Skip to main content

How to handle multiple retrievers


This guide assumes familiarity with the following:

Sometimes, a query analysis technique may allow for selection of which retriever to use. To use this, you will need to add some logic to select the retriever to do. We will show a simple example (using mock data) of how to do that.


Install dependencies​

yarn add @langchain/community @langchain/openai zod chromadb

Set environment variables​


# Optional, use LangSmith for best-in-class observability

Create Index​

We will create a vectorstore over fake information.

import { Chroma } from "@langchain/community/vectorstores/chroma";
import { OpenAIEmbeddings } from "@langchain/openai";
import "chromadb";

const texts = ["Harrison worked at Kensho"];
const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
const vectorstore = await Chroma.fromTexts(texts, {}, embeddings, {
collectionName: "harrison",
const retrieverHarrison = vectorstore.asRetriever(1);

const texts = ["Ankush worked at Facebook"];
const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
const vectorstore = await Chroma.fromTexts(texts, {}, embeddings, {
collectionName: "ankush",
const retrieverAnkush = vectorstore.asRetriever(1);

Query analysis​

We will use function calling to structure the output. We will let it return multiple queries.

import { z } from "zod";

const searchSchema = z.object({
query: z.string().describe("Query to look up"),
person: z
"Person to look things up for. Should be `HARRISON` or `ANKUSH`."

Pick your chat model:

Install dependencies

yarn add @langchain/openai 

Add environment variables


Instantiate the model

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({
model: "gpt-3.5-turbo",
temperature: 0
import { ChatPromptTemplate } from "@langchain/core/prompts";
import {
} from "@langchain/core/runnables";

const system = `You have the ability to issue search queries to get information to help answer user information.`;
const prompt = ChatPromptTemplate.fromMessages([
["system", system],
["human", "{question}"],
const llmWithTools = llm.withStructuredOutput(searchSchema, {
name: "Search",
const queryAnalyzer = RunnableSequence.from([
question: new RunnablePassthrough(),

We can see that this allows for routing between retrievers

await queryAnalyzer.invoke("where did Harrison Work");
{ query: "workplace of Harrison", person: "HARRISON" }
await queryAnalyzer.invoke("where did ankush Work");
{ query: "Workplace of Ankush", person: "ANKUSH" }

Retrieval with query analysis​

So how would we include this in a chain? We just need some simple logic to select the retriever and pass in the search query

const retrievers = {
HARRISON: retrieverHarrison,
ANKUSH: retrieverAnkush,
import { RunnableConfig, RunnableLambda } from "@langchain/core/runnables";

const chain = async (question: string, config?: RunnableConfig) => {
const response = await queryAnalyzer.invoke(question, config);
const retriever = retrievers[response.person];
return retriever.invoke(response.query, config);

const customChain = new RunnableLambda({ func: chain });
await customChain.invoke("where did Harrison Work");
[ Document { pageContent: "Harrison worked at Kensho", metadata: {} } ]
await customChain.invoke("where did ankush Work");
[ Document { pageContent: "Ankush worked at Facebook", metadata: {} } ]

Next steps​

You’ve now learned some techniques for handling multiple retrievers in a query analysis system.

Next, check out some of the other query analysis guides in this section, like how to deal with cases where no query is generated.

Was this page helpful?

You can also leave detailed feedback on GitHub.