support custom models returning vercel chunks

[Light2Dark]

📝 Summary

Supports returning pydantic_ai types / dicts, async or sync chat models.

# Supported yield types
yield {"type": text-start", id: "1"}
yield pydantic_ai.ui.vercel.response_types.TextStart(id="1")

Parking this approach here for discussion.

🔍 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).
• Tests have been added for the changes made.
• Documentation has been updated where applicable, including docstrings for API changes.
• Pull request title is a good summary of the changes - it will be used in the release notes.

[vercel]

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

Project	Deployment	Review	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jan 15, 2026 4:12pm

[Copilot]

Pull request overview

This pull request adds support for custom chat models to return Vercel AI SDK chunks, enabling rich streaming responses with reasoning, tool calls, and other advanced features. The implementation automatically detects and handles both Vercel AI SDK chunks (from pydantic-ai) and plain text strings, supporting both synchronous and asynchronous generators.

Changes:

• Removed the concept of "frontend-managed" streaming in favor of a unified approach where all streaming is handled through Vercel AI SDK chunks
• Added ChuckSerializer (typo - should be ChunkSerializer) class to automatically convert plain strings to Vercel AI SDK format or pass through dict/pydantic chunks as-is
• Updated frontend to always use streaming protocol, removing conditional logic based on frontendManaged flag

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
marimo/_plugins/ui/_impl/chat/chat.py	Core implementation: removes frontend-managed flag, adds ChuckSerializer for automatic chunk conversion, updates streaming response handling
tests/_plugins/ui/_impl/chat/test_chat_delta_streaming.py	Updates tests to verify Vercel AI SDK chunk format instead of accumulated text
tests/_plugins/ui/_impl/chat/test_chat.py	Adds comprehensive tests for dict chunks, pydantic chunks, and ChuckSerializer
frontend/src/plugins/impl/chat/chat-ui.tsx	Simplifies streaming logic by removing frontend-managed conditional branches, always uses streaming protocol
frontend/src/plugins/impl/chat/ChatPlugin.tsx	Updates return type from `string
marimo/_ai/llm/_impl.py	Adds helper to remove None values from dicts for proper serialization
marimo/_ai/_types.py	Sanitizes parts by removing None values before validation
marimo/_smoke_tests/ai/chat-variations.py	New smoke test file demonstrating various chat model patterns
examples/ai/chat/pydantic-ai-chat.py	Adds custom model example showing Vercel AI SDK chunk usage

---

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

[Copilot]

The error message handling has a potential issue. If error.message is undefined, the optional chaining ?.split() will fail on .pop(). This should use conditional logic or provide a fallback error message.

Suggested change

	const strippedError = (error as Error).message
	?.split("failed with exception ")
	.pop();
	const errorMessage =
	(error instanceof Error && typeof error.message === "string"
	? error.message
	: String(error || "Unknown error"));
	const strippedError = errorMessage.includes("failed with exception ")
	? errorMessage.split("failed with exception ").pop() ?? errorMessage
	: errorMessage;

[github-actions]

🚀 Development release published. You may be able to view the changes at https://marimo.app?v=0.19.3-dev45