adk-retrieval 0.0.2

Retrievers (RAG) for Google ADK that uses langchain

Langchain Retrievers for Google ADK

Install

uv add adk-retrieval

Usage

This library exposes LangchainRetrievalTool which is a Google ADK tool.

The tool expects an instance of langchain retriever (See - https://docs.langchain.com/oss/python/integrations/retrievers)

By default LangchainRetrievalTool returns str which is the page_content of the first langchain Document. If that satisfies your needs then look at the basic example below.

In some cases, you would want to post-process the documents and convert them to str as you see fit. In that case, you can either write your own Retriever or use the callback. See the little sophisticated example below.

Basic example

Below is a basic example that shows how to use a retriever from langchain in Google ADK agent.

The first (langchain) document's page_content is returned by LangchainRetrievalTool irrespective of any top_k you may have used in your langchain Retriever.

import wikipedia
from google.adk.agents import LlmAgent
from langchain_community.retrievers import WikipediaRetriever

from adk_retrieval import LangchainRetrievalTool

root_agent = LlmAgent(
    name="adk_wikipedia_agent",
    description="Agent for summarizing Wikipedia articles.",
    instruction="""You are a specialized search & summarizer agent for Wikipedia articles.
      """,
    model="gemini-2.5-flash",
    tools=[
        LangchainRetrievalTool(
            name="wikipedia_retrieval_tool",
            description="Tool for retrieving Wikipedia articles based on user queries.",
            retriever=WikipediaRetriever(wiki_client=wikipedia),
        )
    ],
)

A little sophisticated example

This example shows how you provide your own callback that takes all the documents returned by your langchain Retriever and convert them into a single string.

import wikipedia
from google.adk.agents import LlmAgent
from langchain_community.retrievers import WikipediaRetriever
from langchain_core.documents import Document

from adk_retrieval import LangchainRetrievalTool


# an example post-processing function that combines multiple documents into a single string
async def post_process(docs: list[Document]) -> str:
    sources = []
    for index, doc in enumerate(docs):
        sources.append(f"Source {index + 1}:\n{doc.page_content}\n")

    return "\n\n".join(sources)


root_agent = LlmAgent(
    name="wiki_sophisticated",
    description="Agent for summarizing Wikipedia articles.",
    instruction="""You are a specialized search & summarizer agent for Wikipedia articles.
      """,
    model="gemini-2.5-flash",
    tools=[
        LangchainRetrievalTool(
            name="wikipedia_retrieval_tool",
            description="Tool for retrieving Wikipedia articles based on user queries.",
            retriever=WikipediaRetriever(wiki_client=wikipedia, top_k_results=3),
            post_process=post_process,
        )
    ],
)

Examples

Open this repository in devcontainer and all dependencies will be setup for you

# Run examples using the built-in `adk web`
uv run adk web examples