JuliusBrussee/cavemem
Persistent memory layer for AI coding assistants that compresses session observations into a local SQLite store with hybrid search.

Cross-agent persistent memory for coding assistants. Hooks capture what happens at session boundaries, compress observations using a deterministic caveman grammar that reduces prose tokens by ~75% while preserving code and paths byte-for-byte, and write to a local SQLite database. Agents query their own history through three MCP tools: search, timeline, and get_observations. Supports Claude Code, Cursor, Gemini CLI, OpenCode, and Codex. Provides both SQLite FTS5 keyword search and a local vector index for hybrid retrieval.