hivemind

Why we included this project

Teams running several coding assistants often watch each one repeat the same expensive lessons, like the tricky migration steps or the shell quirks that only make sense after you've hit them once. Hivemind saves those sessions, capturing prompts, tool calls, and responses in a shared store, then distills recurring patterns into SKILL.md files that connected agents pick up automatically. Plugins and hooks attach it to Claude Code, Codex, Cursor, and OpenClaw, so what one engineer's agent figures out on Monday is already in the next person's session on Tuesday. Retrieval mixes lexical and semantic matching to pull up the right prior work without loading the whole history into context, and the maintainers cite LoCoMo results showing 25% cheaper runs, 1.7× fewer tokens, and 31% fewer turns than going without shared memory. It is a good fit for small teams that want institutional agent memory without running their own infrastructure, though the cloud-backed design and Deeplake storage are worth checking if session data placement matters to you.