// pricing — cat PLANS.txt
> Pay for green checks.
Seats are flat. Replays are metered, printed per-hunt, and cheap enough that the median hunt costs less than the coffee you'd have finished debugging by hand. No annual-contract ambush; the enterprise tier exists for compliance, not for hiding the price.
$ plan --oss
Open source
The full CLI under MIT. Hunts run on your machine with your own model API key.
- + local replays, unlimited
- + bring-your-own model key
- + single repo per hunt
- + community support
$ plan --team
most deployedTeam
Hosted replay fleet, the CI sentinel, and draft PRs across every repo in your org.
- + 500 hosted replays / seat / mo included
- + ci sentinel on unlimited repos
- + draft PRs with regression tests
- + codebase memory (beta)
- + slack / pagerduty delivery
$ plan --enterprise
Enterprise
Everything in Team, run inside your network, with the paperwork your auditors want.
- + self-hosted or dedicated VPC
- + SSO / SCIM · signed audit-log export
- + private model endpoints (Bedrock, Azure)
- + 99.9% sentinel SLA · named support
// replay math
> What a replay costs
A replay is one sandboxed execution of your failing check against one commit. The median hunt uses five. Overage is billed per replay, printed live during the hunt, and capped by the per-repo budget you set — bisect stops and asks before exceeding it.
// no surprise invoices: hard caps, not soft warnings
| meter | team | enterprise |
|---|---|---|
| included replays | 500 / seat / mo | pooled, custom |
| overage | $0.04 / replay | committed-use rates |
| median hunt | 5 replays ≈ $0.20 | runs on your compute |
| budget caps | per repo, hard stop | per repo + org-wide |
// faq — man billing
? What counts as a seat
Anyone who starts hunts or reviews bisect's draft PRs in a given month. Read-only viewers of verdicts in Slack are free, always.
? Can bisect hunt in monorepos
Yes — the ownership-graph pruner was built in a 40k-commit monorepo. Path-scoped hunts ("only under services/billing") are a flag: --paths.
? What happens when I hit my replay budget
The hunt pauses at the cap and posts its best current suspect with a confidence score. You can raise the cap for that hunt or accept the partial verdict. Nothing bills past the cap.
? Do you train models on our code
No, contractually, on every tier including free. Sandboxes are destroyed after each hunt; the only persistent artifact is the verdict and the diff you already merged.
? Can we self-host on the Team plan
The replay fleet is hosted on Team. Self-hosting the fleet, the sentinel, and the model gateway is the Enterprise tier — that's most of what you're buying there.
> Fourteen days, no card.
Point the sentinel at one noisy repo and see what it files by Friday.