Hivemind
activeloopai/hivemind
Shared agent memory that turns session traces into reusable skills, with a Hermes install option
Hivemind is a cloud-backed shared memory layer that captures agent sessions as traces and turns repeated patterns into reusable SKILL.md files. Hermes is one of the supported assistants, installed with hivemind hermes install.
What Hivemind does
Hivemind captures every session's prompts, tool calls and responses as structured traces in Deeplake, then mines them for repeated patterns and writes those patterns into reusable SKILL.md files that every agent on a team can use. Search combines lexical and semantic retrieval, and a virtual filesystem over ~/.deeplake/memory/ is backed by SQL. At session end a background worker summarizes sessions into wiki pages.
One installer detects the supported assistants on a machine and wires up hooks, then opens a browser for sign-in. Claude Code, OpenClaw, Codex, Cursor, Hermes and pi each have an install subcommand, and Claude Cowork is marked Alpha. Data can stay in your own GCS, Azure, S3 or on-prem bucket. On the LoCoMo memory benchmark (100 QA pairs, Claude Haiku via claude -p), the README reports 25% lower cost, 1.7x fewer tokens per question and 31% fewer turns than a no-memory baseline.
Key features
- Captures prompts, tool calls and responses as structured traces
- Codifies repeated patterns into SKILL.md files shared across the team
- Hybrid lexical and semantic search over traces and skills
- Per-assistant install, including hivemind hermes install
- AI-generated wiki pages from sessions through a background worker
- Bring-your-own-cloud storage in GCS, Azure, S3 or an on-prem bucket
When to use it
- Share what one engineer's Hermes agent learned with the rest of a team
- Keep one memory layer across Hermes, Claude Code, Codex and Cursor
- Turn repeated debugging patterns into skills automatically
Who it is for: Teams that run several AI agents and want them to share memory and learned skills.
How it fits with Hermes Agent
Hermes is one of the listed supported assistants, with its own install subcommand, and shares the same memory and skills as the other agents.
How to install Hivemind
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
curl -fsSL https://deeplake.ai/hivemind.sh | sh
npm i -g @deeplake/hivemind && hivemind install
hivemind hermes installRequirements: Node 22+ and a writable npm prefix for the npm route, plus sign-in to a Deeplake account (browser or API token)
Note: Hivemind is cloud-backed and needs a sign-in to a Deeplake account, although data can be kept in your own storage bucket.
FAQ
What is Hivemind?
Hivemind is a shared memory layer for AI agents. It records sessions as traces, searches them, and turns repeated patterns into reusable SKILL.md files for every agent on a team.
Does Hivemind work with Hermes Agent?
Yes. Hermes is on the supported assistant list, and hivemind hermes install installs it for that agent only. The general installer also detects supported assistants on the machine.
How do I install Hivemind?
On macOS or Linux, run curl -fsSL https://deeplake.ai/hivemind.sh | sh, or on Windows use the PowerShell installer from the README. The npm route is npm i -g @deeplake/hivemind && hivemind install.
Similar memory for Hermes Agent
All memoryMemory agent that curates, reconciles and shares long-term memory across multiple AI agents
codejunkie99 agentic-stackPortable .agent/ folder of memory, skills and protocols that moves between coding-agent harnesses
MemTensor MemmyLocal memory hub and personal agent that shares one memory across Claude Code, Codex, OpenClaw and Hermes
ClaudioDrews Memory OSSeven-layer local memory system for Hermes Agent with Qdrant, structured facts and a curated wiki
mem9-ai mem9Persistent memory layer for AI agents with a Hermes Agent plugin, hosted API or self-hosted server
AVIDS2 MemorixLocal-first shared project memory for coding agents over MCP, including Hermes Agent
Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles