> ## Documentation Index
> Fetch the complete documentation index at: https://machine-path.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Choose a model provider

> Route Navi through a direct model provider or an intentional proxy.

Navi uses the
[Mastra Model Router](https://mastra.ai/blog/model-router) to turn a
`provider/model` ID into the corresponding adapter from
[AI SDK providers](https://ai-sdk.dev/providers/ai-sdk-providers). Set the
provider's API key and choose the model with `NAVI_MODEL`.

Navi does not maintain a runtime allowlist. Any model the installed Mastra
router supports can be selected when its provider key is available. The model
also needs the capabilities used by the command: text generation for a
repository answer, and tool calling or structured output for some flows.

## Use a direct provider

The table below mirrors Navi's current tested targets in
[`config/tested-models.json`](https://github.com/machinepathindustries/navi/blob/main/config/tested-models.json).
Quick targets cover repository questions. Workflow targets cover flows
that use tools or structured results.

| Lab                                                                                     | API key                        | Quick target                                                                                              | Workflow target                                                                                           |
| --------------------------------------------------------------------------------------- | ------------------------------ | --------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------- |
| [DeepSeek](https://api-docs.deepseek.com/api/list-models)                               | `DEEPSEEK_API_KEY`             | [`deepseek/deepseek-v4-flash`](https://api-docs.deepseek.com/api/list-models)                             | [`deepseek/deepseek-v4-pro`](https://api-docs.deepseek.com/api/list-models)                               |
| [OpenAI](https://developers.openai.com/api/docs/models)                                 | `OPENAI_API_KEY`               | [`openai/gpt-5.6-luna`](https://developers.openai.com/api/docs/models/gpt-5.6-luna)                       | [`openai/gpt-5.6-terra`](https://developers.openai.com/api/docs/models/gpt-5.6-terra)                     |
| [Anthropic](https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5) | `ANTHROPIC_API_KEY`            | [`anthropic/claude-sonnet-5`](https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5) | [`anthropic/claude-sonnet-5`](https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5) |
| [Google](https://ai.google.dev/gemini-api/docs/models)                                  | `GOOGLE_GENERATIVE_AI_API_KEY` | [`google/gemini-3.5-flash-lite`](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite)      | [`google/gemini-3.5-flash-lite`](https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite)      |
| [xAI](https://docs.x.ai/developers/models/grok-4.5)                                     | `XAI_API_KEY`                  | [`xai/grok-4.5`](https://docs.x.ai/developers/models/grok-4.5)                                            | [`xai/grok-4.5`](https://docs.x.ai/developers/models/grok-4.5)                                            |

Start with the default DeepSeek quick target:

```bash theme={null}
export DEEPSEEK_API_KEY="<your-key>"
export NAVI_MODEL="deepseek/deepseek-v4-flash"
npx --no-install navi-cli "Where is retry behavior configured?"
```

To use another direct provider, set the API key and quick target from the same
row. Use the workflow target when you want the tested selection for a flow:

```bash theme={null}
export NAVI_MODEL="deepseek/deepseek-v4-pro"
npx --no-install navi-cli run founder "Should this change ship?"
```

You can put these variables in the project `.env` file instead of exporting
them. A variable already set in the shell takes precedence.

## Use OpenRouter only when you want a proxy

OpenRouter is a separate route with its own key, billing account, and model ID.
Choose it only when you intend to proxy the request instead of calling a model
vendor's direct gateway.

| Route                                      | API key              | Quick target                                                                  | Workflow target                                                                 |
| ------------------------------------------ | -------------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------------- |
| [OpenRouter](https://openrouter.ai/models) | `OPENROUTER_API_KEY` | [`openrouter/openai/gpt-5.6-luna`](https://openrouter.ai/openai/gpt-5.6-luna) | [`openrouter/openai/gpt-5.6-terra`](https://openrouter.ai/openai/gpt-5.6-terra) |

```bash theme={null}
export OPENROUTER_API_KEY="<your-key>"
export NAVI_MODEL="openrouter/openai/gpt-5.6-luna"
npx --no-install navi-cli "Where is retry behavior configured?"
```

Keep the `openrouter/` prefix on this route. It is not interchangeable with the
direct OpenAI ID.

The model manifest is the machine-readable source for Navi's compatibility
test targets; it does not restrict what the router can run. Use
[Troubleshooting](/troubleshooting) when a key or model selection is rejected.