# honeycomb

honeycomb is shared, self-improving memory for AI coding agents. It captures every exchange across Claude Code, Cursor, Codex, Hermes, pi, and OpenClaw into a single loopback daemon, distils facts into a searchable Deeplake memory store, and injects the most relevant context before each new prompt. Memory is scoped personal, team, or org. The Dreaming consolidation loop merges duplicates, prunes stale anchors, and sharpens recall over time.

honeycomb is a collaboration between Legion Code and Activeloop. Licensed AGPL-3.0. Free to use.

## Key pages

- [Home](https://www.theapiary.sh/), Product overview and quick install.
- [Technology](https://www.theapiary.sh/technology), How the daemon, Deeplake store, and Dreaming loop work.
- [Assistants](https://www.theapiary.sh/assistants), The six supported harnesses and their install commands.
- [Teams](https://www.theapiary.sh/teams), Team and org memory scoping.
- [Install](https://www.theapiary.sh/install), One-command install for macOS, Linux, and Windows.
- [Docs](https://www.theapiary.sh/docs), Full documentation.
- [Blog](https://www.theapiary.sh/blog), Notes on agent memory and architecture.

## Supported coding assistants

- [Claude Code](https://www.theapiary.sh/assistants/claude-code), Captures every turn inside Claude Code via the hooks adapter, routes context to the daemon, and recalls before each new prompt.
- [Cursor](https://www.theapiary.sh/assistants/cursor), Wires into Cursor via the MCP server entry, capturing every exchange and injecting recall context before the model sees your prompt.
- [Codex](https://www.theapiary.sh/assistants/codex), Integrates with the Codex CLI via a post-run hook, persisting every decision and snippet so nothing your agent learned disappears when the run ends.
- [Hermes](https://www.theapiary.sh/assistants/hermes), Registers the Honeycomb MCP server inside Hermes so memory flows automatically alongside every agent invocation.
- [pi](https://www.theapiary.sh/assistants/pi), Captures pi sessions via the lightweight loopback adapter, keeping the same shared memory available across every tool you use.
- [OpenClaw](https://www.theapiary.sh/assistants/openclaw), Plugs into OpenClaw via the contracted-tools adapter, giving the daemon full visibility into every tool call and its outcome.

## Blog posts

- [Why agent memory needs receipts](https://www.theapiary.sh/blog/why-agent-memory-needs-receipts) (2026-06-18), An AI agent that cannot show you where it learned something is not a tool you can trust in production. We built Honeycomb around one rule: every recalled fact carries a source, a score, and a timestamp, or it does not ship.
- [The pollinating loop, explained](https://www.theapiary.sh/blog/the-pollinating-loop-explained) (2026-06-11), Dreaming is the name for what honeycomb does while you sleep: it goes back through every captured session, merges duplicate facts, prunes contradictions, and wakes up with a cleaner, denser memory store. Here is the algorithm.
- [One memory, six harnesses](https://www.theapiary.sh/blog/one-memory-six-harnesses) (2026-06-04), Claude Code learns something at 9 a.m. Cursor needs it at 3 p.m. Before honeycomb, those were two separate contexts in two separate windows. Here is how the single-daemon architecture makes cross-harness recall work, and why loopback-only matters.

## Further information

- Sitemap: https://www.theapiary.sh/sitemap.xml
- RSS feed: https://www.theapiary.sh/blog/rss.xml
- License: AGPL-3.0
- Source: https://github.com/legioncodeinc/honeycomb