Embeddings¶
KARLI-hosted embedding models are used through the standard Embedding Model component: drop the component on the canvas, select KARLI as the provider, and pick a model from the dropdown. There is no dedicated KARLI Embeddings component.
Available Models¶
| Model | Notes |
|---|---|
embed_e5_query |
E5-family query embedding endpoint. |
embed_q15_query |
Q15 query embedding endpoint. |
Both models return 1024-dimensional vectors.
Usage in Flows¶
Connect the Embedding Model component to any component that requires embeddings — typically a vector store, retrieval component, or similarity search node.
flowchart LR
Documents --> EM["Embedding Model
(KARLI)"]
EM --> VS["Vector Store"]
Single-text and batch embedding requests are routed to dedicated endpoints (/embed_* and /batch_embed_*) under the configured base URL.
Tip
When choosing an embedding model, consider the language of your documents. Some models perform better on multilingual content than others.