Roo Code

Roo Code is a coding assistant for VS Code. It supports the OpenAI Compatible provider, which lets you fill in a custom Base URL, API Key, and Model ID.

Preparation

First, prepare the following in the Tokeness console:

Field	Value
Base URL	https://n.tokeness.dev/v1
API Key	A key created in the Tokeness console
Model ID	Full model name copied from the Tokeness model marketplace

Roo Code relies heavily on tool calls. For coding tasks, prefer models on Tokeness that are confirmed to support OpenAI tool calling.

Add a provider

Configure the following in VS Code:

1. Open Roo Code.
2. Go to settings.
3. Find API Provider.
4. Select OpenAI Compatible.
5. Set Base URL to https://n.tokeness.dev/v1.
6. Enter your Tokeness API key.
7. Enter the model name copied from the Tokeness model marketplace.
8. Save the configuration.

Field mapping:

Roo Code field	Tokeness value
API Provider	OpenAI Compatible
Base URL	https://n.tokeness.dev/v1
API Key	Tokeness API key
Model / Model ID	Tokeness model name

Model configuration

If Roo Code asks for model context window or output length, start with conservative values:

Setting	Notes
Context Window	Use the context size shown in the Tokeness model marketplace; when unsure, start with a smaller value
Max Output Tokens	For the first test, use 2048 or 4096
Image Support	Enable only if the model explicitly supports image input
Computer Use	Enable only if both the current model and environment support it

Do not enable all advanced capabilities at once. Get chat and simple file edits working first, then gradually enable more complex features.

Verification

Test in the following order:

1. Create a temporary file.
2. Ask Roo Code to explain a small piece of code.
3. Then ask it to modify the temporary file.
4. After each step, check the Tokeness usage logs.

If there are no requests in the Tokeness logs, Roo Code is not currently using the Tokeness provider. Go back to the settings page and check whether the provider was saved successfully.

Tool calling notes

Roo Code uses the native OpenAI tool-calling format. If a model does not support tool calling, it may answer ordinary questions but cannot reliably perform tasks like reading files, editing files, or running commands.

When you encounter tool-calling issues, make two checks first:

1. Can the same model return content through a regular Chat Completions request?
2. Does the model support OpenAI-compatible tool calls?

If the first check passes but the second fails, switch to a model that supports tool calling — do not keep adjusting the Base URL or key.

Common issues

Symptom	Fix
Invalid API Key	Re-copy the key and confirm there are no line breaks, spaces, or non-ASCII quotation marks
Model Not Found	Re-copy the full model name from the Tokeness model marketplace
Connection Error	Check that the Base URL is https://n.tokeness.dev/v1
Chat works but file operations fail	Switch to a model that supports tool calling
Output interrupted	Reduce the task size, or increase Max Output Tokens

References

• Roo Code OpenAI Compatible Provider