Supported Providers

Inference Gateway provides a unified interface to interact with multiple LLM providers. This page details each supported provider, their configuration, and usage examples.

Available Providers

Provider	Auth	Default URL	Vision Support
OpenAI	Bearer Token	`https://api.openai.com/v1`	No
DeepSeek	Bearer Token	`https://api.deepseek.com`	No
Anthropic	X-Header	`https://api.anthropic.com/v1`	No
Cohere	Bearer Token	`https://api.cohere.ai`	No
Groq	Bearer Token	`https://api.groq.com/openai/v1`	No
Cloudflare	Bearer Token	`https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai`	No
Ollama Cloud	Bearer Token	`https://ollama.com/v1`	No
Ollama	Optional API key	`http://ollama:8080/v1`	No
llama.cpp	Bearer Token	`http://llamacpp:8080/v1`	No
Google	Bearer Token	`https://generativelanguage.googleapis.com/v1beta/openai`	No
Mistral	Bearer Token	`https://api.mistral.ai/v1`	No
MiniMax	Bearer Token	`https://api.minimax.io/v1`	No
Moonshot	Bearer Token	`https://api.moonshot.ai/v1`	No
NVIDIA	Bearer Token	`https://integrate.api.nvidia.com/v1`	No
Z-AI	Bearer Token	`https://api.z.ai/api/coding/paas/v4`	No
ElevenLabs	X-Header	`https://api.elevenlabs.io/v1`	No

Every provider above serves /v1/chat/completions except ElevenLabs, which has no chat-completions API: it is audio-only and answers the Audio API alone - speech synthesis with elevenlabs/<model> and an ElevenLabs voice id, plus sound-effect generation through POST /v1/audio/sfx with elevenlabs/eleven_text_to_sound_v2, and music generation through POST /v1/audio/music - it is the only provider serving sound effects and music today. It is also the first provider behind the Videos API: elevenlabs/creatify-aurora renders audio-driven talking avatars at 480p or 720p (chosen through the size field). Routing a chat request to it is not supported.

Vision/Multimodal Support

Several providers support vision/multimodal capabilities, allowing you to process images alongside text. To have the gateway handle image content, enable it in your configuration:

bash

VISION_ENABLED=true

Note: Vision handling is disabled by default. When disabled, the gateway does not inspect image content and forwards it to the provider untouched. When enabled, image parts are stripped only from requests to models known to accept non-image input only (the request continues with text only); models the gateway has no modality information for are passed through. Image content never causes the gateway to reject a request.

Providers with Vision Support

Example Vision Request

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-4-8",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "What is in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://example.com/image.jpg"
            }
          }
        ]
      }
    ]
  }'

Using Providers

Provider Configuration

Each provider requires specific configuration through environment variables:

PROVIDER_API_URL: The base URL for the provider's API
PROVIDER_API_KEY: The authentication key for the provider

Replace "PROVIDER" with the provider name (uppercase): OPENAI, DEEPSEEK, ANTHROPIC, COHERE, GROQ, CLOUDFLARE, OLLAMA_CLOUD, OLLAMA, LLAMACPP, GOOGLE, MISTRAL, MINIMAX, MOONSHOT, NVIDIA, ZAI, ELEVENLABS.

API Endpoints

Inference Gateway offers two main approaches to interact with providers:

1. Unified Generate API

The unified API allows you to generate content with a consistent interface across all providers:

http

POST /v1/chat/completions
Content-Type: application/json

{
  "model": "MODEL_NAME",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "Hello, world!"
    }
  ]
}

2. Provider Proxy

You can also proxy requests directly to the provider's native API:

bash

POST /proxy/{provider}/{path}
Content-Type: application/json

// Provider-specific request body

Provider-Specific Examples

OpenAI Provider

Generate content with OpenAI models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Hello, world!"
      }
    ]
  }'

List all available models:

bash

curl http://localhost:8080/v1/models

DeepSeek Provider

Generate content with DeepSeek models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "What is the capital of France?"
      }
    ]
  }'

List available models:

bash

curl http://localhost:8080/v1/models?provider=deepseek

Anthropic Provider

Generate content with Anthropic Claude models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-4-8",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain quantum computing in simple terms."
      }
    ]
  }'

List available models:

bash

curl http://localhost:8080/v1/models?provider=anthropic

Cohere Provider

Generate content with Cohere models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cohere/command-a-03-2025",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Write a short poem about AI."
      }
    ]
  }'

Groq Provider

Generate content with Groq's high-performance models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "groq/llama-3.3-70b-versatile",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "What are the benefits of quantum computing?"
      }
    ]
  }'

Cloudflare Provider

Generate content with Cloudflare Workers AI:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cloudflare/@cf/meta/llama-3.3-70b-instruct-fp8-fast",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain how neural networks work."
      }
    ]
  }'

Ollama Provider

Generate content with locally-hosted Ollama models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ollama/llama3.3",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Write a function to calculate Fibonacci numbers in Python."
      }
    ]
  }'

Ollama Cloud Provider

Generate content with Ollama's cloud-hosted models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ollama_cloud/gpt-oss:120b",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Summarize the benefits of running models in the cloud."
      }
    ]
  }'

Ollama Cloud models are subscription-gated rather than metered per token. GET /v1/models?include=pricing returns a populated pricing object for every ollama_cloud/* model with subscription: true, so clients classify them as Subscription instead of Free. Models with a published pay-as-you-go rate keep non-zero input_per_token / output_per_token values for reference; the flag still decides billing. See List All Models for the full resolution order.

