Skip to main content
The Ollama API Proxy uses a models.json file to define which models are available and how they map to AI providers. This allows you to customize model names, add new models, and control which provider each model uses.

​
Configuration File

Create a models.json file in your project root (the same directory where you run the proxy). If this file doesn’t exist, the proxy will use built-in defaults (src/index.js:128-149).
The proxy automatically loads models.json from the current working directory at startup (src/index.js:124).

​
File Format

The models.json file is a JSON object where each key is a model name and each value contains the provider and model identifier.

​
Structure

{
  "model-name": {
    "provider": "provider-name",
    "model": "actual-model-id"
  }
}
​
model-name
string
The name you’ll use when making API requests. This can be any string you choose.
​
provider
string
required
The AI provider to use. Must be one of:
  • openai - Uses OpenAI API (requires OPENAI_API_KEY)
  • google - Uses Google Gemini API (requires GEMINI_API_KEY)
  • openrouter - Uses OpenRouter API (requires OPENROUTER_API_KEY)
​
model
string
required
The actual model identifier used by the provider.
  • For OpenAI: gpt-4o, gpt-4o-mini, gpt-4.1-mini, etc.
  • For Google: gemini-2.5-flash, gemini-2.5-flash-lite, etc.
  • For OpenRouter: deepseek/deepseek-r1-0528:free, moonshotai/kimi-k2:free, etc.

​
Example Configuration

Here’s the actual models.json from the repository: 
{
    "gpt-4o-mini": {
        "provider": "openai",
        "model": "gpt-4o-mini"
    },
    "gpt-4.1-mini": {
        "provider": "openai",
        "model": "gpt-4.1-mini"
    },
    "gpt-4.1-nano": {
        "provider": "openai",
        "model": "gpt-4.1-nano"
    },
    "gpt-4o": {
        "provider": "openai",
        "model": "gpt-4o"
    },
    "gpt-5-nano": {
        "provider": "openai",
        "model": "gpt-5-nano"
    },    
    "gemini-2.5-flash": {
        "provider": "google",
        "model": "gemini-2.5-flash"
    },
    "gemini-2.5-flash-lite": {
        "provider": "google",
        "model": "gemini-2.5-flash-lite"
    },
    "deepseek-r1": {
        "provider": "openrouter",
        "model": "deepseek/deepseek-r1-0528:free"
    },
    "kimi-k2": {
        "provider": "openrouter",
        "model": "moonshotai/kimi-k2:free"
    }
}

​
Built-in Defaults

If no models.json file exists, the proxy uses these built-in models (src/index.js:133-143): 
{
    "gpt-4o-mini": { "provider": "openai", "model": "gpt-4o-mini" },
    "gpt-4.1-mini": { "provider": "openai", "model": "gpt-4.1-mini" },
    "gpt-4.1-nano": { "provider": "openai", "model": "gpt-4.1-nano" },
    "gpt-4o": { "provider": "openai", "model": "gpt-4o" },
    "gemini-2.5-flash": { "provider": "google", "model": "gemini-2.5-flash" },
    "gemini-2.5-flash-lite": { "provider": "google", "model": "gemini-2.5-flash-lite" },
    "deepseek-r1": { "provider": "openrouter", "model": "deepseek/deepseek-r1-0528:free" }
}
When using built-in models, you’ll see: ℹ️ Using built-in models. Create a models.json file to customize.

​
Model Validation

When you make an API request, the proxy validates that:
  1. The model name exists in models.json (src/index.js:188-189)
  2. The required provider is available (has an API key set) (src/index.js:191-193)
If a model is defined but its provider isn’t configured, you’ll get an error:
{
  "error": "Provider openai not available"
}
Make sure the corresponding API key environment variable is set.

​
Listing Available Models

You can query which models are currently available using the /api/tags endpoint: 
curl http://localhost:11434/api/tags
This returns only models whose providers have valid API keys configured (src/index.js:494-504). 
{
  "models": [
    {
      "name": "gpt-4o-mini",
      "model": "gpt-4o-mini",
      "modified_at": "2026-03-11T10:30:00.000Z",
      "size": 1000000000,
      "digest": "sha256:gpt4omini"
    }
  ]
}

​
Adding New Models

To add a new model, simply add an entry to models.json and restart the proxy:

​
OpenAI Model

{
  "gpt-4-turbo": {
    "provider": "openai",
    "model": "gpt-4-turbo-preview"
  }
}

​
OpenRouter Model

{
  "claude-opus": {
    "provider": "openrouter",
    "model": "anthropic/claude-3-opus"
  }
}

​
Gemini Model

{
  "gemini-pro": {
    "provider": "google",
    "model": "gemini-1.5-pro"
  }
}
Make sure the model ID matches the actual model identifier from the provider’s documentation.

​
Startup Output

When the proxy loads your models.json successfully, you’ll see: 
✅ Loaded models from /path/to/your/models.json
🚀 Ollama Proxy with Streaming running on http://localhost:11434
🔑 Providers: openai, google, openrouter
📋 Available models: gpt-4o-mini, gemini-2.5-flash, deepseek-r1
If there’s an error loading the file: 
❌ Error loading models.json: Unexpected token in JSON at position 42
JSON syntax errors in models.json will cause the proxy to exit immediately (src/index.js:146-148).

Build docs developers (and LLMs) love

Get started for freeTalk to us