Honeycomb plugin
Shared memory for Claude Code, Codex and Cursor: a decision made in one session is recalled in another tool or on another machine.
- 118 ★ on the repository
- Last commit
Extended access
Why this level
- The daemon records session content and stores it in Deeplake, so code and secrets from conversations may end up in memory.
- The one-line installer runs a script fetched from the network.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/honeycombDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The daemon records session content and stores it in Deeplake, so code and secrets from conversations may end up in memory.
- The one-line installer runs a script fetched from the network.
This is third-party code. Review the repository files before installing.
What it does
Honeycomb runs a local daemon that records each coding-agent session, distills it into short memories and serves them back to any connected tool. Every memory exists at three levels: a one-line key, a short summary and the full session. At session start the agent receives a small index of the most relevant keys and pulls detail on demand. The daemon also mines reusable skills from past sessions and builds a knowledge graph and a code graph. Storage runs on Activeloop Deeplake, and a dashboard opens in the browser at a local address.
Who it is for. Developers and teams who run one project across several agents and re-explain context to each fresh session.
Good fit when
- A decision made in Claude Code in the evening is needed in Cursor or Codex the next morning.
- A new teammate wants to know why a module is built the way it is.
- Recurring techniques from sessions should become a shared skill for everyone.
- An agent should recall how the same error was fixed last time.
Not a fit when
- Session data cannot go to external storage, even encrypted.
- The agent works on a throwaway script and cross-session memory adds nothing.
- The project cannot tolerate a daemon holding a local port.
Example request
Recall how we decided to handle payment errors last week and show the session where it was discussed.Limitations
Storage needs Activeloop Deeplake credentials: sign-in uses the device flow, or you can host your own `pg_deeplake`. The license is AGPL-3.0-or-later, so changes ship under the same open-source license. Claude Code, Codex and Cursor are supported in production; Hermes, pi and OpenClaw are still in progress. The daemon listens on the local port `127.0.0.1:3850`, and the one-line installer sets up Node and the package globally, so read the script before running it. The last commit in the repository dates from October 2026.
How to disable. Run `honeycomb uninstall`: it removes agent hooks, the daemon service and the local `~/.apiary/honeycomb` directory. Clear shared credentials and Deeplake memory separately.
FAQ
What does the agent remember across sessions?
Decisions, project conventions and fixes. Memory is stored in three layers: key, summary and full session, so the agent reads keys first and opens detail only when needed.
How do I forget something?
Claude Code gets `/remember`, `/recall` and `/forget` commands. The last one removes the chosen memory; the daemon's periodic pass also merges duplicates and drops stale facts.
Can it run without the cloud?
Partly. You can host storage yourself on the `pg_deeplake` Postgres extension and point Honeycomb at it with `honeycomb login --endpoint`.
How is this different from a CLAUDE.md file?
A file is edited by hand and lives in one project. Honeycomb collects memory automatically, searches it and serves it to any connected agent.
How do I check that the daemon is running?
Run `honeycomb status`: it shows the state of the daemon and the environment. The dashboard opens at the local address `127.0.0.1:3853`.
Discussion
No questions or reviews yet.
Related
PluginsShared cloud memory for a team's agents: stores session traces and turns repeated patterns into skills
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
hipocampus gives an agent file-based memory: a 3,000-token topic index loads each session and shows what came up before
Knowledge-graph memory for agents: Claude Code and Codex plugins, an MCP server and a Python library
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
One local memory for Codex, Claude Code, Cursor, Hermes Agent and OpenClaw: agents hand off tasks, review each other and refine their skills overnight
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Honeycomb (this card) | no | 118 | |
| Claude-Mem | no | 113,534 | keeps observations for a single agent; Honeycomb maintains shared memory across several tools with a knowledge graph |
| Hivemind | no | 1,646 | predecessor with shared cloud memory; Honeycomb adds a local daemon, a dashboard and three-level memory |
README in short
Memory is captured automatically on every turn and available in all supported tools. Skills found by one engineer are pulled into the others' sessions at start.