lucascs@proton.me Lucas C.S. akitaonrails/ai-memory https://github.com/akitaonrails/ai-memory/issues https://github.com/akitaonrails/ai-memory/releases ai-memory is a long-term memory server for AI coding agents, spoken to over the Model Context Protocol (MCP). It keeps everything on the machine it runs on: pages live in a bundled SQLite database with FTS5 full-text search, a git-versioned wiki records how they changed, and embeddings for semantic search are computed in-process by a pure-Rust BERT model on the CPU, so no text has to leave the host to make search work. Optional providers (Anthropic, OpenAI, Gemini, GitHub Copilot and any OpenAI-compatible endpoint) can be configured for consolidation and hosted embeddings when that trade is wanted. It ships a single ai-memory binary that serves MCP over stdio or HTTP, exposes a small web UI, and installs per-vendor hook scripts under /usr/share/ai-memory/hooks for Claude Code, Codex, Cursor, Gemini CLI, opencode and others. The daemon can run system-wide under the dedicated ai-memory account or per user session; this package provides both an OpenRC service and, with USE=systemd, the upstream units for each scope. This package installs the prebuilt x86_64 or aarch64 executable published by upstream on GitHub Releases, together with the hooks, configuration templates and packaging files carried in the same tarball, so it installs exactly the same file set as the from-source dev-util/ai-memory. The two are mutually exclusive: both install /usr/bin/ai-memory. The binary is linked against glibc and will not work on a musl profile.