Connect an Embedding Model

How to enable text embeddings in Sciwand - provider steps vs Sciwand steps, billing, and indexing your library.

New to this? Start with Connecting AI Providers for what an API key is and which model to pick.

Embedding models power semantic search and MAX (full-context) features by turning your library into vectors. Only one embedding model is active at a time. As with chat models, there are a few steps at the provider and a few in Sciwand.

No subscription needed. Embedding APIs are pay-as-you-go and are usually very cheap; several providers (Voyage, Gemini) include a free tier. A consumer subscription does not include API usage. See Connecting AI Providers.

Part A - at the provider

  1. Sign in to the provider and create an API key (copy it).
  2. Add a little credit under Billing if the provider requires it (Voyage and Gemini have generous free tiers).

Part B - in Sciwand

  1. Open Settings and select the Embeddings tab.
  2. Turn on Text Embeddings.
  3. Click Select (or Change, if one is already set) to open the model picker.
  4. Pick a provider (OpenAI, Gemini, Voyage, Together, or a local option).
  5. Click Configure and enter your API Key (cloud) or Server URL (local), then Save Changes.
  6. Choose the embedding model you want. This becomes the single active model.
  7. Click Generate All to start indexing your library.

Important: switching models clears vectors

Different embedding models produce incompatible vectors. If you change the active model, Sciwand must clear previously generated vectors and re-index. You will be asked to confirm before this happens.

Shared API keys

For OpenAI, Gemini, and Together, the API key is shared with the AI Models tab. If you already connected one there for chat, it is reused here. Voyage is embeddings-only and uses its own key.

Provider guides