Multi Query Retriever
Multi Query Retriever asks a model to write the question in several different ways, searches with each one in another retriever and merges the results. It helps when users ask with different words from those in your documents.
A retriever receives a question and returns the document pieces that best answer it, so an agent or model can reply with your information.
When to use it
- Your customers’ questions are short, ambiguous or use their own words.
- Normal search misses documents that do answer, or does not put them first.
Configuration
| Field | What it does |
|---|---|
| Search Query | The original question. |
| Number of Questions | How many versions of the question the model writes. Each one is a search in Base Retriever. Default: 3. |
| Include Original query | Also searches with the question as it was written. On by default. |
| Base Retriever (required) | The retriever each version is searched in, for example the Retriever output of a vector database. |
| Model (required) | The model that writes the versions. A fast, cheap one is enough. |
Advanced options
| Field | What it does |
|---|---|
| Number of Results | How many results Search Results returns at most: those of the first version, then the new ones of each next version. Default: 20. |
| Prompt | Your own instructions for writing the versions, if the built-in ones do not suit you. Use {question} where the question goes and {number_of_questions} for how many to write. Empty: the built-in prompt. |
Output
- Retriever: the search engine, to connect to components that ask for one, such as Local Rerank or another in this family.
- Search Results: the pieces found for Search Query, which you can pass to a prompt or an agent. Each run’s log lists the questions that were searched.
Use as an agent tool
Connect Multi Query Retriever to the Tools input of an Agent and the agent searches on its own: it writes Search Query with the question. In the Tool tab, Tool Name and Tool Description are the name and explanation the agent uses to recognise the tool. It is called multi_query_search by default; change the description to say what your documents contain.
Example
A bank wants its assistant to find the right policy even if the customer asks “what happens if I lose my card?”.
- Connect the Retriever output of its policies database to Base Retriever and a model to Model.
- Connect the component to the agent’s Tools.
The run’s log shows the searched questions, for example “procedure for a lost or stolen card”, “how to block a lost bank card” and the original one.
Measured figures
Measured on 17 October 2026 on a test knowledge base (9 documents of a fictional repair service, 40 questions; DeepSeek as the model, 3 versions, base retriever with 5 results). It is a single synthetic set: it shows how the technique behaves, not a promise for your data.
- On questions phrased with other words than the documents, the right answer came first in 90 % to 100 % of the questions depending on the run (the model does not always write the same versions), against 80 % with a normal search. It was already among the first five in all of them.
- On other kinds of questions (exact codes, filters) it did not improve on normal search.
- It costs one model call per question: about 1.2 seconds.
Video
This video shows how Multi Query Retriever finds an answer that plain search misses when the question uses other words, and how compression keeps only the useful results.
What you see in the video:
- A question phrased in other words, along the lines of “if a repair breaks again a few months later, do you charge me again?”. Plain search does not bring back the repair warranty section.
- Multi Query Retriever writes 3 versions of the question and also searches with the original one: now it finds that section.
- The run’s log shows the questions it searched with.
- Contextual Compression Retriever, with a model that reviews 10 results and drops the ones that do not answer, keeps only a few: the log says how many it found and how many it kept.
Related components
- Local Rerank: reorders the merged results by relevance.
- Chroma DB: a vector database these retrievers can start from.
To see how search techniques combine, see Search techniques in the RAG guide.