chore(ai): support pydantic-ai v2

[Light2Dark]

📝 Summary

Motivation: less bugs, capability features, better pydantic-ai support

• Remove old pydantic-ai v1 compatibility shims (require_vercel_ai_sdk_support, model_dump fallbacks, AnthropicModelProfile.from_profile, explicit Anthropic temperature handling).
• Adapt to v2 API changes: ModelProfile as TypedDict (.get() access), Agent[object, ...] typing, FunctionToolset without None type param.
• Fix OpenRouter test model IDs to use documented openrouter/<vendor>/<model> slugs — bare openrouter/gpt-4 breaks under v2 because profile resolution is eager and requires the upstream vendor prefix.
• Reorder helper cells in the pydantic-ai chat example so underscore-prefixed helpers are defined before custom_model.

📋 Pre-Review Checklist

• 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).
• Any AI generated code has been reviewed line-by-line by the human PR author, who stands by it.
• Video or media evidence is provided for any visual changes (optional).

✅ Merge Checklist

• I have read the contributor guidelines.
• Documentation has been updated where applicable, including docstrings for API changes.
• Tests have been added for the changes made.

Made with Cursor

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jun 26, 2026 4:44am

[Light2Dark]

pydantic-ai-slim: dependency changes (1.54.0 → 2.0.0)

Comparison of declared dependency constraints between pydantic-ai-slim 1.54.0 and 2.0.0.

Sources:

• pydantic_ai_slim/pyproject.toml @ v1.54.0
• pydantic_ai_slim/pyproject.toml @ v2.0.0

Core deps (always installed with slim)

Dependency	1.54.0	2.0.0
pydantic-graph	==1.54.0	==2.0.0
pydantic	>=2.10	>=2.12 (via pydantic-graph)
genai-prices	>=0.0.48	>=0.0.62
Docstring tooling	griffe>=1.14.0	griffelib>=2.0 (direct; no griffe CLI stack)
httpx	>=0.27	>=0.27 (unchanged)
opentelemetry-api	>=1.28.0	>=1.28.0 (unchanged)
typing-inspection	>=0.4.0	>=0.4.0 (unchanged)

Extra-specific mins (only if you use that extra)

Extra	1.54.0 min	2.0.0 min	Notes
openai	openai>=2.11.0	openai>=2.29.0	Biggest practical bump
anthropic	anthropic>=0.75.0	anthropic>=0.108.0
google	google-genai>=1.56.0	google-genai>=1.70.0	vertexai extra removed; use google
bedrock	boto3>=1.42.14	boto3>=1.42.63	Small floor bump
openai	tiktoken>=0.12.0	tiktoken>=0.12.0	Unchanged
mcp	mcp>=1.25,<2 + fastmcp>=2.14	fastmcp-slim[client]>=3.3.0	Different MCP stack entirely

[akshayka]

If Pydantic AI is popular, then increasing the lower bound in this way will prevent users from using other packages that rely on older pydantic-ai < 2.0.0.

@Light2Dark, did you consider adding support for >= 2.0 without removing support for < 2.0?

[Light2Dark]

If Pydantic AI is popular, then increasing the lower bound in this way will prevent users from using other packages that rely on older pydantic-ai < 2.0.0.

@Light2Dark, did you consider adding support for >= 2.0 without removing support for < 2.0?

It's possible, and maybe we can support it for some time. However, if v2 makes more changes, we can't guarantee newer versions don't break, and we shouldn't cap it imo, a lot of bug fixes gets shipped as new models, APIs get released. It is an increase in our maintenance cost to support both.

I'll see if we can support both for the time being.