Add convenience method to use LangChain community tools #1832

matthewfranglen · 2025-05-26T09:46:23Z

References #1831

This does work however the style and approach are very much open for discussion. In particular:

this creates a function just so it can be inspected, as part of that it has a very bad mapping from json schema types back to python.
the pydantic_ai.tool.Tool class has no base class that describes the desired behaviour.
the LangChain BaseTool defines json schemas for the args and similar which could more directly be translated to the pydantic_ai.tool.Tool class.
future integrations of this type might benefit from a similar approach.

To make my PR simple I have introduced a minimal change. I think it would be good to discuss a more maintainable and extensible approach.

…ther types

…checked, not executed

# Conflicts: # pydantic_ai_slim/pydantic_ai/_output.py # pydantic_ai_slim/pydantic_ai/tools.py

…er to rerun when the length of the example changes

This does work however the style and approach are very much open for discussion. In particular: * this creates a function just so it can be inspected, as part of that it has a very bad mapping from json schema types back to python. * the Tool class current has no base class that defines the behaviour. If it did then creating a separate class that doesn't need the round trip might be appropriate

matthewfranglen · 2025-05-26T09:48:52Z

pydantic_ai_slim/pydantic_ai/tools.py:316:40: F821 Undefined name `langchain_core`

I don't know how you want to handle the typing of the tool to pass in. It feels excessive to make pydantic_ai require langchain for type checking.

Added optional group of langchain and attempted import in TYPE_CHECKING block. If the import fails then a Protocol is defined for the tool.

matthewfranglen · 2025-05-26T09:52:48Z

This file contains the code for adapter.py and test_obama.py which show that the approach works (.py extension isn't allowed as an attachment). Again unsure of how you want to add tests for this stuff in?

adapter-and-test.txt

If the optional group is not installed then a protocol is used to define the tool type used.

hyperlint-ai · 2025-05-26T11:45:31Z

PR Change Summary

Introduced a new installation option for the LangChain library in the documentation.

Added installation instructions for the LangChain library in the install documentation.
Highlighted the need for discussion on the maintainability and extensibility of the proposed changes.

Modified Files

docs/install.md

How can I customize these reviews?

Check out the Hyperlint AI Reviewer docs for more information on how to customize the review.

If you just want to ignore it on this PR, you can add the hyperlint-ignore label to the PR. Future changes won't trigger a Hyperlint review.

Note specifically for link checks, we only check the first 30 links in a file and we cache the results for several hours (for instance, if you just added a page, you might experience this). Our recommendation is to add hyperlint-ignore to the PR to ignore the link check for this PR.

There are a bunch of unstable pre-commit hooks that I can't satisfy. The format one in particular wants to change unrelated code.

Just makes the checks unhappy

https://smolagents.org/docs/tools-of-smolagents-in-depth-guide/#3-toc-title They write well and in a positive way.

matthewfranglen · 2025-05-29T19:07:27Z

Think I have addressed all your comments. I have added a section to the tools documentation. It's heavily based on the smolagents documentation for the same functionality which you can view here. I updated the example to be something that requires a web search to answer correctly.

DouweM · 2025-05-29T19:45:08Z

pydantic_ai_slim/pydantic_ai/tools.py

+        """Creates a Pydantic tool from a function and a JSON schema.
+
+        Args:
+            function: The function to call


@matthewfranglen This discussion still needs to be resolved!

docs/tools.md

DouweM · 2025-05-29T19:53:26Z

pydantic_ai_slim/pydantic_ai/tools.py

+        def proxy(*args: Any, **kwargs: Any) -> str:
+            assert not args, 'This should always be called with kwargs'
+            kwargs = defaults | kwargs
+            return langchain_tool.run(kwargs)


Is there a validation error or something this could raise that we could wrap in ModelRetry, like we do with our own validation errors?

I tried writing a tool that immediately throws an exception and invoking it using a langchain agent. The agent invocation appeared to hang. Maybe this is something that could be further investigated in future.

OK, just wanted to make sure there are no obvious exceptions for us to catch and wrap and reraise here.

DouweM · 2025-05-29T19:55:18Z

docs/tools.md

+
+## Use LangChain Tools {#langchain-tools}
+
+We love LangChain and think it has a very compelling suite of tools. To import a tool from LangChain, use the `from_langchain()` method.


With all due respect to the LangChain team and product, we are competitors so I'd like this to be a bit more neutral :) Something like:

If you'd like to use a tool from LangChain's community tool library with PydanticAI, you can use the Tool.from_langchain convenience method. Note that PydanticAI will not validate the arguments in this case -- it's up to the model to provide arguments matching the schema specified by the LangChain tool, and up to the LangChain tool to raise an error if the arguments are invalid.

I'd expect "community tool library" to be a link.

If we manage to wrap LangChain tool errors in ModelRetry, we may be able to soften the language in the second sentence.

I think we should mention that it's a good way to transition to PydanticAI.

If you want to propose alternative text then please do. Since the documentation is presenting a specific view of the pydantic product that you want to preserve I think it would be faster for all involved if any documentation changes were provided verbatim.

The langchain-community dependency is not installed so it cannot be run

I am fed up with this linter. It's one import. It still fails. It doesn't say how to fix it and there isn't a way to automatically apply the fix.

docs/tools.md

DouweM · 2025-06-02T21:06:45Z

pydantic_ai_slim/pydantic_ai/tools.py

+        )
+
+    @classmethod
+    def from_langchain(cls, langchain_tool: LangChainTool) -> Self:


What do you think about moving this off of Tool and into a new module like pydantic_ai.ext.langchain? We will likely have more wrappers and convenience methods for other frameworks in the future, so my colleague @Viicos correctly pointed out that we should probably start moving them into their own namespace already instead of cluttering up the regular public classes.

