AI code review for pull requests.
- AI code review — line-level findings with committable suggestions and a per-file summary, posted directly on the PR diff
- Multi-provider — supports GitHub and Azure DevOps
- PR chat — reply to a finding or mention
@aporiaanywhere on the PR for a follow-up conversation - Review skills — on-demand domain knowledge (security, framework patterns, etc.) loaded by the reviewer when relevant
- Code graph — indexes repo structure (classes, methods, dependencies) so the reviewer understands cross-file context
- Multiple review engines — pluggable strategies: direct API or GitHub Copilot (Claude Code CLI planned)
Prerequisites: .NET 10, Azure Functions Core Tools, Azurite (or Azure Storage).
-
Copy settings and fill in your credentials (AI keys, provider PAT/App, Cosmos):
cp Aporia/local.settings.example.json Aporia/local.settings.json
-
Start the function host:
cd Aporia && func start
See docs/setup.md for webhook tunnels and full deployment, and docs/observability.md for local telemetry.
Aporia only reviews registered repositories. Unregistered repos are silently ignored. To enroll a
repo, register it via POST /api/manage/repos (requires a function key), then create service
hooks for PR events. Optionally drop a .aporia.json in the repo root to customize review behavior
(see Configuration).
Drop a .aporia.json file in the root of your repository to customize review behavior. All fields
are optional; sensible defaults apply when omitted.
See .aporia.example.json for all available options.
Aporia ships with built-in review skills: domain-specific knowledge the reviewer loads on demand when a PR touches a relevant area.
Each skill is a SKILL.md file under Aporia/Skills/ containing domain-specific review criteria.