fix: route Azure Astra through Responses API

[usnavy13]

Summary

Azure OpenAI agents using gpt-6-astra fail with a 400 when they call a function tool unless the user manually enables the Responses API. Astra support in #15567 limited automatic routing and SDK request constraints to canonical OpenAI endpoints. Azure now receives the same Responses default, with the correct URL and deployment name, and compaction resolves the configured summary model's deployment and credentials.

Fixes #15844.

How it works

Recognize direct Azure OpenAI configurations and canonical Azure hosts, resolving URL placeholders before checking the host. Normalize bare resource URLs to the API path, recover deployments specified only in a URL, and move query parameters into the SDK's query options so it can append /responses correctly. Explicit API opt-outs, custom gateway routing, and Azure's fallback model selection remain intact.

Azure configuration: gpt-6-astra -> production-deployment
  getOpenAILLMConfig
    useResponsesApi: true -> max_output_tokens
    model: gpt-6-astra -> existing SDK request constraints
    modelKwargs.model: production-deployment -> Azure request body
  getOpenAIConfig -> /openai/v1/responses + api-key header

Compaction: gpt-4.1-mini -> summary-production
  shapeSummarizationConfig
    mapModelToAzureConfig -> summary group's resource, key and deployment
    getOpenAIConfig -> target model's API and request parameters
  SDK summarizer -> summary-production

The logical Astra model remains available to the SDK for parameter validation and encrypted reasoning, while the request carries Azure's deployment name. A different summary model resolves through Azure's existing model/group maps; self-summarization keeps the agent's resolved options. No SDK dependency update or new configuration field is required.

Azure's support table lists Astra function calling as Responses-only. Chat Completions remains available without tools; tools are attached after configuration, so the default is chosen at configuration time.

Change Type

• Bug fix

Testing

• 516 tests passed across nine suites: OpenAI endpoint configuration and initialization, Azure utilities, SDK requests, and summarization.
• Request tests use the real agents SDK with an HTTP substitute and assert the final URL, query parameters, API key header, deployment, token limit, tool schema, and Astra parameter constraints. Coverage includes bare origins, trailing slashes, URL templates, deployments specified only in the URL, and the v1 model-name fallback.
• Compaction tests execute the real SDK graph with an HTTP substitute. They verify a separate deployment on the same resource, a different resource/key, and a versioned Astra summary model, while keeping the main agent's configuration intact.
• npx tsc --noEmit passed in packages/api; packages/data-provider also passed for the original comment-only change. ESLint, formatting, import sorting, and pre-commit static checks passed.
• Manual Azure smoke test of automatic routing: an agent with calculator, reasoning_effort: low, and no useResponsesApi setting returned 4 for 2 + 2; after reloading, it called the calculator again and returned 12. The subsequent URL and compaction changes are covered by the SDK tests above.
• npm run lighthouse completed the production build, but the local audit failed during Chrome's temporary-profile cleanup with Windows EPERM. Lighthouse CI passed on the previous reviewed commit; the updated commit needs its own CI result.

Checklist

• Code follows the project's style guidelines.
• Self-review completed.
• Regression tests cover the reported behavior.
• Focused local tests and workspace typechecks pass.

[danny-avila]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-12T01:38:42.907772Z	4237cd7	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4237cd72ab

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

To use Codex here, create a Codex account and connect to github.