02The Story
Merge AI code faster.
Rolt catches the slop and the bugs, like a staff engineer would.
Try it on any pull request: swap github.com for rolt.ai in its link.
01Benchmark
70% recall on Martian’s Code Review Bench. Thorough catches more of the real, human-verified issues than the other AI reviewers tested.*
- Rolt Thorough109/156
- 70%
- Copilot100/156
- 64%
- Cubic94/156
- 60%
*Our run of Martian’s open benchmark, issues matched by code check.
Section 1: Signal merging
- 1Restrict signal merging to root
Section 2: Extend signal override
- 2Replace inherited signal with replaceOption
- 3Remove inherited signal with undefined
03Pricing
Pay per review. Up to 5 authors, no seats. 4 in 5 reviews cost $1 to $2.
Up to 5 authors
$0a seat
Pay only for the reviews you run.
6+ authors
$30an author a month
40 credits each, shared by the team. Extra credits $1. Only authors Rolt reviewed that month count.
Enterprise
Customannual agreement
Volume pricing, SSO and a data processing agreement.
50 free credits for your first 14 days, in any mode. No card needed.
Estimate your month
$41a month
50 free credits for your first 14 days
Pay per review, about $21 an author.
Estimate from real review sizes, before tax and free credits.
Prices by pull request size
| Lines changed | ||||
|---|---|---|---|---|
| Mode | Up to 1,000 | 1,001–5,000 | 5,001–10,000 | 10,001–20,000 |
| Bugs onlyOnly real bugs and security issues | $1 | $2 | $5 | $10 |
| BalancedBugs, plus serious design issues | $1.50 | $3 | $7.50 | $15 |
| ThoroughThe most human-like review: bugs, design and nits like naming | $2 | $4 | $10 | $20 |
04Security
Your code, handled with care. What Rolt keeps, where, and for how long.
- No training on your code
- Never used to train global or cross-customer models.
- Isolated per review
- Each review runs in its own fresh environment, ended with it.
- No lingering copies
- Deleted a day after no open PR is at its commit, 30 days at most.
- Stored in the US
- Encrypted in transit and at rest, on AWS and Neon in Oregon.
- Zero data retention
- The AI models that read your code never store or log it.
- Delete anytime
- Delete your workspace or account in Rolt; a PR's or repo's history by email.
05Questions
Straight answers.
What counts as an active author?
Anyone whose pull request Rolt reviews that month, outside contributors included. Dependency bots like Dependabot don't count.
When do the free credits start?
When you start billing in setup, no card needed: 50 free credits for reviews in any mode over your first 14 days. Add a card when they run out.
Can we cap what a month costs?
Yes. Set a monthly spend limit; reviews past it don't start.
Which languages does Rolt support?
Rolt reads the code's structure in TypeScript and TSX, JavaScript, Python, Go, Java, Rust, C, C++, C#, Objective-C, Kotlin, Swift, Scala, Ruby, PHP, Elixir, Lua, Bash and CSS. Other languages are reviewed from the diff and the code around it.
Does Rolt work with GitLab or Bitbucket?
Not yet. Rolt is a GitHub App: install it on the repositories you choose, and it reviews their pull requests.
Do you train on our code?
No. Rolt does not use your private code for global or cross-customer model training.
Can I try Rolt for free first?
Yes: 50 free credits on your repositories, no card.