Embedding Model
On this page
Purpose
Embedding Model configures an embeddings client for semantic search and retrieval. It does not receive source text directly or return raw vectors.
When to use it
- Configure embeddings for a ChromaDB retrieval workflow.
- Choose Arkus or OpenAI embeddings for a vector store.
- Tune provider batching and retry behavior.
Required setup
Arkus is the default provider and uses the text-embedding-3-small deployment. It does not show an API key or base URL. OpenAI requires an OpenAI API key.
How to use it
- Add Embedding Model and select a provider and model.
- Connect Embedding Model to ChromaDB.
- Connect prepared Data to the vector store for ingestion.
- Run the vector store search path.
Configuration
| Setting | What it controls | Recommended starting point |
|---|---|---|
| Model Provider | Selects Arkus or OpenAI. | Start with Arkus. |
| Model Name | Selects the provider model or Arkus deployment. | Keep the default Arkus deployment unless the flow needs another model. |
| OpenAI API Key | Authenticates OpenAI embeddings. | Use a saved Credential. |
| Dimensions | Requests a compatible vector size for supported models. | Leave blank unless the vector store requires a size. |
| Chunk Size | Number of text chunks in one embedding request. | Leave at 36. |
| Max Retries | Provider retry count. | Leave at 3. |
Expected result
The component returns a configured Embeddings client. ChromaDB uses it to create and search vectors. Chunk Size batches provider requests. It does not split source documents, so use a preparation component before vector-store ingestion.
Reference details
Request Timeout, Show Progress Bar, Model Kwargs, and Dimensions are provider options. OpenAI validates its client with a small test embedding during build. Arkus does not make that validation call.
Provider credentials and credits
Arkus embedding calls use Arkus credits based on input tokens. OpenAI uses the selected OpenAI account and bypasses Arkus embedding metering. A deployed OpenAI key must be a saved Credential reference where credential-bearing deployment is enabled.
Related components
- ChromaDB requires the Embeddings client.
- Data Operations can prepare Data before it reaches a vector store.