Local Embeddings

Turn text into vectors on your own device, so a knowledge base can find passages by meaning without an API key or sending anything out.

What it does

Local Embeddings plugs into the Embeddings slot of a Local Knowledge node. It does not run on its own and has no output. When the knowledge base is filled or searched, an in-browser model turns the text into vectors (lists of numbers that capture meaning). It runs in the background, so the editor stays responsive.

• The default model is all-MiniLM-L6-v2. You can also pick the older Universal Sentence Encoder (to keep older knowledge bases readable) or any embeddings model you installed.
• A knowledge base remembers the model it was built with and always uses it. A new knowledge base uses the model connected on its first fill, or all-MiniLM-L6-v2 when nothing is connected. Connecting another model later does not re-embed it: the knowledge base keeps its own model and the run log shows a warning.
• The model loads on first use (downloading it if needed), then stays ready for the session.

When to use it

• A private, offline knowledge base for the RAG Agent or Q&A Agent.
• Indexing many documents without paying per token.
• You want to pick a specific on-device embeddings model instead of the default.

Since a new knowledge base already uses all-MiniLM-L6-v2 when nothing is connected, you only need this node to choose another on-device model.

Inputs and settings

Einstellung	Typ	Was es bewirkt
Model	Model picker	An on-device embeddings model. all-MiniLM-L6-v2 is the recommended default; Universal Sentence Encoder keeps older stores readable. Standard: Xenova/all-MiniLM-L6-v2

Pick the Model before the knowledge base’s first fill. To switch an existing knowledge base to another model, re-embed it from its settings on the knowledge bases page. Install more models from the Local Models page, or download them from the picker.

Outputs

This node has no output port. It plugs into a slot on another node instead of passing items along.

• Embeddings slot of Local Knowledge

The vectors go to the knowledge base, not to the next node.

Dependencies and credentials

None. This node needs no connection and has no slots for other nodes.

Example workflow

1. A Context Menu trigger on the page.
2. Get All Text with Output field name pageText.
3. Indexer with Input {{ $input.pageText }}, a Recursive Character Text Splitter in its Text Splitter slot, and a Local Knowledge pointing to a new knowledge base in its Knowledge slot.
4. Local Embeddings in the Local Knowledge’s Embeddings slot, with the model you want (for example Xenova/all-MiniLM-L6-v2).

Later, a RAG Agent with the same Local Knowledge answers questions about the pages you saved.

Common issues

• “Embeddings model “…” isn’t available. Install it from Local AI.“ The model is not installed in this browser. Install it from Local Models, or download it from the picker.
• A “Model mismatch” warning in the run log. The knowledge base was built with another model, so that model was used. Disconnect this node, or re-embed the knowledge base with this model.
• The first run is slow. The model is downloading and loading. Later runs in the same session are faster.

Related nodes

• Local Knowledge: the knowledge base it plugs into.
• Indexer: fill a knowledge base from a workflow.
• Ollama Embeddings: on-device embeddings through the Ollama app.