Vector Store
Vector Store saves documents and searches them by meaning in the vector store you choose in Provider: FAISS and Chroma, which are kept in your account with no outside services, or Pinecone, pgvector, Milvus, Redis, Weaviate, Supabase, MongoDB Atlas, AstraDB, Cassandra, Couchbase, Neo4j, Deep Lake, Upstash or Vectara. Inside, it works exactly like that provider’s own component.
The difference is that switching provider does not mean switching node: you change it in the field, or the application that calls the flow changes it through the API.
When to use it
- You start with FAISS or Chroma to test and want to move to a cloud store, such as Pinecone or pgvector, without rebuilding the flow.
- Your application serves several customers and each one has its documents in its own vector store.
If the store will not change, use the provider’s own component, such as FAISS or Pinecone: it shows all its fields directly, such as the metadata filters.
Configuration
The Operation field decides what the component does:
Add(default): saves in the store the documents it receives on Ingest Data.Search: searches for the documents most similar to the text in Search Query.Retriever: offers the store as a retriever for another component to use, for example a question-answering chain.
| Field | What it does |
|---|---|
| Provider | The vector store. Default: FAISS. Choose it first: when you change it, the node asks for that provider’s credential. |
| Credential (required if the provider uses one) | The connection to the chosen provider’s store. The list only shows that provider’s credentials. FAISS and Chroma need none. See Credentials. |
| Operation | What the component does: Add, Search or Retriever. |
Ingest Data (required with Add) | The documents to save, usually those coming out of a text splitter or a file loader. |
| Search Query | With Search. The text to search for. If you leave it empty nothing is returned. |
| Embedding | The connector where you plug the embedding model, such as Embedding Model or OpenAI Embeddings. Always use the same one to save and to search. |
| Collection Name | The store’s collection, index or table, whatever each provider calls it. Empty: the provider’s default. |
The node remembers the credential you choose for each provider: when you go back to a provider you already set up, its credential comes back by itself.
Advanced options
| Field | What it does |
|---|---|
| Number of Results | How many documents each search returns. Default: 4. |
| Search Type | Similarity returns the most similar; Similarity with score threshold, only those above Search Score Threshold. Only shown if the provider supports it. |
| Search Score Threshold | Minimum similarity score, only with the threshold search. Default: 0.1. |
| Provider Settings | Other fields of the provider’s own component, each with its own name. For example, persist_directory = faq with FAISS or Chroma, to choose the storage folder. Fields the provider does not have are ignored. |
| Credentials by Provider | The credential the node remembers for each provider. It fills itself in when you choose a credential. |
Switching the provider through the API
From another application, change the provider, the collection or the search with tweaks, using the node’s ID:
"tweaks": { "MultiProviderVectorStore-v8d2q": { "provider": "Chroma", "Operation": "Search", "collection_name": "customer_manual", "search_query": "How do I change my password?" }}The provider must have been set up on the canvas before, with its credential; otherwise, send credential_id too. More in Changing fields for one call.
Output
The outputs depend on Operation:
- Results: with
Add, the documents that were saved; withSearch, the documents found, which you can pass to a prompt or an agent. - Vector Store: the store, to connect it to components that work with a vector store.
- Retriever: with
Retriever. The retriever, to connect it to components that ask for one.
In tool mode, an agent can use the component to search by itself. See Tool Mode.
Example
An accounting firm tests a search over its guides with FAISS and, once it works, moves it to its Pinecone store.
- Connect the guides, split into chunks, to Ingest Data of Vector Store with Operation set to
Add, and an Embedding Model to Embedding. - With
FAISSin Provider, run the flow and try some searches withSearch. - Switch to
Pinecone, choose its credential and type the index in Collection Name. Run again withAddto load the guides into Pinecone.
Good practices
- Each store keeps its own documents: when you switch provider, the documents do not follow. Load them with
Addinto the new one. - Search with the same embedding model you filled the collection with.
- To filter by metadata or use other options specific to a provider, use its own component or write the field in Provider Settings.
Related components
- Embedding Model: the embedding model where you also choose the provider.
- FAISS, Chroma DB and Pinecone: vector stores with a fixed provider.
- Retrieval QA: answers questions with the documents the retriever finds.