migrate remaining providers to pydantic-ai

[Light2Dark]

📝 Summary

Fixes #7428,
Fixes #6369,
Fixes #6228,
Fixes #7036.
Should fix #5340
Should fix #7040

Also supports other providers (non OpenAI compatible like Mistral), but they may require updating config.

I did not test with Azure, and my config is quite simple. Removes old chat_messages from types.

Todo:

• support custom providers & custom models

OpenAI

RIP $$

Deepseek

Mistral

inline completion

🔍 Description of Changes

📋 Checklist

• I have read the contributor guidelines.
• For large changes, or changes that affect the public API: this change was discussed or approved through an issue, on Discord, or the community discussions (Please provide a link if applicable).
• I have added tests for the changes made.
• I have run the code and verified that it works as expected.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Review	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jan 9, 2026 9:12am

[Light2Dark]

removed timeout because imo we can let users cancel their own requests

[Copilot]

Pull request overview

This pull request migrates the remaining AI providers to use pydantic-ai, unifying the AI provider infrastructure and fixing several reported issues with provider compatibility. The migration simplifies the codebase by replacing custom provider implementations with pydantic-ai's standardized approach.

Key Changes:

• Migrated OpenAI, Azure OpenAI, and custom providers to use pydantic-ai instead of direct SDK calls
• Introduced a new CustomProvider class to handle OpenAI-compatible and other custom providers
• Removed legacy provider abstractions like CompletionProvider and provider-specific conversion functions
• Updated dependencies to include pydantic-ai with OpenAI support in the recommended and dev extras

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
pyproject.toml	Updated dependencies to replace standalone openai>=1.55.3 with pydantic-ai-slim[openai]>=1.39.0 in recommended and dev extras
pixi.lock	Updated lock file to reflect dependency changes with new SHA and editable flag
tests/_server/api/endpoints/test_ai.py	Refactored tests to mock pydantic-ai provider methods instead of OpenAI SDK directly; simplified test structure
tests/_server/ai/test_providers.py	Removed tests for legacy provider implementations; added tests for new provider capabilities (extended thinking, responses API support)
tests/_ai/test_chat_convert.py	Removed tests for deprecated conversion functions that are now handled by pydantic-ai; updated ChatMessage initialization to use empty list for parts
marimo/_server/ai/providers.py	Major refactor: introduced PydanticProvider base class, OpenAIClientMixin, and CustomProvider for flexible provider support; removed legacy CompletionProvider abstraction
marimo/_server/api/endpoints/ai.py	Simplified endpoint handlers by removing provider type checks and using unified pydantic-ai interface
marimo/_ai/_convert.py	Removed provider-specific tool conversion functions (OpenAI, Anthropic, Google) as pydantic-ai handles these internally

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.