> ## Documentation Index
> Fetch the complete documentation index at: https://docs.polpo.sh/llms.txt
> Use this file to discover all available pages before exploring further.

# LLM & Providers

> Complete list of supported LLM providers, model format, and custom provider configuration.

## Model format

Models use the `provider/model` syntax:

```
anthropic/claude-sonnet-4-5
openai/gpt-4o
openrouter/google/gemini-2.5-flash
```

## Built-in providers

| Provider           | Slug                     | Env variable           | Model example                                   |
| ------------------ | ------------------------ | ---------------------- | ----------------------------------------------- |
| Anthropic          | `anthropic`              | `ANTHROPIC_API_KEY`    | `anthropic/claude-sonnet-4-5`                   |
| OpenAI             | `openai`                 | `OPENAI_API_KEY`       | `openai/gpt-4o`                                 |
| Google             | `google`                 | `GEMINI_API_KEY`       | `google/gemini-2.5-pro`                         |
| xAI                | `xai`                    | `XAI_API_KEY`          | `xai/grok-3-mini`                               |
| OpenRouter         | `openrouter`             | `OPENROUTER_API_KEY`   | `openrouter/anthropic/claude-sonnet-4-5`        |
| Groq               | `groq`                   | `GROQ_API_KEY`         | `groq/llama-3.3-70b`                            |
| Mistral            | `mistral`                | `MISTRAL_API_KEY`      | `mistral/mistral-large`                         |
| Cerebras           | `cerebras`               | `CEREBRAS_API_KEY`     | `cerebras/llama3.1-70b`                         |
| MiniMax            | `minimax`                | `MINIMAX_API_KEY`      | `minimax/minimax-m2.5`                          |
| MiniMax (China)    | `minimax-cn`             | `MINIMAX_CN_API_KEY`   | `minimax-cn/minimax-m2.5`                       |
| Amazon Bedrock     | `amazon-bedrock`         | `AWS_ACCESS_KEY_ID`    | `amazon-bedrock/anthropic.claude-v2`            |
| Azure OpenAI       | `azure-openai-responses` | `AZURE_OPENAI_API_KEY` | `azure-openai-responses/gpt-4o`                 |
| Google Vertex      | `google-vertex`          | `GOOGLE_CLOUD_PROJECT` | `google-vertex/gemini-pro`                      |
| Hugging Face       | `huggingface`            | `HF_TOKEN`             | `huggingface/meta-llama/Llama-3-70b`            |
| Vercel AI Gateway  | `vercel-ai-gateway`      | `AI_GATEWAY_API_KEY`   | `vercel-ai-gateway/anthropic/claude-sonnet-4-5` |
| z.ai               | `zai`                    | `ZAI_API_KEY`          | `zai/zai-default`                               |
| Kimi (Moonshot)    | `kimi-coding`            | `KIMI_API_KEY`         | `kimi-coding/kimi-k2.5`                         |
| GitHub Copilot     | `github-copilot`         | `COPILOT_GITHUB_TOKEN` | `github-copilot/gpt-4o`                         |
| OpenAI Codex       | `openai-codex`           | `OPENAI_API_KEY`       | `openai-codex/codex-mini`                       |
| Google Gemini CLI  | `google-gemini-cli`      | `GEMINI_API_KEY`       | `google-gemini-cli/gemini-2.5-pro`              |
| Google Antigravity | `google-antigravity`     | `GEMINI_API_KEY`       | `google-antigravity/gemini-2.5-pro`             |
| OpenCode           | `opencode`               | `OPENCODE_API_KEY`     | `opencode/opencode-default`                     |
| OpenCode (Go)      | `opencode-go`            | `OPENCODE_API_KEY`     | `opencode-go/opencode-default`                  |

## Auto-inference

If you don't specify a provider, Polpo infers it from the model name:

| Prefix                                | Provider  |
| ------------------------------------- | --------- |
| `claude-`                             | anthropic |
| `gpt-`, `chatgpt-`, `codex-`          | openai    |
| `o1-`, `o3-`, `o4-`                   | openai    |
| `gemini-`                             | google    |
| `grok-`                               | xai       |
| `mistral-`, `codestral-`, `devstral-` | mistral   |
| `llama-`, `llama3`                    | groq      |

```text theme={null}
"model": "claude-sonnet-4-5"
```

is equivalent to:

```text theme={null}
"model": "anthropic/claude-sonnet-4-5"
```

## Custom providers

Point to any OpenAI-compatible endpoint by adding a provider in `.polpo/polpo.json`:

```json theme={null}
{
  "providers": {
    "my-provider": {
      "baseUrl": "https://my-endpoint.com/v1",
      "api": "openai-completions"
    }
  }
}
```

Then use: `"model": "my-provider/model-name"`

### API modes

| Mode                 | Description                           |
| -------------------- | ------------------------------------- |
| `openai-completions` | OpenAI Chat Completions API (default) |
| `openai-responses`   | OpenAI Responses API                  |
| `anthropic-messages` | Anthropic Messages API                |

### Custom model definitions

Define model metadata for accurate context window and cost tracking:

```json theme={null}
{
  "providers": {
    "my-provider": {
      "baseUrl": "https://my-endpoint.com/v1",
      "api": "openai-completions",
      "models": [
        {
          "id": "my-model",
          "name": "My Custom Model",
          "contextWindow": 128000,
          "maxTokens": 16384,
          "reasoning": true
        }
      ]
    }
  }
}
```

## Model Gateways

Polpo can route all model calls through a gateway configured in `settings.gateway` (see [Model Gateway](/docs/platform/llm-gateway)). You can also register gateways as custom providers for model-level routing:

### LiteLLM

```json theme={null}
{
  "providers": {
    "litellm": {
      "baseUrl": "https://my-litellm-proxy.com/v1",
      "api": "openai-completions"
    }
  }
}
```

```text theme={null}
"model": "litellm/gpt-4o"
"model": "litellm/claude-sonnet-4-5"
```

LiteLLM routes to 100+ providers through a single proxy. The model ID follows LiteLLM's naming convention.

### Vercel AI Gateway

```json theme={null}
{
  "providers": {
    "vercel": {
      "baseUrl": "https://gateway.ai.vercel.app/v1",
      "api": "openai-completions"
    }
  }
}
```

```text theme={null}
"model": "vercel/anthropic/claude-sonnet-4-5"
"model": "vercel/google/gemini-2.5-pro"
```

The Vercel AI Gateway provides automatic failover, cost tracking, and caching across providers.

### Ollama (local)

```json theme={null}
{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "api": "openai-completions"
    }
  }
}
```

```text theme={null}
"model": "ollama/llama3.1:70b"
```

### vLLM

```json theme={null}
{
  "providers": {
    "vllm": {
      "baseUrl": "https://my-vllm-server.com/v1",
      "api": "openai-completions"
    }
  }
}
```

The model ID is passed as-is to the gateway or server — no need to pre-define every model in the config.