Claude Code · OpenClaw · Codex · Cursor · Hermes · pi
Your agents learn as a team.
LoCoMo benchmark, 100 QA, Claude Haiku via claude -p. $6.65 vs $8.94 / 100 QA, 1,008 vs 1,700 tokens / Q, 6.2 vs 8.9 turns / Q. Hybrid lexical + semantic retrieval.
HOW IT WORKS
Install and forget
One command wires every assistant on your machine into a shared memory layer. Every interaction is captured to a persistent store that any agent can read from. Claude Code, OpenClaw, Codex, Cursor, Hermes, pi, and your custom agents. One brain, no silos, no RAG pipeline to babysit.
Auto-Capture
Every prompt, tool call, and response is captured automatically into your shared Deeplake workspace. No manual saving. No extra commands.
Instant Recall
Ask any agent what any team member was working on. Search by keyword, project, or topic. BM25 full-text by default; semantic embeddings opt-in.
Shared Across Agents
Every assistant in your org reads from and writes to the same memory. One agent's context becomes every agent's context, in real time.
FOR TEAMS
Built for teams, isolated by design
Every org gets its own brain. Every workspace, its own room.
Hivemind gives your team shared agent memory without the crosstalk. Org-level isolation keeps your agents' context walled off from everyone else on the planet. Workspace-level isolation keeps your growth team's agents out of your finance team's ledger, and vice versa.
One company, many workspaces, zero leaks.
Org Isolation
Your agents' memory lives in a tenant only your org can reach. Not shared. Not co-mingled. Not one misconfigured IAM policy away from a breach.
Workspace Isolation
Spin up a workspace per team, per project, or per customer. Each one is its own sealed context, and agents only see what they're invited to.
Granular Access
Invite teammates to the workspaces they need. Revoke in one click. Audit everything.
Think of it like Slack: your company has one org, your teams have their own channels, and nothing leaks between them unless you say so.
SECURITY & STORAGE
Your memory, your cloud
Hivemind Cloud is the default. When that isn't enough, point Hivemind at storage in your own cloud. We handle the orchestration. Data never leaves the perimeter.
Same agents, same recall, backed by storage you own.
Encrypted in transit and at rest
TLS between every agent and Deep Lake. AES-256 on the bytes once they land. Your cloud credentials live in Deep Lake's vault, and Hivemind never sees the raw keys.
Tenant-isolated by design
Org and workspace boundaries are enforced at the storage layer, not just the API. Your sessions never share a row, a partition, or an index with another workspace.
You stay in control
Disable capture per session with HIVEMIND_CAPTURE=false. Local credentials are written with 0600/0700 permissions. Delete a workspace and the underlying objects go with it.
BRING YOUR OWN CLOUD
BYOC: connect your own cloud storage
Connect Hivemind to storage you already run. GCS, Azure Blob, and AWS S3 are supported today. S3-compatible on-prem is available on request.
GET STARTED
One command. Every agent on your machine.
The unified installer detects every supported assistant, wires up the hooks, and opens your browser once for SSO. Restart your assistants and they all share the same brain.
npm install -g @deeplake/hivemind && hivemind installAuto-detects Claude Code, OpenClaw, Codex, Cursor, Hermes, and pi.
After npm install -g, target a single assistant instead of all of them:
Claude Code
hivemind claude installOpenClaw
hivemind claw installCodex
hivemind codex installCursor
hivemind cursor installHermes
hivemind hermes installpi
hivemind pi installCheck what's wired up
hivemind statusRemove from every detected assistant
hivemind uninstallDisable capture for a single session with HIVEMIND_CAPTURE=false. Enable verbose hook logs with HIVEMIND_DEBUG=1.
FEATURES
Everything your agents need to remember
Built for teams that run multiple AI agents and need them to share context without manual wiring.
AI Session Summaries
A background worker writes a wiki-style summary of every session — key decisions, code changes, next steps — and indexes it for recall.
Org-Level Access Control
Invite team members with ADMIN, WRITE, or READ permissions. Switch between orgs and workspaces on the fly.
Virtual Filesystem
Memory looks like a folder under ~/.deeplake/memory/. Use cat, ls, grep — the same commands you already know. No new APIs to learn.
Real-Time Sync
All writes go to Deeplake Cloud and are instantly available to every agent in the org. Sub-second propagation.
Zero Infrastructure
No databases to run. No servers to maintain. Deeplake Cloud handles storage, indexing, and search as a managed service.
Privacy Controls
Opt out of capture anytime with HIVEMIND_CAPTURE=false. Credentials stored locally with strict file permissions (0600 / 0700).
USE CASES
Memory that compounds
The more your team uses Hivemind, the smarter every agent gets.
TEAM ONBOARDING
New engineers get full context on day one
Every architectural decision, debugging session, and code review is captured. New team members can ask their agent about any past decision and get the full story, not just what's in git blame.
you: What approach did the team take for the auth refactor?
agent: Based on session history, the team chose JWT with rotating keys. Decision made on March 12, driven by compliance requirements flagged by legal.MULTI-AGENT WORKFLOWS
Agents that pick up where others left off
Run a research agent in one session, a coding agent in another. The coding agent already knows what the research agent found. No copy-pasting context between sessions.
agent-1: Analyzed 3 API options. Recommended Stripe for billing. Findings saved to memory.
agent-2: Building billing module. Found agent-1's analysis in memory. Proceeding with Stripe integration using recommended patterns.INCIDENT RESPONSE
Never debug the same issue twice
When your agent debugs a production issue at 2am, the fix, root cause, and full investigation are captured. Next time, any agent on the team can find it instantly.
you: Have we seen this OOM error before?
agent: Yes, session from Feb 8. Root cause was unbounded query results in the analytics pipeline. Fix: added LIMIT + pagination. PR #847 has the full diff.Give your agents a shared brain
One command, every agent. Free for individual developers.
npm install -g @deeplake/hivemind && hivemind install