Hosted or local
- Hosted
- Any OpenAI-compatible or Anthropic model, with your own key.
- Local
- A local model through Ollama or vLLM keeps your code in your network.
Runs on your model.
Use a hosted provider with your own key, or a local model. Give the review a strong, a standard and a fast model, and a daily budget.
Three sizes
- StrongJudges what the reviewers found.
- StandardReviews the change and answers replies.
- FastReviews small changes.
Daily budgets
Budgets per organisation and project keep spend in check.
For example: Today: 120,000 of 2,000,000 tokens · 3 reviews
Reads like a colleague.
It reads more than the diff and comments on the lines your merge request changed.
Reads the whole repository
It opens any file it needs, not only the diff.
Comments on the lines
It starts threads on the changed lines, with suggested changes a person applies in one click.
Only what changed since
After a push, only new changes raise new findings. Fixed findings close and their threads resolve.
Follows your rules
It follows your repository's own AGENTS.md and REVIEW.md, taken from the target branch.
It talks back.
Reply in one of its threads and it answers. A member's reply can mark a finding won't fix or false positive, and a qualor review note asks for a review, drafts included.
Accounting exports a few dozen invoices a month. Can this wait for the next release?
I’d recommend fixing it before the next release: an export with 1–50 matching invoices contains no invoice rows, so accounting’s usual exports are likely affected. Changing the loop condition to page <= pages includes the final page; tests around the 50-invoice boundary would help prevent a recurrence.
qualor review
Reviewing this commit (4c7d2e91a0b3).
Better with your analyzers.
- No repeats. The reviewer is shown what your analyzers found, so it doesn't repeat it.
- One issue list. AI findings are Qualor issues, beside your analyzers' findings.
- Your agent fixes both. Agents read both kinds of finding through MCP.
- The gate stays yours. AI findings never count in the quality gate, and the reviewer never approves or commits.
Issues feature/csv-export
- HighData from the HTTP request reaches the text of an SQL query.
src/invoices/export.ts:9 - MediumAIThe final page of invoices is never exported
src/invoices/csv.ts:25
Quality gate failed: 1 new issueThe AI finding is not counted.
Switch it on in three steps.
Connect a model
A hosted provider with your own key, or a local model.
Switch on AI review
Pick the projects to review. It stays off until you do.
Scan your merge requests
Add Qualor's scanner to GitLab CI or GitHub Actions. Reviews then appear in your merge requests.
Questions
Which models does AI review work with?
Any OpenAI-compatible or Anthropic model, hosted with your own key, or a local model through Ollama or vLLM. You can give the review a strong, a standard and a fast model.
Does my code leave my network?
Qualor stores everything on your server. AI review sends the merge request, the files the model reads and the findings on them to the model you chose, with secrets masked; the assistant sends the lines of the issue you ask about. With a local model, your code never leaves your network.
What does AI review cost?
Nothing from us: it is in the free community edition. You pay your model provider, or nothing with a local model, and daily budgets keep spend in check.
Does it work with GitLab and GitHub?
Yes: GitLab.com and self-managed GitLab, GitHub.com and GitHub Enterprise Server, with your scans in GitLab CI or GitHub Actions.
Does it approve or reject merge requests?
No. It never approves, rejects or commits, and its findings never count in the quality gate. A person applies every suggestion.
Free in every edition.
AI code review is part of the MIT‑licensed community edition, with no lines‑of‑code licence.