Setup
- Open the extension provider settings and select OpenAI Compatible rather than the official OpenAI account flow.
- Enter the TokenReply base URL ending in /v1 and a TokenReply API key.
- Paste an exact model ID from the live catalog.
- Use the built-in verification action, then run a read-only task before allowing edits or commands.
Provider: OpenAI Compatible
Base URL: https://api.tokenreply.com/v1
API key: sk-your-token-here
Model ID: YOUR_MODEL_ID
How to verify
- The provider verification action succeeds with the saved base URL and key.
- A simple prompt returns text before native tools are enabled.
- The chosen model completes a harmless native tool call before repository work begins.
Limitations
- Roo Code requires native OpenAI-style tool calling for agent tools; a chat-only model is not sufficient.
- Cline and Roo settings labels can differ by extension version.
- Manually entered context and price metadata in a client is not TokenReply billing authority.
Troubleshooting
| Symptom | What it usually means | What to do |
|---|
| Chat works but file tools fail | The selected model or route does not support native OpenAI tool calls. | Switch to a catalog model whose tool support you have verified with the same endpoint. |
|---|
| Model not found during provider verification | The client contains a friendly display name instead of the exact catalog ID. | Copy the raw model ID from the TokenReply models page. |
|---|
