The engine is free. The controls are the product.
Self-hosted AI infrastructure, licensed per device — a flat license per host, unlimited decisions. We don't meter your AI agents and we couldn't: the engine runs inside your perimeter and nothing phones home. One license, one host: nobody gets to run a datacenter on a hobby plan.
One license covers one device — one host running the engine. Decisions on it are unlimited: no quotas, no overage, no telemetry.
Community
The full engine on your hardware.
- Complete self-hosted core — API, SDKs, MCP
- 1 device · 1 worker
- 1 seat
- Community support
Pro
For a product in production. Licensed per device.
- Everything in Community
- 5 workers per device
- 1 seat
- Email support
Team
For platforms serving many surfaces. Licensed per device.
- Everything in Pro
- 20 workers per device
- 15 seats
- Staging + production environments
- Priority support
Enterprise
For infrastructure that must answer to auditors.
- Everything in Team
- Air-gapped mode
- Fleet licensing — your device count, one agreement
- 24/7 SLA · dedicated engineer option
- Security review support
Four steps. One engine.
Tiers change headroom — workers, seats, governance — never capability. The runtime under your agents on the $0 step is the runtime on the $24k step.
No meter. No telemetry. No surprises.
Per-decision pricing needs a counter watching your agents — which means your infrastructure reports to someone else's. Corin refuses the premise. The license is flat per device, verified locally, and the engine sends nothing back. Your usage is not our business.
- Decisions
- Unlimited. No quotas, no overage rates, no per-call pricing. Throughput is whatever your hardware sustains — that is the point of owning it.
- Per-device licensing
- A device is one host running the engine — VM, bare metal, or container node. The license key carries the device and worker count, enforced locally.
- Usage tracking
- None. The engine reports nothing back — no counters, no phone-home, no analytics. We couldn't see your usage if we wanted to.
- Air-gapped licensing
- License-key capacity, verified offline. The strict-privacy profile stays strict — activation included.
- Predictability
- The invoice is devices × flat license. It changes when you add hardware, not when your agents get busy.
The usual questions, answered plainly.
Is Community really the full engine?
Yes — the same runtime, kernels, and surfaces. The differences are capacity and governance — workers, seats, environments — never capability. Benchmarks reproduce on Community.
What counts as a device?
One licensed device = one host (VM, bare metal, or container node) running the engine. Replicas on the same host don't multiply the count; new hosts claim a device slot from your license.
How does annual billing work?
Pay for 12 months, get roughly 20% off the monthly price. Enterprise is annual-only.
Do you meter or track my usage?
No. Decisions on a licensed device are unlimited, and there is no telemetry — the engine runs inside your perimeter and reports nothing back. We couldn't count your decisions if we wanted to.
Can I move from cloud-optional to air-gapped later?
Yes. The deployment is Docker Compose either way; the air-gapped profile is a configuration change plus an Enterprise license key.
What if a license lapses?
The deployment downgrades to Community. The engine keeps running on your hardware at the free tier's capacity — paid workers, seats, updates and support pause until the license renews. Nothing reaches into your deployment; the key simply expires locally.