agent-memory

Give your AI agent a memory that persists across sessions.

Agent Memory stores knowledge as Markdown files you can read, edit, and version — just like Obsidian, but built for agents. It finds relevant memories through hybrid search that combines full-text search (FTS5) with vector similarity (local embeddings, no API calls) and connects them in a knowledge graph with bidirectional links.

Everything runs locally. No cloud, no API keys, no external services.

Key Features

• Markdown-first — Human-readable files with YAML frontmatter. Your data, your format.
• Hybrid search — BM25 keyword matching + cosine vector similarity, merged via Reciprocal Rank Fusion.
• Local embeddings — Runs a MiniLM model on your machine via ONNX. Zero network calls.
• Knowledge graph — Bidirectional connections between entries with automatic discovery.
• Git versioning — Every change committed with semantic messages via pure-JS git.
• Per-project + global — Project-scoped memories with a shared global knowledge base.

Do not run the full test suite — see Commands. Tests run per module.

Architecture

Five isolated modules plus an orchestrator, an extension runtime, and migrations. Modules never import from each other — the orchestrator in src/index.ts is the sole wiring point.

Modules

Memory Store

Each memory is a Markdown file with YAML frontmatter, organized by type:

A memory file looks like this: