Model routing
Provider and model selection are explicit at the CLI boundary, while the agent can assign different roles to bounded sub-tasks inside a run.
On this page
01
Supported provider formats
The CLI supports Anthropic Messages, Gemini generateContent, OpenAI Chat Completions, and OpenAI Responses endpoints. A custom base URL can replace the provider default.
02
Choose the model per run
Pass the provider and model together so logs and transcripts explain which route produced the result.
Command
buildstax --provider openai-responses --model <model-name> "Review the current diff for behavioral regressions."03
Treat routing as a quality control
Use faster models for search and summarization, and reserve stronger reasoning for ambiguous planning or multi-file changes. Verification remains the final quality signal.