Connect a Model Provider

Captain Who does not include a model account that is ready to use. You must supply an API URL, Token, and model identifier that you are authorized to use. The selected service provider determines the associated cost, data processing, and availability.

Current adapter types

“Compatible” means that the protocol shape meets the relevant adapter's requirements. It does not mean that Captain Who certifies every gateway, proxy, or service-provider version.

Configure the global connection

1. Open “Settings → Configuration.”
2. Enter the API URL and API Token in model settings.
3. Enable the models you need under “Available Models.”
4. Open “Manage Models” to add or edit model metadata.

Models without their own overrides inherit the global URL and Token. A model-level connection must provide both its URL and Token; the application will not combine a model-level URL with the global Token.

Add a model

Enter the model identifier, display name, and context window. The model identifier is sent to the Provider unchanged; the display name is only used in the interface.

Advanced options include:

• a model-level API URL and Token;
• API vendor/compatibility configuration;
• whether image input is supported;
• estimated per-thousand-Token prices for input, cached input, and output.

Prices are used only for local estimates. They are not the service provider's bill, and currency is not identified automatically. The context window and image support are also configured by the user; incorrect values may cause requests to fail or capacity estimates to be inaccurate.

When DeepSeek V4 Chat is selected, you can also set the reasoning mode and reasoning effort. Provider-native private reasoning continuation is not exposed as ordinary chat text and cannot be migrated losslessly to every other Provider.

Verify the configuration

After saving, begin with a low-risk test:

1. Create a new conversation and select the model.
2. Send a short question and confirm normal streamed output.
3. Request one read-only Tool invocation and confirm that Tool Call and Tool Result complete the loop.
4. Then test declared capabilities such as image input or a longer context.
5. Check the request and billing against the service provider's dashboard.

A complete configuration means only that a connection can be attempted. It does not verify the remote Token, model name, or network. When switching Providers, the application may need to adapt or reject a direct switch if the current conversation contains continuation state specific to another Provider.

Data and credentials

Model requests may contain the current conversation, system instructions, file contents read for the task, and necessary projections of Tool results. Connect only to Providers you trust and whose data policies you have read.

Signed macOS releases store the model Token in the system keychain, while the local database retains only an opaque reference. Reading Settings never returns the stored Token; replacing it requires entering a new value. Do not share screenshots containing Tokens or credential-diagnostic material. See Data and Permissions for details.