fix: enable handling video_url in litellm and chat completions models

[seokhyunan]

Summary

• add video_url support to the chat-completions converter in openai/openai-agents-python
• allow OpenRouter-compatible video inputs to pass through the standard Agents SDK message conversion path

Problem

The SDK currently handles text, image, audio, and file content parts in the chat-completions converter, but it rejects OpenRouter-style video inputs shaped like:

{
    "type": "video_url",
    "video_url": {
        "url": data_url,
    },
}

Change

Updated src/agents/models/chatcmpl_converter.py so that:

• raw dict multimodal content items can pass through the converter
• Converter.extract_all_content() recognizes video_url
• the converter forwards video_url content instead of rejecting it

Verification

• uv run python -c "from agents.models.chatcmpl_converter import Converter; print(Converter.extract_all_content([{'type': 'input_text', 'text': 'x'}, {'type': 'video_url', 'video_url': {'url': 'data:video/mp4;base64,abc'}}]))"

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a9dee6b7f6

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: fbc09d1a86

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".

[seratch]

I took another pass on the latest head (4fef52e4) and re-checked it with a live runtime verification.

The previous tool-output regression looks fixed in this revision, but I still see one remaining blocker:

• src/agents/models/chatcmpl_converter.py now accepts video_url in the user-message path, and OpenAIChatCompletionsModel always sends that converted payload through the default Chat Completions path (src/agents/models/openai_chatcompletions.py). On official OpenAI, that still does not work.
• I verified this through the SDK/runtime path, not just by reading the converter: a normal image input succeeds, but a video_url input returns a backend 400 invalid_value from Chat Completions. The error says the supported content-part types are text, image_url, input_audio, refusal, audio, and file.
• That means this PR currently broadens the default official-OpenAI behavior and turns what used to be a local validation failure into a backend failure.

Given that, I do not think this is ready as an unconditional change for the default OpenAI Chat Completions path. If the goal is support for non-OpenAI OpenAI-compatible providers, I think this needs provider/base-URL gating so the official OpenAI path still rejects video_url locally. Also, if LiteLLM adapter layer (or LiteLLM itself) is the right place to enhance for this need, it'd be a more natural way.

[seokhyunan]

@codex review again

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

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".

[seratch]

Thanks for working on this. Now I see the value for developers who want to use video input supported models on openrouter. Having some unit tests would be ideal, but it's okay to skip them this time