ProbeRun Cloud remembers every gate. Open source is how you try it.
The command-line tool, its libraries and the GitHub Action are open source under Apache 2.0. The self-hosted server and dashboard are source-available under FSL-1.1-ALv2, and each version becomes Apache 2.0 two years after its release. ProbeRun Cloud is the hosted version for teams.
ProbeRun Cloud
From $29 a month The first 10 teams pay a founding price for their first 12 months. Team, Scale and Enterprise start with a conversation; Hobby opens with self-serve checkout.
Everything in self-hosted, plus:
- A hosted workspace for your team, run by us.
- Drift history: each night's sentinel result, remembered as a trend on the project.
- Audit export, and audit retention set per team.
- Drift monitoring on hosted runners, so nobody has to schedule it.
- Run your scenarios against your agent from the dashboard, remember a gate, and trigger runs from your deploy, without the CLI.
- A review queue: the runs your team should label next, picked by hand or sampled at random.
- Gate history: ProbeRun Cloud remembers every gate it runs, so you can see the commit where a scenario's verdict or tool path first changed, and which scenarios passed and failed on the same commit.
Self-hosted
$0 The CLI under Apache 2.0, the server and dashboard under FSL-1.1-ALv2. Yours to run and change.
This is the free tier: run it yourself, with no usage limits and no time limit.
- Capture, import and replay your agent's runs.
- Compare two runs on verdict flips and path drift.
- A CI check that fails when a verdict flips:
probe gate. - Every gate states what it can catch.
- Spend caps, and a cost estimate before any live run.
- The drift sentinel, on a schedule you run.
- The self-hosted server and dashboard: login through your identity provider, roles, an audit log, the trajectory diff.
How Cloud plans work
| Plan | Hobby | Team | Scale | Enterprise |
|---|---|---|---|---|
| Price | $29 a month | $199 a month | $599 a month | From $2.5k a month, billed annually |
| Founding price, first 12 months | No | $149 a month | $449 a month | By agreement |
| Projects | 1 | 3, more at $79 a month each | 10, more at $79 a month each | By contract |
| Monitored scenarios | None | 10, checked daily; more at $49 a month per 10 | 50, as often as every 6 hours; more at $49 a month per 10 | By contract |
| Run history kept | 30 days | 365 days: a year of remembered gates | 3 years | By contract |
| Runs | 10,000 live runs a month, then live runs stop until next month; replays keep working; never an overage bill | Fair use: 250,000 a month | Fair use: 1,000,000 a month | By contract |
| Runs ProbeRun starts for you | 1,000 a month, one at a time | 5,000 a month, three at a time | 25,000 a month, eight at a time | By contract |
| Seats | 1 | Not counted | Not counted | Not counted |
| Remembered gates and baseline by branch | Yes | Yes | Yes | Yes |
| SAML single sign-on | No | No | Included | Yes |
| Support | Community support on GitHub; bug fixes for everyone | First reply within 2 business days | First reply within 1 business day | In the contract |
Try it with the self-hosted edition first; Cloud is billed monthly and you can cancel any time. Seats are not counted on Team and above. Hobby is for one engineer: need a second project, a teammate, history beyond 30 days or nightly monitoring? That is Team. What each limit means, and what happens at it:
- Runs
- Runs are never billed per run. On Hobby, live runs past 10,000 in a month stop until the next month while replays keep working, so there is never a surprise bill. Team and above have a monthly fair-use line. Past 80% and again past 100% your team's admins get an email, and at 100% we start a conversation about the right plan. A run in CI is never slowed, dropped or failed by it. Runs ProbeRun starts for you, from the dashboard or a deploy hook, have their own monthly count and a number that can run at once; past either, a new one is refused before anything runs. Runs you push from CI, replays and monitoring never count against it.
- Projects
- A set number per plan. At the limit, a new project is refused with the reason and what to do; every existing project keeps working, including after a plan change.
- Drift monitoring
- Each plan includes a number of monitored scenarios, each counted once however many live runs it gets per check, and a shortest interval between checks. Larger plans check more often; extra scenarios come in packs of 10, which never shorten the interval.
- Support
- By email, business hours US Eastern, with a written first-reply target: 2 business days on Team, 1 business day on Scale. Targets are best effort and carry no service credit. Enterprise targets are set in the contract. Our bugs, outages and how-to questions are always included. Custom work, such as writing your scenarios or building an integration, is $250 an hour or $2,000 a day, or a 20-hour block for $4,500 used within 12 months. Work that has to start within 2 business days costs 50% more.
Model credits
Bring your own model key and ProbeRun never charges for model calls. If you would rather ProbeRun run the model for you, Cloud sells prepaid credits: $25 for 2,500, $100 for 10,500, or $500 for 55,000. A run uses 1.5 times the model's list price, at 100 credits a dollar, rounded up to a tenth of a credit. A run that fails because of ProbeRun costs nothing.
| Model | Credits per million input tokens | Credits per million output tokens |
|---|---|---|
| Claude Fable 5.1 / 5 | 1,500 | 7,500 |
| Claude Opus 5.5 | 600 | 3,000 |
| Claude Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | 750 | 3,750 |
| Claude Sonnet 5.5 / 5 | 300 | 1,500 |
| Claude Sonnet 4.6 / 4.5 | 450 | 2,250 |
| Claude Haiku 4.5 | 150 | 750 |
| GPT-4o | 375 | 1,500 |
| GPT-4.1 | 300 | 1,200 |
| GPT-4.1 mini | 60 | 240 |
| GPT-4o mini | 22.5 | 90 |
| GPT-5.4 mini, ProbeRun's judge | 123.75 | 742.5 |
List prices come from the model vendors' public price pages: Anthropic's checked 2026-10-16, OpenAI's 2026-08-29. Models a vendor has retired are not listed; Claude Sonnet 4.5 retires on 2026-11-30. ProbeRun's judge runs GPT-5.4 mini on Azure OpenAI, so its row uses Azure's list price for that deployment, checked 2026-10-12. A price change is announced 30 days ahead and never changes credits you already bought; a vendor price cut is passed on. Credits do not expire. A purchase can be refunded within 24 hours if none of its credits were used; otherwise credits are not refundable. Auto-recharge is off unless you turn it on, with a monthly cap you choose and a receipt for every charge.
Your runs and results export as JSON from the CLI and the API in every edition, so leaving costs you nothing. Need it on your own infrastructure, with support? Email hello@proberun.com.
Start with a conversation.
Tell us about your agent and we set up your workspace with you, starting with a baseline recorded together. The first 10 teams get a founding price of $149 or $449 a month for their first 12 months. In return, they pay the year up front or agree to a case study we can publish. Teams that do neither pay the list price.
We reply by email.