Google Provider

Generate content with Google's Gemini models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-3-flash",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain the concept of machine learning in simple terms."
      }
    ]
  }'

Note: The default Google base URL ends in /v1beta/openai, which is Google's OpenAI-compatible endpoint. The gateway speaks the OpenAI protocol (/chat/completions, /models), so this suffix is required for routing to work. It is distinct from Google's native Gemini API at /v1, which uses a different request/response format and is not OpenAI-compatible. Keep the /v1beta/openai suffix when overriding GOOGLE_API_URL.

Mistral Provider

Generate content with Mistral AI models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistral/mistral-large-3",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain the differences between supervised and unsupervised learning."
      }
    ]
  }'

Moonshot Provider

Generate content with Moonshot AI models:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshot/kimi-k2-thinking",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "What are the key principles of clean code?"
      }
    ]
  }'

Generate content with a Moonshot vision model, using an image URL:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshot/kimi-latest",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "What is in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://example.com/image.jpg"
            }
          }
        ]
      }
    ]
  }'

NVIDIA Provider

Generate content with NVIDIA NIM models hosted on the build.nvidia.com API catalog:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nvidia/meta/llama-3.1-8b-instruct",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain the concept of GPU acceleration."
      }
    ]
  }'

List available models:

bash

curl http://localhost:8080/v1/models?provider=nvidia

Note: NVIDIA is additive alongside Groq. Groq provides ultra-low-latency inference via LPU hardware, while NVIDIA offers a broad GPU-served catalog including Nemotron, Llama, DeepSeek, Mistral, and Qwen. They are complementary providers.

llama.cpp Provider

Generate content with self-hosted GGUF models via llama.cpp's llama-server:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llamacpp/llama-3.2-3b-instruct",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain how llama.cpp works in simple terms."
      }
    ]
  }'

List available models:

bash

curl http://localhost:8080/v1/models?provider=llamacpp

Note: LLAMACPP_API_KEY is optional. When your llama-server is started with --api-key, set LLAMACPP_API_KEY to the same value and the gateway forwards it to llama.cpp as an Authorization: Bearer <key> header. Leave it empty or unset for local, unauthenticated servers - no Authorization header is sent, so existing deployments are unaffected.

Z-AI Provider

Generate content with Z-AI models for direct access to open-weight models like GLM:

bash

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai/glm-5.2",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Explain the benefits of open-weight models."
      }
    ]
  }'

List available models:

bash

curl http://localhost:8080/v1/models?provider=zai

ElevenLabs Provider

ElevenLabs is audio-only: speech, sound effects, music and avatar videos. It has no chat-completions API.

ElevenLabs' own models endpoint lists speech models only. The gateway appends the sound-effect (eleven_text_to_sound_v2), music (music_v2, music_v2_5) and video (creatify-aurora) models to /v1/models so clients can discover them. Filter on modalities to pick the ones you need:

bash

curl -s 'http://localhost:8080/v1/models?provider=elevenlabs&include=modalities' \
  | jq '.data[] | select(.modalities.output == ["audio"]) | .id'

pricing and context_window are null for every ElevenLabs model: billing is credit-based and there is no token window.

Supported Providers ​

Available Providers ​

Vision/Multimodal Support ​

Providers with Vision Support ​

Example Vision Request ​

Using Providers ​

Provider Configuration ​

API Endpoints ​

1. Unified Generate API ​

2. Provider Proxy ​

Provider-Specific Examples ​

OpenAI Provider ​

DeepSeek Provider ​

Anthropic Provider ​

Cohere Provider ​

Groq Provider ​

Cloudflare Provider ​

Ollama Provider ​

Ollama Cloud Provider ​

Google Provider ​

Mistral Provider ​

Moonshot Provider ​

NVIDIA Provider ​

llama.cpp Provider ​

Z-AI Provider ​

ElevenLabs Provider ​

Supported Providers

Available Providers

Vision/Multimodal Support

Providers with Vision Support

Example Vision Request

Using Providers

Provider Configuration

API Endpoints

1. Unified Generate API

2. Provider Proxy

Provider-Specific Examples

OpenAI Provider

DeepSeek Provider

Anthropic Provider

Cohere Provider

Groq Provider

Cloudflare Provider

Ollama Provider

Ollama Cloud Provider

Google Provider

Mistral Provider

Moonshot Provider

NVIDIA Provider

llama.cpp Provider

Z-AI Provider

ElevenLabs Provider