Frequently Asked Questions

General

What is Revue?

Revue is a multi-agent AI code reviewer with two modes:

Claude Code skill (primary): Install /revue in Claude Code and run it on your staged diff before you commit. Six specialist agents review your code inline, before it ever reaches CI.

CI pipeline (secondary): Add Revue to your GitHub Actions, GitLab CI, or Bitbucket Pipelines to automatically review every pull request or merge request.

What platforms does Revue support?

  • Claude Code (primary): via the /revue skill, installed with one command
  • GitHub: GitHub Actions
  • GitLab: GitLab CI/CD
  • Bitbucket: Bitbucket Pipelines

Does Revue read my code?

No. Revue never reads your code. Your code diff goes directly to the AI provider you configure, using your own API key. Revue's servers only handle license validation and usage counts. No source code is ever transmitted to Revue.

Diff limits

What is the diff limit?

By default, Revue stops reviewing if the diff exceeds 2,000 lines and suggests breaking the PR into smaller pieces. This keeps AI costs predictable and reviews fast.

You can raise the limit in .revue.yml:

review:
  max_diff_lines: 4000

What counts toward the diff limit?

Every line in the unified diff: context lines, additions, and deletions across all changed files.

What happens when the limit is exceeded?

Revue posts a summary comment explaining that the diff is too large and suggesting a breakdown strategy. No agent reviews are run.

BYOK (Bring Your Own Key)

Do I need to provide an AI API key?

Yes. Revue uses your own AI provider key; it does not have a shared AI backend. This means: - Your code never leaves your infrastructure - You choose your AI provider (OpenRouter, Anthropic, OpenAI, Azure, etc.) - You control costs directly

Which AI providers are supported?

  • OpenRouter: default (DeepSeek-V4-Pro, cost-optimised)
  • Anthropic (Claude)
  • OpenAI (GPT-4o, etc.)
  • Azure OpenAI
  • Custom / corporate gateway (any OpenAI-compatible endpoint)

How do I configure my AI provider?

Set the API key as a CI secret and reference it in .revue.yml:

ai:
  provider: openrouter              # default, most cost-efficient
  model: deepseek/deepseek-v4-pro
  api_key_env: OPENROUTER_API_KEY   # name of your CI secret

To use Anthropic instead, set provider: anthropic, model: claude-sonnet-4-5, api_key_env: ANTHROPIC_API_KEY.

Blocking behaviour

Can Revue block a PR from merging?

Yes. Set fail_on_critical: true in your CI configuration and Revue will exit with a non-zero code when critical findings are found, causing the pipeline to fail.

GitHub Actions:

- uses: revue-io/action@v1
  with:
    fail_on_critical: "true"
    ai_api_key: ${{ secrets.OPENROUTER_API_KEY }}

GitLab CI:

variables:
  REVUE_FAIL_ON_CRITICAL: "true"

What severity levels does Revue use?

Level Description
critical Must fix before merge: security vulnerabilities, data loss risks
high Should fix: significant bugs or performance issues
medium Consider fixing: code quality, maintainability
low Minor suggestions: style, naming, minor improvements

Fix suggestions

How do fix suggestions work?

After the specialist reviewers surface findings, a resolver step evaluates each one and, where it can generate a safe scoped fix, posts a platform-native code suggestion (1-click accept in GitHub/GitLab/Bitbucket).

How do I control the confidence threshold?

Set min_confidence in .revue.yml:

review:
  min_confidence: 70  # 0-100, default 70

Only suggestions above this confidence score are posted. Setting it higher (e.g. 85) means fewer but higher-quality suggestions.

Can I disable fix suggestions?

Yes. Use team-quick or any team that does not include the resolver step. Alternatively, set min_confidence: 100 to suppress all suggestions.

Free tier

How many reviews do I get on the Free tier?

10 reviews per month, and every review runs the full agent panel (the same specialists paid plans use). A "review" is one diff processed by the review pipeline, whether run locally via /revue or in CI.

What happens when I hit the limit?

Revue returns a 429 response and posts a comment on the PR explaining that the limit has been reached and linking to the upgrade page.

Does the counter reset monthly?

Yes, on the first day of each calendar month.

Troubleshooting

The review ran but posted no comments

  • Check that your CI runner has the pull-requests: write permission (GitHub Actions).
  • Verify that the GITHUB_TOKEN / CI_JOB_TOKEN is being passed correctly.
  • Check the pipeline log for errors from the revue review step.
  • The diff may have been empty or all files matched ignore_patterns.

"License key is invalid"

  • Copy your key fresh from the dashboard: keys are long and easy to truncate.
  • Check for leading/trailing whitespace in the CI secret value.
  • If the key was recently regenerated, update the secret.

The review is very slow

  • Increase agent_timeout_seconds to 120 in .revue.yml.
  • Use team-quick for rapid feedback on small changes.

I'm getting "diff too large" on every PR

  • Increase max_diff_lines in .revue.yml.
  • Add large generated files to ignore_patterns (e.g. *.lock, migrations/*).
← .revue.yml Reference