Ensemble Retriever
Ensemble Retriever queries several retrievers at once and merges their results into one list. The usual setup combines one that searches by words, such as BM25 Retriever, with one that searches by meaning.
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
- You want to get both literal questions and questions in other words right.
- Your information is spread across several databases and you want to search all of them at once.
Configuration
| Field | What it does |
|---|---|
| Search Query | The question. |
| Retrievers (required) | The retrievers to combine. You can connect several. |
Advanced options
| Field | What it does |
|---|---|
| Number of Results | How many results Search Results returns at most. Default: 20. |
| Weights | How much each retriever counts, in the order they are connected, separated by commas, for example 0.7, 0.3. Empty: all count the same. |
Output
- Retriever: the search engine, to connect to components that ask for one, such as another in this family or a question-answering component.
- Search Results: the pieces found for Search Query, which you can pass to a prompt or an agent.
Use as an agent tool
Connect Ensemble Retriever to the Tools input of an Agent and the agent searches on its own: it writes Search Query with the question. It is called hybrid_search by default; in the Tool tab change the description to say what your documents contain.
Example
A repair service wants to find manuals both by exact model and by fault description.
- Connect a BM25 Retriever and the Retriever output of its manuals database to Retrievers.
- Connect the question to Search Query and pass Search Results to the agent.
Good practices
- Start with equal weights. In our test on 17 October 2026 (9 documents, 40 questions, a synthetic set), BM25 Retriever and search by meaning in equal parts got every code question in the top 5 results. Change the weights only if your own questions show that one retriever does better than the other.
- To order the combined list better, connect the Retriever output to Local Rerank and ask each retriever for more results than you keep.
Video
This video compares plain search with hybrid search plus reranking, on a question about a part reference.
What you see in the video:
- Qdrant alone, with its top 5 results, misses part VL-2210.
- BM25 Retriever and Qdrant joined in Ensemble Retriever with Weights
0.5, 0.5, followed by Local Rerank: it asks for 20 candidates and keeps the best 5. - The first result is the spare parts list of the Vellani Aurora manual, where “VL-2210: PT1000 temperature probe” appears, with a relevance of 0.56.
- The same combination, connected as an Agent tool, answers the question correctly.
Related components
- BM25 Retriever: searches by words.
- Chroma DB: a vector database these retrievers can start from.
- Local Rerank: reorders the combined list by relevance.
To see how search techniques combine, see Search techniques in the RAG guide.