Cavemem gives coding agents memory
Cavemem is an open-source, local-first memory layer for coding assistants that stores compressed session observations in SQLite, indexes them with FTS5 and vector search, and exposes recall through MCP for Claude Code, Cursor, Codex, Gemini CLI, and OpenCode.
Cavemem is attacking the right pain point: coding agents are useful until every new session starts with amnesia. The bet is that memory needs to be private, inspectable, and cheap before developers will trust it.
- –Local SQLite storage and no default network calls make this more plausible for real codebases than cloud-first memory products
- –MCP support turns memory into shared agent infrastructure instead of a feature trapped inside one IDE
- –Caveman compression is the differentiator, but also the risk: compressed memories need excellent fidelity or agents will retrieve confident nonsense
- –The project is early, but it fits the broader shift from bigger context windows toward curated, searchable, persistent context
DISCOVERED
45d ago
2026-04-23
PUBLISHED
45d ago
2026-04-23
RELEVANCE
AUTHOR
Github Awesome
