Cline
Wire: OpenAI. Base URL: https://sator-api.princep.org/v1. State: verified against Cline's provider settings.
Install#
Install the Cline extension in VS Code (or a compatible editor) from the marketplace.
Configure#
Open Cline's settings and fill in the provider fields:
| Field | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://sator-api.princep.org/v1 |
| API Key | your sk-sator-v1-… key |
| Model ID | an id from Models, for example deepseek-v4-flash |
Cline does not detect a model's limits for a custom provider, so set them:
| Field | Value |
|---|---|
| Context Window Size | the model's context window, e.g. 1000000 for deepseek-v4-flash |
| Max Output Tokens | the model's max output, e.g. 384000 |
| Supports Images | tick only if you send images; see Tool calling |
Cline fills its model dropdown from GET /v1/models, which needs no key, so the list appears as soon as the base URL is entered.
Verify#
Ask Cline for something trivial: Reply with exactly: ok. A working setup answers ok, streams, and shows token counts; the request appears in your dashboard.
Troubleshooting#
- 404 on every request — the base URL is missing its
/v1. - Model dropdown is empty — the base URL is wrong; the dropdown is filled from
{base URL}/models, which works without a key. - Responses stop early — Max Output Tokens is what Cline sends as
max_tokens; raise it to the model's max output. - Tool calls are mangled — Cline relies on native tool calling. Every Sator model supports it on this wire, and tool-call deltas are forwarded byte-for-byte; if you see this, check that the model id is a catalog id and not a name.