Pricing

Pricing for the runtime.

You pay for the licensed limits and the support that comes with them. The model is billed to you by your own provider, at your own rates, and we never touch it. Prices are in US dollars and exclude taxes.

The four plans

Billing period

Community

Run a whole AI team on your own machine, free.

Freeno licence needed, no expiry

projects open
1
agents per mesh
8
turns at once
4
  • The full runtime: event-sourced kernel, policy enforcement, scheduler, artifacts, replay
  • Dashboard, designer, CLI and the project host with its spend ceiling
  • Token-free demo and stub runtime
  • One open project, up to 8 seats, 4 concurrent turns

Support: Community (issues and documentation)

Start free

Team

Several projects under one host, with usage reporting.

$124per month, billed annually ($1,488 a year)

$149per month, billed monthly

projects open
5
agents per mesh
12
turns at once
8
  • Everything in Community
  • Up to 5 projects open under one host, 12 seats per mesh, 8 concurrent turns
  • Usage export (JSON and CSV) and Prometheus metrics

Support: Email, next business day

Get Team

Business

A fleet of projects, with priority support and a deployment you can hand to ops.

$499per month, billed annually ($5,988 a year)

$599per month, billed monthly

projects open
25
agents per mesh
30
turns at once
24
  • Everything in Team
  • Up to 25 projects, 30 seats per mesh, 24 concurrent turns
  • Support for the container image and Helm chart deployment, with upgrade and backup runbooks
  • Security questionnaire support

Support: Priority email, 4 business hours first response

Planned, not included: Single sign-on (OIDC); Per-operator identity and roles; Audit-log export with operator identity

Get Business

Enterprise

Unlimited scale, your terms, your environment.

Let’s talkquoted for your terms

projects open
Unlimited
agents per mesh
Unlimited
turns at once
Unlimited
  • Everything in Business
  • No seat, project or concurrency limits
  • Air-gapped and private-registry installs
  • Custom terms, security review and a support agreement

Support: Named contact and a support agreement

Planned, not included: Single sign-on (OIDC); Per-operator identity and roles; Audit-log export with operator identity

Talk to us

Plan prices are for the runtime. Model usage is separate: the agents run on model credentials you own (any OpenAI-compatible provider, the Anthropic API, Bedrock, Vertex or Foundry), and your provider bills you for it.

Planned, not included: single sign-on, per-operator identity and roles, and audit-log export with operator identity are on the roadmap. Today there is one shared operator credential: put the instance behind your VPN or an authenticating proxy.

Compare

The plans, side by side.

The four plans side by side: price, limits, features and support. Prices are in US dollars and exclude taxes.
PlanCommunityTeamBusinessEnterprise
Per month, billed annuallyFree$124 ($1,488 a year)$499 ($5,988 a year)Quoted
Per month, billed monthlyFree$149$599Quoted
Projects open at once1525Unlimited
Agents (seats) per mesh81230Unlimited
Concurrent agent turns4824Unlimited
Usage export and Prometheus metricsNoYesYesYes
SupportCommunity (issues and documentation)Email, next business dayPriority email, 4 business hours first responseNamed contact and a support agreement

A seat is one agent in a mesh; the operator at the console is not a seat. A project is one mesh, and the limit is how many are open at once: registering more is free. Concurrent turns is how many agent turns may run at the same moment across the whole host.

What you are paying for

Licensed limits, and the support that comes with them.

A plan is a licence to run the runtime up to a size, plus the help that goes with it. It is not a price for the model, and it is not a lock.

Licensed limits

How many projects may be open at once, how many agents a mesh may have, and how many agent turns may run at the same moment. The Community plan’s limits are what the source licence lets anyone run in production for free; a paid plan is the licence to run beyond them.

Features

Usage export (JSON and CSV) and Prometheus metrics, from the Team plan. The event log that the usage report is computed from is yours on every plan.

