Use case
Code review on a GitHub repository with several AI models
Milly Lab reviews code in a connected GitHub repository with Swarm Mind, part of the Max ($42/month) and Advanced ($72/month) plans. The models read the files they need, write up what they find and stage any fixes; nothing reaches GitHub until you press Apply and choose the default branch, a new branch or a new repository.
Last updated:
What does an AI code review in Milly Lab do?
It gives a repository a second reader: AI models list, search and read its files, report problems in the form you ask for, and can propose fixes as staged file changes that you review before anything is committed.
How do you run a code review on a repository?
Start a Project or Council session in Swarm Mind, connect GitHub from the composer, pick the repository and describe the review; the run reads the code, reports, and stages fixes for you to apply or discard.
- Open Swarm Mind and choose Project Mode for one structured review with a report, or Council Mode for several independent reviews of the same code that you can compare.
- In the composer, open the attachment menu, choose GitHub, approve access in the window GitHub opens, and pick the repository from the list of repositories your account can reach.
- Write the brief: which folder or module to review, what to look for (security, errors, readability), the format of the report, and where fixes should go, for example on a new branch.
- Follow the run in the activity log as the models list folders and read files; each file read returns at most 48 KB.
- Read the report. Proposed fixes appear in the changes panel with each file's path and whether it is created, edited or deleted; nothing has touched the repository yet.
- Press Apply and choose the default branch, a new branch or a new repository, or press Discard. A new branch leaves the default branch untouched, so you can open a pull request on GitHub and review the change there.
What should a code review request say?
A good request names the part of the repository, the kind of problems to look for, the shape of the report and where any fixes should go.
Review the authentication module in the connected repository. List security weaknesses with a severity for each, and propose fixes for the three most serious. Put the fixes on a new branch.
This request has stages that depend on each other: review, then rank, then fix. That is the kind of work Project Mode splits between models; for a quick second opinion on one file, a normal chat costs less.
Which plan includes code review on a repository?
Code review on a connected repository runs in Swarm Mind, which is included in Max ($42 per month), Advanced ($72 per month) and Enterprise; Free, Starter and Pro do not include it.
| Model | Max ($42 per month) | Advanced ($72 per month) |
|---|---|---|
| GPT-5.6 Terra | 285 | 394 |
| GPT-5.6 Sol | 68 | 102 |
| Claude Sonnet 5 | 265 | 440 |
| DeepSeek V4 Pro | 666 | 1,083 |
| Gemini 3.7 Flash | 359 | 598 |
| GPT-5.6 Luna | 1,180 | 1,476 |
| Claude Haiku 4.5 | 270 | 337 |
| Claude Opus 5 | — | 114 |
| ChatGPT Codex | — | 226 |
Numbers are monthly allowances in messages (ChatGPT Codex: in sessions). Every model call in a run counts against that model's allowance, as a chat message does.
What are the limits of a repository review?
A run reads at most 48 KB of each file, so the end of a longer file is cut off, and staged changes are capped at 256 KB per file and 2 MB per session.
| Limit | Value |
|---|---|
| Read per file | 48 KB; longer files are truncated, not skipped |
| Entries per folder listing | 300 |
| Staged change per file | 256 KB |
| Staged changes per session | 2 MB |
| File tool rounds per model turn | 12 |
| Council and Project sessions running at once | 4 per account |
| Lock after an Apply that failed midway | Released after 2 minutes |
What can Milly Lab see in my GitHub account?
Milly Lab asks GitHub for the repo scope, which reaches every repository your GitHub account can access; Milly Lab's own code confines a session to the repository you chose and rejects file paths outside it.
- If that is broader than you want, connect a GitHub account that can reach only the repositories you plan to review.
- Files a run reads are sent to the AI model providers as context. Do not connect repositories that hold passwords, API keys or personal data you may not share.
- Reading files counts toward your allowance, because file contents are sent to the models as input; a large repository makes a run cost more.
- Disconnecting the workspace in Milly Lab removes the stored connection; to withdraw the grant completely, also revoke Milly Lab in your GitHub settings.