docs: add documentation for OpenAI model support in ADK agents

[hanorik]

Added documentation for the experimental OpenAI model support in ADK Go.

[netlify]

✅ Deploy Preview for adk-docs-preview ready!

Name	Link
🔨 Latest commit	a3de53d
🔍 Latest deploy log	https://app.netlify.com/projects/adk-docs-preview/deploys/6a679caa7a8ef500089cc5db
😎 Deploy Preview	https://deploy-preview-2000--adk-docs-preview.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[karolpiotrowicz]

A few suggestions to round this out:

1. Link the runnable example. examples/openai/ is a great resource and linking repo examples is the norm on other pages. Suggest adding to the end of "Get started":

For a complete, runnable sample, see [examples/openai/](https://github.com/google/adk-go/tree/main/examples/openai) in the ADK Go repository.

2. Add a short "Supported features" and "Limitations" section. For an experimental, text-focused integration, it helps to set expectations up front:

## Supported features

- Text generation (streaming and non-streaming)
- Function (tool) calling
- Structured output via `OutputSchema` (JSON schema)
- Reasoning models (e.g. o-series), including reasoning-token accounting
- Token logprobs

## Limitations

- **Text only** — multimodal input (images, audio, files) is not supported.
- **Function tools only** — built-in tools (Google Search, code execution, etc.) are not supported.
- **Structured output uses OpenAI strict mode** — every field declared in an `OutputSchema` is treated as required.
- Some `GenerateContentConfig` options return an error rather than being silently ignored: `TopK`, stop sequences, multiple candidates, frequency/presence penalties, request labels, and safety settings.

3. Admonition type. !!! example "Experimental" renders as a green "Example" box; a caveat reads better (and stands out more) as a warning:

!!! warning "Experimental"

4. Minor polish.

• Heading case: "Configuration Options" → "Configuration options" (the other headings use sentence case).
• The snippet uses ctx and log without showing where they come from — worth a one-line note (e.g. "assuming a context.Context named ctx") or including the context/log imports so it's copy-paste-ready.
• There's a stray tab on the blank line after the model instantiation.

With the example link and the capabilities/limitations added, this will be a really solid page. Thanks again!

[karolpiotrowicz]

One more suggestion, on how this page relates to LiteLLM. Since it sits right next to the LiteLLM page — and ADK Python reaches OpenAI through LiteLLM — a reader landing here could read "OpenAI = experimental, Go-only" and miss that OpenAI is fully usable from Python today via LiteLLM.

The Claude page already handles this nicely by making the intro language-aware. Suggest doing the same here — replace the single-line intro with:

You can use OpenAI models (such as GPT-4o and GPT-4o Mini) with ADK. How you connect depends on the language:

- **Go — native support (this page):** ADK Go provides a direct `openaimodel` package that implements the `model.LLM` interface, targeting the OpenAI Responses API. It is currently experimental.
- **Python — via LiteLLM:** ADK Python accesses OpenAI models (and many other providers) through the LiteLLM connector. See [LiteLLM](/agents/models/litellm/).

The rest of this page covers the Go-native integration.

This keeps the page focused on the Go-native model while making it clear, up front, that Python users have a stable path via LiteLLM — so no one mistakes "experimental, Go" for "OpenAI isn't well supported."

(Optional, separate follow-up: a one-liner on the LiteLLM page's OpenAI mention pointing Go users back to this native page, for symmetry.)

[hanorik]

@karolpiotrowicz thanks for the comments! I've added suggested sections and text adjustments except the Example box. Actually it renders as a purple box and seems informative enough for the purpose. This box is also used in the Model routing article so to keep it more consistent I'd prefer to leave it as it is.