Supported providers

OpenAI

• Base URL: https://api.openai.com/v1
• Default model: text-embedding-3-small
• API endpoint: POST /embeddings

Models

Model	Dimensions	Description
text-embedding-3-small	1536	Small, fast, cost-effective
text-embedding-3-large	3072	Higher quality, supports custom dimensions
text-embedding-ada-002	1536	Legacy model

Supports the dimensions parameter to reduce embedding size (for text-embedding-3-* models).

---

Azure OpenAI

• Base URL: Computed from Azure resource configuration
• API endpoint: POST /embeddings

Azure OpenAI supports two modes depending on the api_version setting:

v1 mode

When api_version is set to "v1", behaves identically to OpenAI. Uses the standard connection config with base_url and token.

Legacy mode

Uses Azure-specific connection settings:

{
  "connection": {
    "resource_name": "my-azure-resource",
    "deployment_id": "my-embedding-deployment",
    "api_version": "2024-02-01",
    "api_key": "xxx"
  }
}

Parameter	Type	Description
resource_name	string	Azure resource name
deployment_id	string	Deployment ID (used as model)
api_version	string	API version (use "v1" for OpenAI-compatible mode)
api_key	string	API key (alternative to bearer token)

---

Azure AI Foundry

• Base URL: https://<resource>.services.ai.azure.com/models
• Default model: text-embedding-3-small
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

Mistral 🇫🇷 🇪🇺

• Base URL: https://api.mistral.ai
• Default model: mistral-embed
• API endpoint: POST /v1/embeddings

Models

Model	Dimensions	Description
mistral-embed	1024	Multilingual embedding model

---

Ollama (Local Models)

• Base URL: http://localhost:11434
• Default model: snowflake-arctic-embed:22m
• API endpoint: POST /api/embed

Ollama uses its native embed API (not OpenAI-compatible format). The token is optional for local installations.

Models

Any embedding model available in Ollama can be used, for example:

• snowflake-arctic-embed:22m
• nomic-embed-text
• mxbai-embed-large

---

Cohere

• Base URL: https://api.cohere.com
• Default model: embed-multilingual-v3.0
• API endpoint: POST /v2/embed

Cohere uses its own API format (not OpenAI-compatible). The extension handles the translation automatically.

Models

Model	Dimensions	Description
embed-multilingual-v3.0	1024	Multilingual, 100+ languages
embed-english-v3.0	1024	English-optimized

---

Gemini

• Base URL: https://generativelanguage.googleapis.com/v1beta/openai
• Default model: gemini-embedding-001
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

X-AI (Grok)

• Base URL: https://api.x.ai
• Default model: v1
• API endpoint: POST /v1/embeddings

---

Deepseek

• Base URL: https://api.deepseek.com
• Default model: deepseek-r1
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

Scaleway 🇫🇷 🇪🇺

• Base URL: https://api.scaleway.ai/v1
• Default model: qwen3-embedding-8b
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

Cloud Temple 🇫🇷 🇪🇺

• Base URL: https://api.ai.cloud-temple.com/v1
• Default model: embeddinggemma:300m
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

Huggingface 🇫🇷 🇪🇺

• Base URL: https://api-inference.huggingface.co/v1
• Default model: Qwen/Qwen3-Embedding-8B
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

Nebius AI Studio

• Base URL: https://api.studio.nebius.ai/v1
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

SambaNova

• Base URL: https://api.sambanova.ai/v1
• API endpoint: POST /embeddings

Uses OpenAI-compatible API.

---

OpenRouter

• Base URL: https://openrouter.ai/api/v1
• API endpoint: POST /embeddings

Uses OpenAI-compatible API. Provides access to multiple embedding models from different providers.

---

OpenAI Compatible

• Base URL: your own endpoint (set in the connection config)
• API endpoint: POST /embeddings

A generic provider for any embedding service that speaks the OpenAI /embeddings format. Point it at your endpoint with base_url and token, and give it a display name with provider_name — that name is what shows up for this provider everywhere in the gateway. When the target API needs it, you can also override the request headers, the param_mappings, and the additional_body_params.

{
  "connection": {
    "base_url": "https://my-endpoint/v1",
    "provider_name": "My Provider",
    "token": "xxx"
  },
  "options": {
    "model": "your-embedding-model"
  }
}

---

OVH AI Endpoints 🇫🇷 🇪🇺

• Base URL: https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
• API endpoint: POST /embeddings

Uses OVH's unified OpenAI-compatible API. Just set your token and the embedding model you want to use.

---

All MiniLM L6 V2 (local)

• No API call — runs entirely in-process via ONNX runtime
• Model: all-minilm-l6-v2
• Dimensions: 384

This model is embedded in the extension and requires no external connection or API token. It is used internally by the semantic cache.

Token usage is not tracked for this model.

---

Provider comparison

Provider	API format	Default model	Dimensions	Special features
OpenAI	OpenAI	text-embedding-3-small	1536	Custom dimensions
Azure OpenAI	OpenAI or Azure	(deployment)	varies	Two auth modes
Azure AI Foundry	OpenAI	text-embedding-3-small	1536	—
Mistral	OpenAI	mistral-embed	1024	—
Ollama	Ollama native	snowflake-arctic-embed:22m	varies	Local, no token needed
Cohere	Cohere v2	embed-multilingual-v3.0	1024	Multilingual
Gemini	OpenAI	gemini-embedding-001	varies	—
X-AI	OpenAI	v1	varies	—
Deepseek	OpenAI	deepseek-r1	varies	—
Scaleway	OpenAI	qwen3-embedding-8b	varies	—
Cloud Temple	OpenAI	embeddinggemma:300m	varies	—
Huggingface	OpenAI	Qwen/Qwen3-Embedding-8B	varies	—
Nebius	OpenAI	—	varies	—
SambaNova	OpenAI	—	varies	—
OpenRouter	OpenAI	—	varies	Multi-provider
OpenAI Compatible	OpenAI	(your model)	varies	Configurable name & headers
OVH AI Endpoints	OpenAI	—	varies	Unified API 🇫🇷 🇪🇺
All MiniLM L6 V2	Local ONNX	all-minilm-l6-v2	384	No API call