# Memory Sidecar

An installable, agent-agnostic external memory system for Claude Code, Codex, Cursor, Hermes and similar agents, archiving sessions and building long-term recall without modifying the agent itself.

Memory Sidecar is an external memory system that runs alongside an AI agent, reading the agent's own data directory to archive sessions, build long-term knowledge, and inject relevant recall back into future work. It layers recall across hot, warm, cold and curated knowledge notes, using an agent-agnostic install driven by an AGENT_HOME path so it works across different agent tools.

The current release adds privacy-safe synthetic and private memory evaluation registries, quality metrics, governance policy metadata and dry-run-first edge planning for its gbrain component. It is built to run on top of a Hindsight and gbrain PostgreSQL environment, and emits machine-readable health reports so operators can tell when archival or recall silently stops working.

Website: https://github.com/mage0535/hermes-memory-installer