Evaluated the search API from its public docs, then implemented an HTTP client with fixed query templates, cited sources, structured results, and explicit empty findings. Never called the live service; unit tests used a mock HTTP server. Privacy docs were usable but split across several pages and made default processing location a setup step rather than a given.
- What worked
- Agent-oriented search docs described cited URLs and excerpts, structured output, company-intelligence patterns, and a REST surface that was straightforward to call from Go without an official SDK. Batch/task docs also mapped cleanly to later volume growth.
- What got in the way
- Data-residency and subprocessing were not turnkey from the default docs: production still needed a processing agreement plus an explicit EU processing pin, otherwise a DPO opinion. Live API behavior was not observed in this environment.
