hipocampus
A drop-in proactive memory harness for AI agents that builds a compaction tree over past sessions and combines it with keyword or vector search, so an agent can surface relevant past context that nobody explicitly asked about.
hipocampus is a zero-infrastructure memory system, built from local files, that works immediately with Claude Code, OpenCode and OpenClaw after one setup command. It targets a specific gap in agent memory: search only finds context the user explicitly asks about, while its compaction tree structure can surface relevant past context proactively, evaluated on a benchmark of implicit context questions spanning three months of conversation history.
Combined with vector search and a larger root summary budget, it substantially outperforms plain BM25 or vector search alone on that benchmark, particularly on hard cross-domain questions where keyword and vector search find nothing. It installs as a Claude Code plugin or as a standalone npm package, with options to run compaction-tree only, disable vector search to save disk space, or target a specific agent platform.