Roo Code + TOKNADO
How to use Roo Code with TOKNADO's OpenAI-compatible endpoint
Roo Code is a VS Code coding agent that runs in several modes (code, architect, debug and custom ones) and is model-agnostic: any provider with an OpenAI-compatible API works. TOKNADO qualifies.
The key decision is which provider type to pick, because Roo has two that look similar and use different API routes. This guide shows the right one for a gateway and how to use profiles to give each mode its own model.
Start free trial Jump to setup
What you need
- VS Code with Roo Code installed.
- Your TOKNADO endpoint and API key.
- One or more model IDs from your TOKNADO dashboard.
Setup: Roo Code with TOKNADO
Open settings and pick the right provider
Click the gear icon and set API Provider to OpenAI Compatible. This uses the chat completions route that gateways implement. The separate “OpenAI” provider targets the Responses route and is for OpenAI-style endpoints that support it.
Enter the connection details
Paste the three values. Pasting avoids the typo and stray-whitespace errors Roo's docs list as common failures.
API Provider: OpenAI Compatible Base URL : https://toknado.up.railway.app/v1 API Key : YOUR_TOKNADO_API_KEY Model : toknado/autoSet model limits explicitly
Open Model Configuration and set the context window and max output tokens. A compatible endpoint doesn't guarantee that Roo can discover complete limits.
Optional: one profile per mode
Create configuration profiles and assign them to modes. For example, a stronger model for architect mode and a faster one for code mode. Switching is a model-ID change per profile, not a new deployment.
Run a tool-call sanity check
Ask Roo to read a file in a scratch project. If it can read and edit, tool calling works end to end through your endpoint.
Notes for real-world use
- Several modes can mean several simultaneous models and many requests. Check the rolling hourly limit on your plan if you run orchestrated multi-step tasks.
- If a model misbehaves with tools, try it in a plain chat first to separate a model problem from a configuration problem.
- Roo's docs advise using the OpenAI Compatible provider if you hit routing issues with the OpenAI provider type.
Troubleshooting Roo Code + TOKNADO
| Symptom | What to check |
|---|---|
| Connection error | Verify the Base URL (including /v1) and that the endpoint is reachable from your machine. |
| “Model not found” | Use the exact ID from your TOKNADO dashboard. |
| Strange routing or 404s | You selected OpenAI instead of OpenAI Compatible. Switch the provider type. |
| Truncated or cut-off output | Set context and max output tokens under Model Configuration. |
| 401 / invalid key | Paste the key again with no spaces; confirm your plan or trial is active. |
FAQ
OpenAI or OpenAI Compatible, which should I choose?
For a gateway like TOKNADO, choose OpenAI Compatible. It uses chat completions; the OpenAI provider uses the Responses route.
Can different Roo Code modes use different models?
Yes. Use configuration profiles and assign a profile to each mode.
Do I need to set a context window?
It's strongly recommended for custom endpoints, since limits may not be discovered automatically.
Is the free trial enough to test Roo Code?
Yes. The three-day trial allows up to 250 requests per rolling hour.