Embeddings

Model Registry supports Embedding models for generating vector embeddings. Use them with the gateway’s POST /openai/v1/embeddings endpoint or when configuring vector storage (e.g., Knowledge Base, RAG).

Creating an Embedding Model

1. In FloTorch Console, go to Model Registry and click Create FloTorch Model (top right).
2. In the form, set Type to Embedding.
3. Enter a Name and optional Description. The name must be unique.
4. Choose an Embedding provider (e.g., OpenAI, Azure, Google, Cohere, Amazon Bedrock, OpenRouter, or OpenAI Compatible).
5. Select the Embedding model from that provider (e.g., text-embedding-3-small).
6. Optionally add Embedding model parameters (key/value) if the provider supports them.
7. Submit the form.

The embedding model is created with a default version and is ready to use—no version configuration canvas or publish step is required. It appears in the Model Registry table with type Embedding.

Using Embedding Models

• Embeddings API – Call POST /openai/v1/embeddings with model: "flotorch/<your-embedding-model-name>" to generate embeddings.
• Vector storage – When creating or editing a vector storage (e.g., ChromaDB, Pinecone, PgVector, LanceDB), select your FloTorch embedding model so the gateway uses it to embed documents and queries.

Embedding vs Chat Models

Aspect	Chat	Embedding
Purpose	Chat completions	Vector embeddings
Version canvas	Yes (router, cache, guardrails)	No
Publish step	Required to use	Not required; usable after create
Code snippet in Console	Available for published chat models	Not shown (embedding models use the embeddings API)