress(), forget(), and a few helpers.
What it is not
- Not an AI agent framework
- Not an orchestration runtime
- Not a hosted backend service
- Not a general-purpose vector database product
It is simply the shared semantic memory layer your agents can all talk to.
Why agentOrc?
- Local-first — data stays on disk in a SQLite file
- Multi-agent — concurrent writers are handled for you
- Model-agnostic — any OpenAI-compatible embedding and chat endpoint
- Small API — easy to integrate into existing agents
Core ideas
Memories are embedded from content.text, stored with opaque metadata, and retrieved by semantic similarity. Compression can summarize related memories through a separately configured LLM.