ai-memory stores agent memory as a git-backed markdown wiki with zero LLM calls in the write path
akitaonrails/ai-memory, a single Rust binary running MCP and HTTP servers over a local data directory, hit 7,975 stars (about 217 added on 22 September, last push that morning) on the argument that agent memory belongs to the repository rather than the vendor. Agents write observations through lifecycle hooks, which consolidate into ordinary markdown you can grep, edit in Obsidian and diff as commits, with retrieval by full-text search, entity matching and optional embeddings — no summarization model in the write path and a claimed write ceiling near 700/second. It registers with 20+ harnesses including Claude Code, Codex, Cursor, Gemini CLI and OpenCode, and version 1.39 handles multiple agents on one project. Note this is a May 2026 project trending again, not a new launch.
↳ Follow the thread