Embedding providers
Turning text into vectors for storage, search and retrieval.
startup.embedding turns text into vectors for /embed, /search/text and retrieval in
/api/generate. A collection's dimension is fixed by the model that first fills it.
provider | What it talks to |
|---|---|
openai | OpenAI with OPENAI_API_KEY, or any server speaking the format (TEI, vLLM, llama.cpp, llama-swap) when base_url is the full embeddings endpoint |
ollama | an Ollama server |
piramid | a Qwen3 embedding checkpoint run in-process; model is the directory, options takes device (cpu or cuda:N), dtype and max_tokens |
startup:
embedding:
provider: openai
model: nomic-embed-text
base_url: http://127.0.0.1:8100/v1/embeddings