Agent Memory is the shared memory brain for Agent OS and remote nodes. It stores, retrieves, and serves knowledge across machines using a local-first architecture.
This repo turns Obsidian vault content and remote source summaries into a searchable memory layer with hybrid retrieval.
- Loads content from an Obsidian vault.
- Builds local embeddings with Ollama or a compatible embedder.
- Stores memory in LanceDB for hybrid vector + keyword search.
- Exposes memory via HTTP and MCP APIs.
- Supports remote memory ingestion from other machines.
- Preserves provenance, source metadata, and health reporting.
- Durable memory storage backed by Obsidian notes.
- Hybrid retrieval combining semantic vectors and keyword search.
- Remote node registration and health reporting.
- Memory search API for agents and dashboards.
- Provenance metadata attached to all memory entries.
cd "Agent Memory"
cp .env.example .env
npm installUpdate .env with your vault path and any optional values such as MEMORY_API_KEY or MEMORY_HOST.
npm run index
npm run serveFor MCP integration:
npm run serve:mcpdocs/TAILSCALE.md- secure mesh setup for remote nodes.docs/INTEGRATION.md- how agents connect to memory via HTTP and MCP.
Use this memory service as the shared brain for Agent OS and remote nodes. Configure the MEMORY_API_KEY in consuming agents and connect via the provided API.