DocsCline
Connect Cline’s OpenAI Compatible provider to Runtoken.
Prerequisites#
Before you start, prepare:
- An installed Cline IDE extension and a test workspace.
- A Runtoken platform API key.
- A Chat-compatible model ID with the capabilities your task requires.
Configure the provider#
Open Cline's settings with the gear button and choose the provider. Fill in the connection values below.
| Setting | Value | Purpose |
|---|---|---|
| API Provider | OpenAI Compatible | Selects the custom compatible endpoint. |
| Base URL | https://api.runtoken.net/v1 | API prefix; do not append a chat-completion path. |
| API Key | YOUR_PLATFORM_KEY | Runtoken platform credential. |
| Model ID | YOUR_MODEL_ID | Exact Chat-compatible model identifier. |
Replace YOUR_PLATFORM_KEY and YOUR_MODEL_ID with your platform key and model ID from the catalogue. Keep the key in the credential field, not in task prompts or workspace files.
Use advanced model settings only when the values are established for your selected model.
| Model setting | How to choose it |
|---|---|
| Context window | Use the supported model limit; increasing the client value does not enlarge it. |
| Maximum output tokens | Stay within the model's output limit. |
| Image or computer-use capabilities | Enable only the capabilities supported by the model and route. |
Client-side price fields are not platform price configuration. Use Runtoken's request and billing records to check actual usage rather than relying on an editor estimate.
Check a request#
Save the settings, open a new task and send a brief prompt. Locate the corresponding Runtoken request by time, key and model. If the task still uses another provider, check the active task or mode selection before replacing credentials.
Start with Plan mode to discuss a small change, then move to Act when you are ready to edit.
Describe how you would add greeting.py with a greet(name) function.
It should return "Hello, " plus name, or "Hello, friend" for an empty name.
Do not install dependencies or use the network.Review the plan, switch to Act and ask Cline to implement it. Inspect the proposed tool operations and actual file. Follow up with a request to read the file and explain the empty-name behavior, checking that the tool result is handled successfully.
If your version offers separate model choices for Plan and Act, check both. Testing Plan alone does not establish which provider is used after switching modes.
Tools and permissions#
Cline combines model replies with local tool operations. A successful text request does not establish that every tool or input type is supported.
| Workflow | What must work |
|---|---|
| Planning | Model replies and the intended context are available. |
| File modification | Compatible tool calls, returned results and workspace permissions. |
| Shell or browser operations | Both client permission and the selected model's required capabilities. |
| Long tasks | Context handling, model limits and gateway request support. |
Keep normal tool approvals during setup. Do not enable broad auto-approval to diagnose an API failure. Runtoken does not convert protocols or grant local filesystem permissions.
Troubleshooting#
Read the error body and inspect the active provider before changing advanced options.
| Symptom | What to check |
|---|---|
401 | Platform key, credential field and inherited task settings. |
404 or rejected model | Base URL, exact model ID and selected protocol. |
| Plan works but Act fails | Act's active model and the rejected tool request. |
| Context or output-limit error | The actual model limit and Cline's configured values. |
| Cost estimate differs | Compare platform usage records; client estimates are not settlement. |
429 | The specific limit and reset information returned by Runtoken. |
Restore and next steps#
Select the previous provider and model in Cline settings to restore your earlier workflow. Remove only the credential and endpoint entries you added.
See OpenAI Chat, Cline's provider settings and Plan & Act.