Support

The Community plan is supported through the issue tracker and the documentation. Team and Business add email support with the response targets in the table, and Business adds support for the container image and Helm chart deployment. Enterprise has a named contact and a support agreement.

What you are not paying for

  • Model usage.The agents run on credentials that belong to you, and your provider bills you for them, with your own discounts and rate limits. We do not resell it and we do not add a margin to it.
  • A hosted service.Curule is software you run, on your own infrastructure. We cannot see, start or stop anything on it.
  • Anything on the roadmap.A feature is listed as planned until it exists and is tested, and it is never sold as included.

Estimate

Which plan, and roughly what will it cost?

An estimate, not a quote. The model figure comes from one measured mission, and yours will differ.

The mission it is based on

We have measured one mission class: five agents built a small library in 31 minutes for $5.52 of model usage, and an independent check scored it 99.6%. One run is not a rate. The table prices the same tokens at each model’s list price; another model would take a different number of turns, so read the spread as a range of prices and not as a forecast.

The tokens of the measured mission priced at each model's list price of 2026-10-01, and a month of 10, 25 or 100 such missions
ModelThis mission10 a month25 a month100 a month
Claude Haiku 4.5 what it ran on$5.52$55.20$138.00$552.00
Claude Sonnet 5.5$11.03$110.30$275.75$1,103.00
Claude Opus 5.5$15.60$156.00$390.00$1,560.00
Claude Fable 5.1$30.93$309.30$773.25$3,093.00

Each model’s list price on 2026-10-01. The product’s own estimate of spend is a backstop and not an invoice: your provider’s bill is the one that counts.

How licences work

A key by email, checked on your machine.

There is no checkout, no account and no billing system on this site. A paid plan starts with an email, and the licence key is issued by hand.

  1. Write to us

    Say which plan, how many projects and agents you run, and the name the licence should carry.

  2. We issue a key by hand

    A licence is one line of text, signed with a key that stays with us. We send it to you by email.

  3. Install it on the host

    curule license install <key>

    Or put it in MESH_LICENSE in the environment. A running host reads it again within 30 seconds, so there is nothing to restart.

  4. It is checked offline

    The signature is verified on your machine. Nothing phones home, and an air-gapped install verifies a licence exactly as a connected one does.

  5. When it expires

    You are told 30 days ahead. After the date the plan’s limits stay in force for a grace period (14 days by default), and then the instance is on the Community plan. Nothing is deleted and nothing that is running is stopped; install a renewed key and the plan returns.

Limits are reported, not enforced, by default. MESH_LICENSE_ENFORCEMENT is warn unless you change it: an instance over its plan’s limits says so in the log, the dashboard, the API and the metrics, and keeps running. With enforce, a mesh over its seat limit will not start and a project over the open-project cap will not open; what is already running is never stopped. The licence terms, not a lock in the code, are what bind a customer to a plan.

Questions

Pricing questions.

Is the price per agent?

No. A price is flat for a plan. The plan sets how many projects may be open, how many agents one mesh may have and how many agent turns may run at the same moment.

Does the price include the model?

No. The agents run on model credentials you own (any OpenAI-compatible provider, the Anthropic API, Bedrock, Vertex or Foundry), and your provider bills you for the model. The calculator above shows the two figures separately, and the model figure is an estimate from one measured mission.

What happens if we go over a limit?

By default you are told and nothing is refused. Running beyond the Community plan’s limits in production is what a paid plan licenses; without one it is outside the source licence, whatever the runtime lets you do.

Can the price change under us?

A licence carries its plan, its limits and its expiry, so a change to the list price takes effect at renewal, when a new licence is issued.

How do we buy?

Write to us: see how licences work above. There is no checkout, and we agree the terms with you before we issue a key.

Can you host it for us?

Not today. It is software you run on your own infrastructure, with your own provider credentials.

Talk to us about a paid plan.

Tell us which plan, how many projects and agents, and whether you would pay monthly or annually.