This wouldn't need to be class method in that case, I imagine:

from pydantic_ai import Agent from pydantic_ai.ext.langchain import langchain_tool from langchain_community.tools import DuckDuckGoSearchRun agent = Agent(..., tools=[langchain_tool(DuckDuckGoSearchRun])

I've moved it and named the function from_langchain_tool.

DouweM · 2025-06-02T21:08:26Z

pydantic_ai_slim/pydantic_ai/tools.py

+        def proxy(*args: Any, **kwargs: Any) -> str:
+            assert not args, 'This should always be called with kwargs'
+            kwargs = defaults | kwargs
+            return langchain_tool.run(kwargs)


OK, just wanted to make sure there are no obvious exceptions for us to catch and wrap and reraise here.

DouweM and others added 22 commits May 20, 2025 19:44

Output handling refactoring borrowed from output_mode PR

b3f8bd2

Support functions as output_type, as well as lists of functions and o…

2ebe6a9

…ther types

Fix tests

ab576d7

Make Python 3.9 happy

6c4fcec

Support output_type = bound instance method

1bd16dc

Support RunContext arg on output_type function using same logic as tools

98e64d4

Improve test coverage

60d789e

Start output tool name disambiguation counter at 2

14e69d0

Stop requiring explicitly specifying type_ kwarg name on ToolOutput

ee95a80

Remove runtime assertion from typed_agent.py as the file is only type…

f1f093e

…checked, not executed

Add typing tests for Agent(output_type=)

66e5405

Merge remote-tracking branch 'origin/main' into output-type-callable

53d25f9

# Conflicts: # pydantic_ai_slim/pydantic_ai/_output.py # pydantic_ai_slim/pydantic_ai/tools.py

Update typing test for mypy

3aad6fc

Treat str in an output_type list the same as in a union

3ff6e74

Set ToolRetryError as cause on UnexpectedModelBehavior when available

a7fd8ac

Drop end line from example test parameterized test ID to make it easi…

5399df2

…er to rerun when the length of the example changes

Document output functions

e503edb

Fix docs

f436797

Update output_type typing tests

046b813

Update output_type typing tests

dc39e68

Document when automatic output_type type inference may fail

a578745

Add optional group of langchain to allow type checking

7f897b1

If the optional group is not installed then a protocol is used to define the tool type used.

matthewfranglen added 4 commits May 26, 2025 14:15

Fix linting problems

e3b3c7f

There are a bunch of unstable pre-commit hooks that I can't satisfy. The format one in particular wants to change unrelated code.

Add a test for tool conversion

8597ae4

Drop attempted import of langchain tool

2a92ba0

Just makes the checks unhappy

Older pythons don't enjoy type unions with |

2d22c32

matthewfranglen added 8 commits May 29, 2025 12:06

Use JsonSchemaValue to describe arguments

9194930

Make name and description arguments

9f89e02

Revert unrelated formatting changes

d2462d8

uv run ruff format

1c99adf

Handle $refs in the langchain schema

c709402

Replace arg iteration with assertion

f075d7d

Add flag to allow testing additional properties branch

65a04be

Write langchain tool section HEAVILY BASED ON SMOLAGENTS DOCS

b91cd66

https://smolagents.org/docs/tools-of-smolagents-in-depth-guide/#3-toc-title They write well and in a positive way.

matthewfranglen added 3 commits May 29, 2025 21:23

formatting for docs

5300f3f

Testing imports in docs is neat

a06f814

import sorting

24507f9

DouweM requested changes May 29, 2025

View reviewed changes

matthewfranglen added 5 commits May 30, 2025 10:29

Rephrase and skip tests over code

3ab281f

The langchain-community dependency is not installed so it cannot be run

Clarify the requirements over the function

e67b0ab

Add section on from_schema

d96fb5b

replace double quotes with single quotes, add comma

47e9d7b

docs/tools.md:415:1: I001 [*] Import block is un-sorted or un-formatted

17969fb

I am fed up with this linter. It's one import. It still fails. It doesn't say how to fix it and there isn't a way to automatically apply the fix.

DouweM requested changes Jun 2, 2025

View reviewed changes

matthewfranglen added 3 commits June 3, 2025 09:51

Move langchain tool converter

77f6a98

Add example of using tool from Tool.from_schema

0142858

remove assertions over _function_tools

77a7bee

DouweM changed the title ~~Suggested code for Tool.from_langchain~~ Add convenience method to use LangChain community tools Jun 6, 2025

DouweM added 4 commits June 6, 2025 18:32

Satisfy linter

669b299

Rename helper to tool_from_langchain

d83ef10

Update LangChain example

fe14acd

Update LangChain example

e558fc7

DouweM enabled auto-merge (squash) June 6, 2025 19:03

Merge branch 'main' into langchain-tools

3646fb3

DouweM merged commit 9e717dc into pydantic:main Jun 6, 2025
17 checks passed


		## Use LangChain Tools {#langchain-tools}

		We love LangChain and think it has a very compelling suite of tools. To import a tool from LangChain, use the `from_langchain()` method.

Add convenience method to use LangChain community tools #1832

Add convenience method to use LangChain community tools #1832

Uh oh!

Conversation

matthewfranglen commented May 26, 2025

Uh oh!

matthewfranglen commented May 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

matthewfranglen commented May 26, 2025

Uh oh!

hyperlint-ai bot commented May 26, 2025

PR Change Summary

Uh oh!

matthewfranglen commented May 29, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

matthewfranglen commented May 26, 2025 •

edited

Loading