Local Knowledge
Local Knowledge
Section titled “Local Knowledge”Local knowledge
What it does
Section titled “What it does”Stores and retrieves document embeddings locally in the browser for semantic search.
When to use it
Section titled “When to use it”Connect this to RAG or Q&A agents to provide a local knowledge base for question answering and information retrieval.
Inputs and settings
Section titled “Inputs and settings”| Setting | Notes |
|---|---|
| Knowledge | The knowledge store to read from and write to. |
Outputs
Section titled “Outputs”Returns a vector store dependency for retrieval and RAG workflows.
Dependencies and credentials
Section titled “Dependencies and credentials”- Embeddings (optional) — the model used to turn text into vectors. Connect Local Embeddings to stay fully on-device, or another embeddings node. See Embedding models for how the choice is remembered.
Embedding models
Section titled “Embedding models”Each knowledge store remembers the embedding model it was built with, and is always searched and filled with that model. Vectors from different models live in different spaces and can’t be compared, so they are never mixed.
- New stores use the embeddings connected on their first run — or, if none is connected, the on-device all-MiniLM-L6-v2 model. That model is then recorded on the store.
- Older stores (created before stores remembered their model) keep the Universal Sentence Encoder they were built with, so existing search results stay correct.
- Mismatched embeddings — if you connect a different model than the one a store was built with, the store’s own model is used for that run and a warning is written to the run log. To move to another model, create a new store and index your documents again.
If a store was built with a non-local embeddings node, connect that same node to use it; the store can’t rebuild those vectors on its own.
Example workflow
Section titled “Example workflow”Connect Local Embeddings to Local Knowledge, and Local Knowledge to the Knowledge port of a Q&A Agent or RAG Agent. Index documents with the Indexer, then ask questions.
Troubleshooting
Section titled “Troubleshooting”- “Model mismatch” warning — the connected embeddings differ from the store’s model; the store’s model was used. Disconnect or match the embeddings, or use a new store.
- ”… was built with … embeddings. Connect that embeddings node to use it.” — the store uses a non-local embeddings model; connect it.
- Poor search results after changing embeddings — changing the connected model doesn’t re-embed an existing store. Create a new store for the new model.
- If the node uses browser page data, run it on the target tab after the page has loaded.