dspy-langchain-bridge

A production-grade compatibility bridge between DSPy and LangChain.

The Problem

DSPy and LangChain are the two most widely used LLM frameworks — and they don't talk to each other.

• LangChain has a massive ecosystem: tools, retrievers, agents, memory, LangGraph.
• DSPy has the best prompt optimizers in the ecosystem (BootstrapFewShot, MIPROv2, COPRO).

If you want DSPy's optimization inside a LangChain pipeline — or LangChain's retrievers inside a DSPy RAG module — you're currently on your own. dspy-langchain-bridge solves that.

Installation

pip install dspy-langchain-bridge

With optional extras:

pip install "dspy-langchain-bridge[openai,langgraph]"   # OpenAI + LangGraph support
pip install "dspy-langchain-bridge[all]"                # Everything

Quick Start

Use a DSPy module inside a LangChain chain

import dspy
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from dspy_lc_bridge import DSPyRunnable, LangChainLM

# Configure DSPy to use any LangChain LLM
llm = ChatOpenAI(model="gpt-4o")
dspy.settings.configure(lm=LangChainLM(llm))

# Build a DSPy module
cot = dspy.ChainOfThought("question -> answer")

# Drop it into an LCEL chain
prompt = ChatPromptTemplate.from_template("{question}")
chain = prompt | DSPyRunnable(cot)

result = chain.invoke({"question": "What is the capital of France?"})
print(result["answer"])

Use a LangChain retriever inside DSPy RAG

from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
from dspy_lc_bridge import LangChainRetriever

vectorstore = FAISS.load_local("my_index", OpenAIEmbeddings())
retriever = LangChainRetriever(vectorstore.as_retriever(), k=3)

class RAG(dspy.Module):
    def __init__(self):
        self.retrieve = retriever
        self.generate = dspy.ChainOfThought("context, question -> answer")

    def forward(self, question):
        passages = self.retrieve(question).passages
        context = "\n".join(passages)
        return self.generate(context=context, question=question)

Optimize a LangChain prompt with DSPy

from langchain_core.prompts import ChatPromptTemplate
from dspy.teleprompt import BootstrapFewShot
from dspy_lc_bridge import optimize_langchain_prompt

prompt = ChatPromptTemplate.from_template("Answer the question: {question}")

optimized = optimize_langchain_prompt(
    prompt=prompt,
    optimizer=BootstrapFewShot(metric=my_metric),
    trainset=my_examples,
)
# optimized is a ChatPromptTemplate with few-shot examples injected

Compatibility Matrix

Public API

from dspy_lc_bridge import (
    # Core
    DSPyRunnable,              # DSPy module → LangChain Runnable
    LangChainLM,               # LangChain LLM → DSPy LM backend

    # Signatures
    signature_to_prompt,       # dspy.Signature → PromptTemplate
    prompt_to_signature,       # PromptTemplate → dspy.Signature

    # Retrievers
    LangChainRetriever,        # LangChain retriever → DSPy Retrieve
    DSPyRetriever,             # DSPy Retrieve → LangChain BaseRetriever

    # Tools & Agents
    DSPyTool,                  # DSPy module → LangChain Tool
    LangChainTool,             # LangChain Tool → DSPy tool
    DSPyNode,                  # DSPy module → LangGraph node

    # Memory
    LangChainMemoryAdapter,    # LangChain memory → DSPy history

    # Optimizers
    optimize_langchain_prompt, # DSPy optimizer → optimized LangChain prompt
)

Roadmap

Contributing

See CONTRIBUTING.md. PRs, issues, and discussions are welcome.

License

MIT