Investigations in action
All investigationsRepository activity
Investigations
Select an investigation
Start with a GitHub issue or open a saved report.
Maintainer response
UnreviewedApproval is saved locally. Publish the response manually on GitHub.
New investigation
Usage
Request activity
Token counts come from the model provider. Historical runs may not have request counters.
Connections
Local configuration
# api/.env GITHUB_TOKEN=your_github_token OPENAI_API_KEY=your_model_key OPENAI_MODEL=gpt-4.1-mini
Configure credentials on the backend and restart the server. Issue text and code excerpts are sent to your configured model provider in AI mode.
Documentation
Investigate a GitHub issue
Open New Investigation and paste a public issue URL. The investigator reads its description and discussion, then searches related issues and repository code.
Review the evidence
Findings link to observed sources. Code excerpts point to a fixed commit and line range. Confidence describes an assessment; it does not establish a reproduced root cause.
Approve a response
Edit the response, save changes, and approve the draft. Copy the approved response to GitHub. Editing it requires approval again.
Investigation tools
- search_issues
- Find related reports in the same repository.
- search_paths
- Find candidate files using path keywords.
- search_code
- Search up to eight ranked source files plus cached files. Coverage is recorded.
- read_file
- Read up to 100 lines from a known file.
Boundaries
Public repositories only. Repository code is read, never executed. The snapshot is current default-branch code. Retrieval mode collects evidence without AI analysis.
Each run allows 36 GitHub requests, 10 model turns, 18 tool operations, and six minutes. Interrupted runs retain collected evidence.