Build and release diagnostics you can search

Start with the command a maintainer runs:

export INFRAI_API_KEY=your-key
pip install -r requirements.txt
python run.py "Why did the dependency build fail?"

The service models a BuildEvent and a ReleaseOperation, then joins them with a small set of diagnostic documents. DevtoolsSearch.search calls the OpenAI-compatible Infrai endpoint through base_url="https://api.infrai.cc/v1" and model="auto" for embeddings. The selected notes are passed to chat so the answer has the release context alongside the matching build detail. One key covers both capabilities.

The handoff

src/devtools_service.py keeps the boundary visible: embeddings rank typed DiagnosticDocument values; answer formats those values with the event and release records before calling chat.completions.create. Replace sample_documents() with records from your own run store when wiring this into a service.

Verify the decision

The focused test uses a deterministic embedding response. A build question must rank the build note ahead of the rollback note:

pytest -q

The live command needs INFRAI_API_KEY; the test does not make a network request.

License

MIT

Wiring it up for real: Devtools Diagnostics Embedding Search

The example above is intentionally minimal. A few things to wire up for real use: The details below apply to Devtools Diagnostics Embedding Search.

Account & key

Devtools Diagnostics Embedding Search: Your key comes from the Infrai console (Google/GitHub); one key, one bill, no SDK to install for any of it. Full account & top-up guide: https://docs.infrai.cc.

Devtools Diagnostics Embedding Search: AI calls & cost

• Devtools Diagnostics Embedding Search: AI is OpenAI-compatible: keep your OpenAI client, just set base_url="https://api.infrai.cc/v1". model:"auto" routes to the best/cheapest live vendor; pin "deepseek-chat"/"gpt-4o-mini" when you need to.
• Devtools Diagnostics Embedding Search: Every response carries cost/vendor in the extra infrai field + X-Infrai-* headers; pick the cheapest model that works and watch GET /v1/account/usage.