Connect Cline

Cline has its own provider settings. Configuring VS Code’s built-in chat does not configure Cline. This guide covers an existing Cline extension and the separate Cline CLI.

Have your service key, available balance and an exact model ID from our models endpoint ready. Choose a text model with streaming and tool-calling support for agent tasks.

Configure the editor extension

Open the Cline panel and click its settings gear. Enter:

Use the API base, without adding /chat/completions. This provider uses the OpenAI-compatible Chat Completions format. Enter the secret only in Cline’s key field.

In Model Configuration, use verified context and output limits for the selected model. Leave image and computer-use capabilities off for this text setup. Do not copy another model’s limits or price estimates. Save the settings, using Done or the equivalent control in your version. These fields are documented in Cline’s OpenAI-compatible setup.

If your installation keeps separate Plan and Act model settings, configure the mode you will use and inspect the other mode before switching. A different provider in one mode can produce requests outside this connection.

Configure Cline CLI separately

For an existing terminal installation, check its version and available auth options, then open the interactive setup:

cline --version
cline auth --help
cline auth

Choose the option to use your own API key, then OpenAI Compatible. Enter the same Base URL, service key and exact model ID in the interactive setup. Finish the wizard and launch a new CLI session:

cline

The CLI documents its own interactive auth flow and saved provider configuration. Do not assume credentials saved by the extension are present in a separate CLI profile or on another machine. The current CLI README and auth implementation describe this terminal route.

Verify the selected client

Start a new task in the extension or CLI you configured:

Reply with CONNECTION_OK. Do not use tools or change files.

Check the displayed provider/model, the reply and its matching service request record. Then try a small file-reading task with normal approvals to verify tool calls. Both prompts use API balance; saved settings alone do not test the connection.

If authentication fails, re-enter the service key. For a missing model, check its exact ID and access in /v1/models. For an endpoint error, keep /v1 in the Base URL and select OpenAI Compatible, rather than a vendor login or native provider. If only tool actions fail, verify the model’s tool-call format and the tool’s own configuration. See our API troubleshooting.