Setting Up Code Review
Connect your source control provider
Go to Code Review in the Entelligence dashboard and connect your provider:
Install the EntelligenceAI GitHub App on the org or specific repos you want reviewed. The app requests: Pull requests (read/write — to post reviews and comments), Contents (read — to fetch diffs), Checks (read/write — to post a review status check), Issues (read/write — so @mention commands work), and Metadata (read). If you see a 403 related to Checks after install, re-approve the app with the Checks permission granted.
Enable the repos you want reviewed
In the onboarding wizard, toggle on each repo you want Entelligence to review. You can enable more repos later from Code Review → Codebases.
Configure review settings
For each repo, set:
- Priority level — how much gets flagged:
- Critical — only the most severe, high-confidence issues (highest bar, least noise)
- Medium — medium-priority issues and above (a reasonable default for most teams)
- Low — everything, including minor/nitpick-level findings
- Guideline docs (optional) — upload a style guide or best-practices doc (
.txt,.md,.doc,.docx, up to 10MB) and assign it to one or more repos. Review comments will reflect what's in the doc, not just generic rules. You can also have Entelligence generate a guideline automatically by mining your repo's existing PR review-comment history, instead of writing one from scratch. - Additional per-repo toggles available from the same settings screen: review on every push vs. only on open, auto-resolve threads when you reply, show/hide the confidence score, cross-repo/cross-org learnings, inline diagram generation, and a max-comments-per-PR cap.
You're done
Open a new pull request on any enabled repo — Entelligence reviews it automatically within moments of it being opened, and again on every subsequent push.
Every push to an open PR triggers a re-review (unless you've turned that off in settings) — new commits get fresh findings, and previously-resolved issues won't resurface as long as the underlying code didn't regress. Draft PRs are skipped automatically until marked ready for review.
Talking to the bot in a PR
Mention the bot (@entelligence / @entelligenceai) in any PR comment:
| Say this | It does this |
|---|---|
@entelligence review (or "re-run review", "retrigger review") | Runs a fresh full review |
@entelligence summary | Regenerates just the PR summary comment |
@entelligence resolve_all (or "resolve all") | Resolves every open bot-created thread |
@entelligence config | Posts the bot's current configuration for this repo |
@entelligence link <PR URL or org/repo#123> | Links a related PR in another repo for cross-repo context |
| Any other question | Answered directly using the PR's context |
You don't need to @mention the bot when replying directly to one of its own comments — it treats a reply as addressed to it automatically.
Two shorthand commands also work directly in a reply thread: /updateGuideline and /addGuideline to adjust the guideline doc from inside a PR conversation, and /updateCommit to have the bot re-scan after you've pushed a fix.
Tuning the bot over time
- Thumbs down a comment that isn't useful — Entelligence infers a "learning" from this and reduces similar false positives on that repo going forward.
- Reply to a finding to explain context; the bot factors it into its next comment on that thread.
- Update guideline docs anytime from Code Review → Codebases — changes apply to the next review.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Bot doesn't comment on a new PR | App/integration not enabled for that specific repo, or PR opened as a draft | Check Code Review → Codebases, confirm the repo is enabled; open the PR as ready-for-review (or wait until it's marked ready) |
| Bot stopped commenting after previously working | GitHub App permissions were revoked or scope reduced (common after an org permissions audit) | Reinstall/re-approve the GitHub App with the permissions listed above |
| Findings feel too noisy | Priority level set to Low | Raise the priority level to Medium or Critical in Codebase settings |
| Same false positive keeps coming back | The bot hasn't seen a 👎 or clarifying reply on it yet | Thumbs-down the comment or reply with context — this is what feeds the learning loop |
| Bitbucket bot has no write access | The one-time admin-authorization step during setup wasn't completed | Re-run setup and complete the "grant write access" step with an account that has admin rights on the workspace |
Next steps
- Want to check code before it even becomes a PR? See Reviewing from the CLI.
- Curious how confidence scores and findings feed into your team's metrics? See Team Insights.