3.2 Embedding Configuration
An Embedding Configuration is a saved connection to an embedding model: a model that turns text into vectors so it can be searched by meaning rather than by exact words. Graph Memories use one to vectorize the content of the entities they cover, which is what makes similarity-based retrieval possible when an Agent looks up relevant data.
An Embedding Configuration follows the same pattern as a Provider Configuration: it stores a provider type and a Config (JSON) block with the credentials, endpoint, and model. The two are separate because embedding models and chat (LLM) models are different services with a much smaller provider list for embeddings.
Supported providers
Four embedding providers are available in the Provider dropdown:
- OpenAI: Config keys
api_keyandmodel. - Azure OpenAI: Config keys
api_key,api_version,api_base, andmodel. - OAuth2 Proxy: For an embedding endpoint behind an OAuth2 proxy. Config keys
model,auth_base_url,auth_token_url,client_id,client_secret, andcert_path. - SBERT: A self-hosted Sentence-BERT endpoint. Config keys
modelandapi_base.

Azure and SBERT use api_base for the endpoint here, not the base_url key used by the LLM Provider Configuration. Copy the key names from the pre-filled Config (JSON) rather than reusing an LLM configuration.
Choose an embedding model that matches the language and domain of your data. The model also fixes the vector dimensionality: once a Memory has been initialized with a given embedding model, keep using the same configuration for it, because vectors from different models are not comparable.
Create, edit, and delete
- Open the Embedding Configuration section from the sidebar and click Add new.
- Enter a title (and an optional description).
- Select a provider from the Provider dropdown. The Config (JSON) field fills with that provider's default keys.
- Replace the placeholder values, then click Save. The JSON is validated on save.
Use Duplicate configuration to copy an existing one, and Delete configuration (or the list's checkbox and bulk delete) to remove one. The Search field filters the overview list.