Contravariance of `Tool[AgentDepsT]` does not seem to be respected by inferred `Tool[None]`

[lukekh]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

If my agent has any deps_type specified then tools that have no dependencies (e.g. the DuckDuckGoSearchTool from pydantic_ai.common_tools.duckduckgo) are type Tool[None] and cause the tool to complain by contravariance of the Tool type.

Shouldn't a Tool with no dependencies default to Tool[Any] to correctly use contravariant types?

I've gotten around this in my own code by just explicitly typing my tools like this:

ddgs_tool: Tool[Any] = duckduckgo_search_tool(max_results=5)

to avoid the type complaints from my linter (Pylance).

Example Code

from pydantic import BaseModel
from pydantic_ai import Agent
from pydantic_ai.common_tools.duckduckgo import duckduckgo_search_tool

class AgentDeps(BaseModel):
    foo: str
    bar: int

agent = Agent(
    deps_type=AgentDeps,
    tools=[duckduckgo_search_tool()]
)

Python, Pydantic AI & LLM client version

Python 3.11.13
pydantic-ai==1.0.10
openai==1.108.1

[DouweM]

@lukekh Changing the default on AgentDepsT from None to Any sounds reasonable, but I'd want to understand properly why we chose None to begin with before changing that, so a smaller change we could make right away that addresses the issue you're seeing is to explicitly type hint the duckduckgo_search_tool and tavily_search_tool functions to return Tool[Any]. Would you be up for submitting a PR to that effect?

[lukekh]

@DouweM Likewise! I was hoping someone could explain the logic too as I'm not 100% sure that I've considered every aspect of the type implications. That being said, I'm happy to make a P.R. for this.

P.S. These lines here are where AgentDepsT is defined to be contravariant (which I believe is correct) and perhaps default should be set to Any? I think this would fix they type for Tools but I don't know if that messes up other agent related types...

[lukekh]

Btw @DouweM, I have this fix but I can't make a branch and your contributing docs don't have any instructions for submitting a PR.

Do you need to make a branch and assign the issue to me, or something?

[DouweM]

@lukekh You can fork the repo, clone it (or add it as a remote to an existing local repo), create a branch, make the change (be sure to run make install and make typecheck), push it to your forked repo, and then GitHub will let you create a PR to get that branch merged into our main repo.