Repolode
Submit
activeloopai

activeloopai/hivemind

Shared, auto-learning memory layer that captures coding-agent sessions, mines them into reusable skills, and shares them across Claude Code, Codex, Cursor and other agents.

View on GitHubWebsite
Stars
1,632
Forks
112
Open issues
125
Last commit
12 hours ago

What is it?

Hivemind gives a team of coding agents a shared, cloud-backed memory. It installs hooks into supported assistants such as Claude Code, OpenClaw, Codex, Cursor, Hermes Agent and pi, records each session's prompts, tool calls and responses as structured traces in Deeplake, and makes them searchable by any connected agent.

A background worker mines those traces for repeated patterns and writes them as SKILL.md files scoped to a workspace, so a fix one engineer's agent discovered can be picked up by teammates' agents. It also generates wiki-style session summaries using the host agent's own CLI, and builds a codebase graph from captured activity to guide search and recall.

Data is stored per org and workspace, with options to route a directory to a different workspace via a .hivemind file, disable capture per directory, or bring your own GCS, Azure or S3 bucket. Claude Cowork support is marked Alpha.

Key features

  • 01Captures prompts, tool calls and responses from supported coding agents as structured traces.
  • 02Mines repeated patterns from traces and codifies them into SKILL.md files shared across the team.
  • 03Hybrid lexical and semantic search over traces and skills, with optional local embeddings.
  • 04Per-directory .hivemind config to route traces to a workspace or disable capture.
  • 05Bring-your-own-cloud storage on GCS, Azure Blob or S3.

Use cases

  • A team wants an agent to reuse a migration or debugging pattern another engineer's agent already solved.
  • Developers running several coding assistants want one shared memory instead of per-tool context.
  • An organization needs coding-agent history kept in its own cloud bucket.

Quick start

  1. 1

    Install via npm (Node 22+ required) and run the installer

    npm i -g @deeplake/hivemind && hivemind install
  2. 2

    Check which assistants are wired up

    hivemind status
  3. 3

    Restart your assistants after install

Things to know

  • Requires Node 22 or newer for the npm install path.
  • Shared memory needs a Deeplake account or API token; without sign-in, only hooks are installed.
  • Claude Cowork support is Alpha, and auto-capture covers only Local Agent Mode sessions.

Topics

#ai-agents#ai-memory#claude-code#coding-agents#long-term-memory#rag#embeddings#postgres#ai#anthropic

FAQ

What is hivemind?
Shared, auto-learning memory layer that captures coding-agent sessions, mines them into reusable skills, and shares them across Claude Code, Codex, Cursor and other agents. Hivemind gives a team of coding agents a shared, cloud-backed memory. It installs hooks into supported assistants such as Claude Code, OpenClaw, Codex, Cursor, Hermes Agent and pi, records each session's prompts, tool calls and responses as structured traces in Deeplake, and makes them searchable by any connected agent.
Is hivemind free to use? What license does it have?
hivemind is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is hivemind still maintained?
Actively maintained. The last commit to hivemind was on Oct 22, 2026.
What can I use hivemind for?
A team wants an agent to reuse a migration or debugging pattern another engineer's agent already solved. Developers running several coding assistants want one shared memory instead of per-tool context. An organization needs coding-agent history kept in its own cloud bucket.
How do I install hivemind?
Install via npm (Node 22+ required) and run the installer: npm i -g @deeplake/hivemind && hivemind install Check which assistants are wired up: hivemind status Restart your assistants after install
Where can I get hivemind?
The source code and installation instructions are on GitHub: https://github.com/activeloopai/hivemind

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects