Add a model provider in Mandri

A provider gives Mandri access to models. You can connect a remote service such as OpenRouter, OpenAI, Anthropic or Gemini, or a local server such as LM Studio. The tool that runs your agent, such as Codex or Pi, is a separate choice.

Before you start

For an API provider, have its API key ready. API access and a chat subscription are different: a ChatGPT subscription does not provide OpenAI API credit. Model availability and charges depend on the provider. Remote providers receive the requests sent to their models.

For a local provider, start its model server and note the server address. The LM Studio guide covers that setup.

If your tool uses its own model configuration, follow that tool’s authentication instructions. This guide covers adding a provider in Mandri’s settings.

Add the connection

These steps cover API keys and local servers. If you select ChatGPT (subscription), Mandri offers a browser sign-in instead. Follow Use your ChatGPT subscription with Mandri for that setup.

1. In Mandri, open Settings, then Providers.
2. Select Add provider.
3. In Name, enter a name you will recognize, such as openrouter-personal. It must be different from your other provider names in Mandri. It is not your account name at the provider.
4. In Kind, select the service you want to connect.
5. For an API provider, paste its key into API key. For a local server, enter its address in Base URL. Complete any other required fields shown for your choice.
6. Select Save and verify.

Mandri checks access to the provider’s model catalog. A successful check confirms this connection; it does not guarantee that every listed model will work with every tool.

For a complete example using an API key, follow the OpenRouter guide.

Choose a model for a task

The model catalog in Settings lets you inspect what the provider exposes. To use a model, return to the task or session and open its model menu. Choose a model there, alongside the tool and permission settings.

Continue with your first task for the full sequence.

If verification fails

• Authentication error: check that the key belongs to the selected provider, is complete and has not expired or been revoked.
• Rate limit: wait before trying again and check your provider’s account limits.
• Local server unreachable: make sure the server is running and that Base URL matches its address, including the port.
• Model missing or unsupported: check the provider’s catalog and the selected tool. A model’s presence in the catalog does not establish compatibility with that tool.