Pricing
Pay for work done,
not seats filled
We meter the agent run — one task, executed end to end by one specialist. It is the thing that costs us money and the thing that leaves you with something. Viewers are free. Going over bills; it never blocks a deploy.
Free
One project, real gates, no card.
1 project · 100 agent runs a month
Everything you need to see it work:
- 3 core roles — manager, full stack, reviewer
- 1 repository, 1 project
- Manual autonomy
- 1 run at a time
- Community support
No card. No trial clock. It stays free.
Starter
For one team shipping one product.
1,000 agent runs a month
Everything in Free, plus:
- 8 specialist roles
- 3 projects, unlimited repositories
- Supervised autonomy
- 1 deployment target with rollback
- 3 concurrent runs
- Email support, one working day
14-day trial. Viewers never count as users.
Pro
For teams that want it to run itself.
5,000 agent runs a month
Everything in Starter, plus:
- All 14 roles plus custom roles
- Unlimited projects
- Autonomous autonomy
- Unlimited deployment targets
- Production investigations
- 10 concurrent runs
- Priority support, four hours
14-day trial. Move down a plan any time.
Business
For agencies and multi-client work.
20,000 agent runs a month
Everything in Pro, plus:
- Bring your own model keys
- Per-client project isolation
- Audit log export
- 25 concurrent runs
- Named contact and onboarding
Platform fee plus seats. Seats are per person who acts.
Autonomy is the upgrade trigger, not a paywall on your own data: Manual is free, Supervised is Starter, Autonomous is Pro. You pay for trust as you earn it. Compare every line.
Enterprise
We are not ready for your enterprise. Here is exactly why.
Most pricing pages put a Contact sales button in this space and let a discovery call find out what is missing. That costs you three weeks and a meeting you did not need.
So we publish the list instead. Read it, and you can rule us in or out in about thirty seconds. If everything you need is already there, start on Business today. If one line is a hard requirement, we are not your platform yet, and we would rather you knew that now than after procurement.
Join the waitlist and we will write to you when the specific item you need has shipped. Not before, and not to check in.
What we do not have yet
- SSO and SAML at identity-provider grade — no enforced sign-in through Okta, Entra ID or Google Workspace, and no SCIM provisioning
- A completed SOC 2 audit — no report, no Type I, no Type II, no bridge letter, and no auditor engaged today
- A contractual uptime SLA — we publish status openly and we do not sign availability terms or service credits
- Dedicated tenancy — every customer runs on shared, multi-tenant infrastructure, with no single-tenant or on-prem option
- A procurement motion — no MSA redlines, no security-questionnaire desk, no vendor onboarding portal, no named legal counsel on standby
Four of the five sit on our roadmap with the quarter we expect them. The fifth arrives when we hire someone to run it. Read what we do and do not have on the security page.
Compare
Every line, on one page
Nothing is hidden behind a sales call. Where a row is empty for every plan, it is empty because we have not built it.
| Feature | Free | Starter | Pro | Business |
|---|---|---|---|---|
| Projects & repositories | ||||
| Projects | 1 | 3 | Unlimited | Unlimited |
| Connected git repositories | 1 | Unlimited | Unlimited | Unlimited |
| Requests on the board | Unlimited | Unlimited | Unlimited | Unlimited |
| Plain-English plans with a task graph | ||||
| Structured reports on every task Summary, files modified, risks, assumptions, blockers |
||||
| Design previews for front-end work | — | |||
| Per-client project isolation | — | — | — | |
| Export your requests, plans and reports | ||||
| The agent team | ||||
| Specialist roles | 3 core | 8 | All 14 | All 14 |
| Editable Markdown role contracts | — | |||
| Custom roles you write yourself | — | — | ||
| Capability substitution when a role is absent Recorded on the task, never silent |
||||
| Model and effort overrides | — | Per project | Per project and request | Per project and request |
| Bring your own model keys | — | — | — | |
| Concurrent runs | 1 | 3 | 10 | 25 |
| Autonomy & control | ||||
| Autonomy levels | Manual | Manual, Supervised | All three | All three |
| Plan approval before anything runs | ||||
| Mark any role as requiring human review Dependent work stops at every autonomy level |
||||
| Agents pause to ask you a question | ||||
| Isolated branch and git worktree per task | ||||
| Dependency graph across tasks | ||||
| Live run streaming | ||||
| Deployment & operations | ||||
| Deployment targets over SSH | — | 1 | Unlimited | Unlimited |
| Health checks and diagnostics | — | |||
| Automatic rollback to the last healthy commit | — | |||
| Deployment history with merge and deploy output | — | |||
| Production investigations Read-only access by default; write access is an explicit grant |
— | — | ||
| Investigations open follow-up requests | — | — | ||
| Included agent runs a month | 100 | 1,000 | 5,000 | 20,000 |
| Going over the allowance A deploy, a rollback and an investigation always run |
Runs pause until the month resets | Billed, never blocked | Billed, never blocked | Billed, never blocked |
| Support & administration | ||||
| Support | Community | Email, one working day | Priority, four hours | Named contact |
| Guided onboarding | — | — | — | |
| Read-only viewers | Free | Free | Free | Free |
| Owner and admin roles | ||||
| Audit log export | — | — | — | |
| SSO and SAML Not built. On the roadmap for Q4 2026 |
— | — | — | — |
| SCIM provisioning Not built, and not yet scheduled |
— | — | — | — |
| Dedicated tenancy Not built. Shared infrastructure for every customer |
— | — | — | — |
| Contractual uptime SLA Not offered. We publish status instead |
— | — | — | — |
| SOC 2 report No audit completed. SOC 2 Type II sits in 2027 on the roadmap |
— | — | — | — |
Metering
How the meter works
One unit, counted the same way on every plan. Read this before you pick a tier — it decides the number that matters far more than the price does.
