agent-memory
View on GitHubLong-term memory runtime for AI agents — plain Markdown as the source of truth, local ranked retrieval, and an independent sleep-time Manage layer. Claude Code and Codex share one store. No API key.
Local-first long-term memory runtime for AI agents: Markdown files are the source of truth, a rebuildable SQLite/FTS5 index provides ranked path-based recall, and a sleep-time Manage layer consolidates memories. Works via CLI or MCP for Claude Code, Codex and any shell-capable host, with no API key.
Use Cases
Give coding agents persistent memory across sessionsShare one memory store between Claude Code and Codex CLIRetrieve ranked memories as file paths instead of pasted chunksConsolidate and forget memories during idle sleep passesReplace opaque vector stores with greppable Markdown filesTrack superseded decisions with as-of date recallExpose memory to any agent via CLI or MCPRebuild indexes without losing knowledgeBenchmark agent memory against LongMemEval
Built With
- Language
- Python
- Frameworks
- MCP · SQLite FTS5 · Claude Code · Codex CLI · uv · pytest · ruff
Tags
agent-memory · long-term-memory · markdown · local-first · sqlite · fts5 · bm25 · retrieval · mcp · claude-code · codex · no-api-key · consolidation · session-persistence · python · git-friendly