langchain - 💡(How to fix) Fix Add langchain-torchagentic partner package - differentiable planning tool [1 pull requests]

Submission checklist

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

I'd like to propose adding a new partner package called langchain-torchagentic — it wraps torchagentic's differentiable planning primitives (Value Iteration, MCTS) as a standard LangChain BaseTool.

The idea is to let LLM agents call differentiable planners the same way they'd call a calculator or a search tool — just another tool in the toolbox, but for planning action sequences. torchagentic gives you PyTorch-native planning: value iteration, Monte Carlo tree search, successor representation, all differentiable and composable.

I've already set this up following the same structure as the other partner packages in libs/partners/ (exa, xai, perplexity, etc.). The PR is at #37795 with tests passing, so it's ready for review — just needs a maintainer to take a look and assign the issue so the bot unblocks it.

Use Case

I'm building agents that need to plan sequences of actions using learned reward models — things like navigation, resource allocation, or game-playing. Right now there's no built-in way to do differentiable planning as a LangChain tool. You either roll your own planner or call an external solver, which breaks the seamless agent experience.

With langchain-torchagentic, you'd just register the tool and the agent can invoke planning as a natural step in its reasoning loop. The package already lives at libs/partners/torchagentic/ with tests passing — it's really just waiting for a maintainer to give it the green light.

Proposed Solution

A partner package at libs/partners/torchagentic/ exposing a TorchAgenticPlannerTool (subclass of BaseTool) that runs value iteration or MCTS on reward/transition models and returns a summary of the planned states/actions.

Usage would look like:

from langchain_torchagentic import TorchAgenticPlannerTool

tool = TorchAgenticPlannerTool(
    num_states=64,
    num_actions=8,
    planner_type="vi",  # or "mcts"
)

result = tool.invoke({"task": "Find optimal route from A to B"})

The implementation is already done in PR #37795 — just needs a reviewer.

Alternatives Considered

No response

Additional Context

No response