bugs squashed 0
Private preview · GitHub teams

AI writes fast. Rolt reads faster.

Every finding points at a line. Clean pull requests stay quiet.

How it works

Understand, inspect, decide.

  1. 01

    Orient

    Head, size, CI, riskiest chapter first.

  2. 02

    Read in order

    Callers, tests and config become chapters.

  3. 03

    Open the evidence

    Every claim links to a line or a check.

  4. 04

    Decide

    Unproven stays visible. Never a score.

Calibrated language

A citation says where, not whether.

Observed1 / 4

Computed fact

Read from code, git, GitHub or CI.

Possible2 / 4

Static inference

Deterministic, possibly incomplete.

Verify3 / 4

Model finding

A falsifiable reading of the evidence.

Confirmed4 / 4

Outcome

A decision, a check, or a later change.

One engine, two routes

Free on your machine. Paid for the team.

01 · Free localPreview pending

Review before you push

Any PR or local change as a Story in your browser.

  • Works without a model
  • State stays on your machine
  • Bring your own Claude Code or Codex

Install command coming with the packaged CLI.

02 · Team gatePrivate preview

Review every enabled pull request

A GitHub App reviews every head and links to a private Story.

  • Hosted models on Rolt accounts
  • Per-repository controls
  • Credits only for completed reviews
Real product, public code

A Story from a live Node.js pull request.

nodejs/node #65761, recorded 3 September 2026.

Opennodejs/node #65761head 30748a2CI passed
Intent from PR

generatedSourceMapCache grows for every evaluated source with a unique source URL.

Computed fact2 source files · 1 generated file
lib/internal/source_map/source_map_cache.jsadded
1+while (cacheSize > sizeLimit) {
2+ deleteOldestSourceMap();
3+}

"Concern" is a deterministic signal, not proof of a defect. Node.js source under its license.

Pricing

Local is free. Teams pay per review.

Free · local
$0

Local Story reader. Your own Claude Code or Codex.

Team
$30 / seat / mo

Per active seat per month. 20 pooled credits per seat.

Retries and failures never cost a credit.Size bands and overage
Know the boundary

Local code stays local.

Local

Processed on your machine. Only GitHub and the agent you invoke see it.

Cloud

Authorized source goes to Rolt and an approved model. Never global training data.

Retention

Reviews and Stories are kept 30 days, then deleted.

The full data map

Questions

Straight answers.

Why Rolt when Copilot already reviews our PRs?

Rolt shows the evidence behind every finding and stays quiet when there is nothing to say. It learns from what your team confirms, whichever agent wrote the code.

What is a review credit?

One completed review of one head. Later heads review only the delta.

Does Rolt train on our code?

No. What it learns stays scoped to your organization, and you can inspect and delete it.

What does the Story get wrong?

A citation proves where a claim came from, not that it is true. Rolt labels the difference instead of hiding it behind a score.

Ship what the AI wrote. Not what it broke.

Nothing runs or charges until you enable a repository.