Free for individuals. Per-seat for teams.
A seat is a person — every AI client they use (Claude Code, Cursor, Codex, Claude Desktop, Hermes, MCP clients) and every agent they run, unlimited. No call meters, no usage anxiety, no tax on coverage.
Individual
One person. Every client. Every agent. No limits.
- Unlimited calls, agents & clients
- Identity, policy, limits, PII & audit
- Full tracing & cost X-ray
- Hardline floor & inline approvals
- Self-optimization MCP tools
- 30-day audit retention
- Community support
Team
Starts the moment you add a second person. Less than any one of the AI tools it governs.
- Everything in Individual, per person
- Unlimited calls & agents per seat
- 5 service-agent identities per seat
- Shared policies & org-wide rules
- Team approvals & standing grants
- Org-wide audit & cost rollup, per person
- 1-year audit retention
- Email support
Enterprise
For compliance, scale, and deployment control.
- Volume seat pricing
- SSO / SAML · SCIM
- VPC / on-prem / self-host
- Unlimited audit retention
- Unlimited service-agent identities
- SOC 2 evidence exports · DPA / BAA
- Dedicated support & SLA
A seat is a human. Their agents — coding sessions, autonomous runs, subagents, delegation chains — are covered by that seat, unlimited. Headless agents that belong to no person (CI bots, scheduled agents) use service-agent identities.
Why free for individuals, forever?
No trick. We want individuals to try ACP, run it on everything, and bring it to work once we've proven the value. Governing one person's agents costs us almost nothing to serve — the product earns its keep when a team shares one policy surface, one audit trail, and one bill. That's what seats pay for.
Self-host for free
All six governance layers are available as MIT-licensed npm modules. Use them individually or compose the full pipeline. No account required. Deploy anywhere.
Pricing FAQ
A human being. One seat covers that person's entire AI surface — every client they use (Claude Code, Cursor, Codex, Claude Desktop, Hermes, any MCP client) and every agent, subagent, and delegation chain those clients spawn. We don't count calls, agents, or tools. Humans are the unit because identity, approvals, and accountability are the unit.
No call caps on any plan, and ACP never cuts you off — if anything ever needs attention we warn you, loudly, and keep governing. Blocking your agents mid-task is the one thing a control plane must never do to you.
Agents that belong to no person — CI bots, scheduled agents, production services. Each Team seat includes five; Enterprise is unlimited. Agents a person runs from their own clients don't count against this — those are covered by their seat.
Self-hosted uses the open-source npm modules in your own infrastructure. Managed (ACP Cloud) adds the web console, built-in integrations, agent-to-agent governance, multi-tenant isolation, and we handle deployment and scaling.
Yes. Individual is free forever with no card. Team is monthly per seat — add and remove seats as your team changes.
ACP is not yet certified for SOC 2, HIPAA, or ISO 27001 — those certifications are on the roadmap, and we say so plainly on our trust page. What ACP gives you today is the evidence layer those frameworks ask for: identity-attributed logs of every tool call, redaction findings, policy decisions, and per-user attribution — audit exports built to serve as evidence in your own certification work.
Govern every AI client your team runs
Start free as an individual. Add your team when you're ready — one seat per person, everything included.