HiveMind

What is HiveMind?

HiveMind is a shared dashboard from Weights and Biases that records your team's coding agent sessions. A small daemon on each machine picks up Claude Code, Codex, Cursor, and Gemini CLI work, strips secrets, and uploads it. You can then search, replay, and fork sessions and see what each one cost.

Top Features:

• Cross-agent capture: one daemon logs sessions from Claude Code, Codex, Cursor, and more.
• Spend tracking: estimated cost by person, repo, and model, down to each merged PR.
• Session search: find past work by prompt, file, tool, or repo, then fork it.

Use Cases:

• Cost reviews: show where the agent token bill went, team by team.
• Workflow reuse: find an earlier team session and pick up where it left off.
• Adoption tracking: see which teams use agents and how much agent code ships.

Who Can Use HiveMind?

• Developers: review your own agent sessions and reuse prompts and skills that worked.
• Engineering leaders: follow adoption, spend, and merged output across teams and repositories.
• IT admins: roll the daemon out through JAMF or MDM with SSO and SCIM.

Pricing

• Free plan ($0): hosted dashboard with monthly quotas for AI analysis and PR walkthroughs.
• Self-host trial (30 days free): run it on your servers with every premium feature.
• Team and Enterprise (contact sales): higher limits, SSO, insights, and production self-host licenses.

Pros and Cons

Pros:

• One view: every supported agent lands in the same dashboard without per-agent setup.
• Secret scrubbing: the daemon strips secrets before session data leaves your machine.
• Self-host option: teams can run the whole stack on their own Docker infrastructure.

Cons:

• Cost estimates: spend figures come from list prices, not your actual vendor bills.
• Unclear pricing: paid plan prices are not public, so you must contact sales.
• Transcript uploads: session transcripts go to a hosted service unless you self-host.

FAQs:

1) Is HiveMind free?

Yes, a hosted Free plan has monthly AI quotas, and paid tiers raise them.

2) Which agents are supported?

Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, OpenCode, and Pi all work.

3) Can I self-host it?

Yes, the CLI runs a Docker Compose stack on a Linux host.

4) How do I install it?

Run the install script or uv, then start the daemon and sign in.

5) Are cost figures exact?

No, they are estimates from published model prices and session token counts.