One task, one run
An agent run is one task executed end to end by one specialist: it reads the code, works on its own branch in its own worktree, and files a structured report. Plan a request into six tasks and you spend six runs, plus the planning run.
Watching is free
Reading the board, opening a plan, reviewing a diff, approving a gate, browsing reports and sharing a design preview all cost nothing. We meter work, not attention.
Our failures are on us
A run that fails because our platform errored is not counted. If one slips through, tell support and we credit it. A run that finishes and reports the work cannot be done does count — that answer is the value.
Overage bills, never blocks
Cross your included runs on a paid plan and the extra runs are billed on your next invoice. A deploy, a rollback and an investigation always run. Stopping a pipeline to force an upgrade is a dark pattern and we have banned it.
Controls before the bill
Set the model and effort per project, and per request when one job deserves more care. Concurrency caps how much can run at once. Usage is tracked per agent account so you can see where the runs went.
Move when it stops fitting
Change plan whenever you like. Upgrades apply straight away and we prorate the difference. Downgrades apply at your next renewal, so you keep what you paid for until then.
Billing questions
The awkward ones, answered
What is an agent run, exactly?
One task, executed end to end by one specialist. The agent reads your code, does the work on its own branch inside its own git worktree, and ends with a structured report — summary, files modified, risks, assumptions, recommendations, blockers.
A request that the Project Manager plans into six tasks spends six runs, plus one for the planning itself. A review gate that rejects work and sends it back costs another run when the fix is made.
What happens when I hit the included limit?
On a paid plan, nothing stops. Runs past your allowance are billed at the overage rate and land on your next invoice.
We will never block a deploy, a rollback or an investigation because you crossed a threshold. Holding your pipeline hostage to sell you a tier is a dark pattern, and it is banned in writing.
On the Free plan there is no overage and no card on file, so runs pause until the month resets.
Do read-only viewers cost anything?
No, and there is no cap on them. Invite your co-founder, your designer, your accountant or your client to watch the board for nothing.
You pay for people who act: filing requests, approving plans, dispatching work, managing deployment targets. A stakeholder who only ever looks at the board is not a seat.
Can I change plan in the middle of a month?
Yes. Upgrade whenever you like — it takes effect immediately, the new allowance applies at once, and we prorate the difference on your next invoice.
Downgrades take effect at your next renewal date. You keep the plan you already paid for until then, rather than losing it the moment you click.
Do you give refunds?
Yes, within 30 days of a charge, if the platform did not do what this page says it does. Email support@capybari.com and tell us what happened.
For annual plans cancelled part-way through, we refund the unused whole months on request. We would rather return the money than keep a customer who has stopped using us.
How does annual billing work?
You pay for ten months and get twelve — 2 months free. That is why Starter falls from $49 to $41 a user a month, Pro from $149 to $124, and Business from $499 to $416.
We quote it as two months rather than a percentage on purpose. A percentage discount on a per-user, per-month price is arithmetic you should not have to do.
What if I go a long way over my allowance?
You get an email at 80% and again at 100% of your included runs, and the number is on the board the whole time.
If your overage is consistently larger than the gap to the next tier, support will write and tell you to move up. We would rather have the smaller invoice and the customer who trusts it.
Do you charge for runs that fail?
Not when the failure is ours. A run that dies because our platform errored is not counted, and if one is counted by mistake we credit it on request.
A run that completes and reports bad news — the tests fail, the approach will not work, the blocker is in a system we cannot reach — does count. Finding that out is the work.
Do you offer non-profit, education or student pricing?
Yes. Registered charities and non-profits get a standing discount, and students and academic researchers can use a paid plan free for the length of a course or project.
Write to support@capybari.com from your organisation or institution address and ask. There is no form, and no call.
Which currency do you bill in, and what about VAT?
Prices on this page are in US dollars and every invoice is issued in US dollars, wherever you are.
VAT or local sales tax is added at checkout where we are required to charge it, and shown as a separate line on the invoice. Business customers in the UK and EU can enter a valid VAT number; reverse charge is applied where it applies.
Do I need a card to start?
Not for the Free plan — one project, 100 runs a month, Manual autonomy, real review gates, no expiry.
Starter and Pro trials run for 14 days and do take a card, so that nothing pauses on day 15 mid-request. Cancel inside the trial and you are not charged.
What happens to my work if I cancel?
Your code was never ours. It lives in your repositories, on branches you own, deployed to servers you control.
Export your requests, plans and reports at any time, before or after you cancel. We keep your account readable for 30 days after cancellation, then delete it. The security page sets out where data sits and who can reach it.
Are you ready for my enterprise?
Probably not, and the panel above lists precisely what we are missing. If you need SSO at identity-provider grade, a completed SOC 2 audit, a contractual uptime SLA or dedicated tenancy today, we are not your platform yet.
Join the waitlist and we will write when the item you need has actually shipped.
Something here still not clear? Ask us — a person replies.
Start on Free. Move when it stops fitting.
One project, 100 agent runs a month, Manual autonomy and the same review gates as every paid plan. Connect a repository and file one request that has been waiting a month.
No card required · Viewers always free · Export everything, any time