Cline¶
Cline works in many small steps, and each one is a model call. Point it at the gateway and every step is governed: checked against your policies, charged to your team and recorded in Request logs, with limits on the key to keep an autonomous task's spend in bounds.
What it is¶
Cline is a coding agent for VS Code (and editors built on it). It reads files, edits them and runs commands, asking a model what to do next at each step. It calls the model from your machine through the API Provider you choose; two of them can point at the gateway.
The two settings¶
Open Cline's settings and choose an API Provider:
| Field | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://<your-gateway>/v1 |
| API Key | Your Vidai key (starts vk_) |
| Model | A model name such as gpt-4.1, claude-sonnet-4.5 or gemini-3.5-flash |
Under Model Configuration, set the Context Window size and Max Output Tokens to the model's own, and tick image support for models that read images.
| Field | Value |
|---|---|
| API Provider | Anthropic |
| Anthropic API Key | Your Vidai key (starts vk_) |
| Use custom base URL (tick it) | https://<your-gateway> (no /v1: Cline adds /v1/messages) |
| Model | A Claude model your organisation approved |
The OpenAI Compatible provider sends the key as Authorization: Bearer, the Anthropic one as x-api-key; the gateway
accepts both. The gateway's address is under Connect your app on the Keys page and in System › Settings › Gateway;
the key comes from Keys › Issue a key (its secret is shown once).
A minimal call¶
- Fill in one of the forms above.
- Start a new task with "Reply with exactly: ok".
- A short answer means the address and key are right.
Streaming¶
Cline streams every step. Through the OpenAI Compatible provider, a stream from Claude or Gemini arrives as OpenAI chunks; through the Anthropic provider, in Anthropic's own events.
Tool calls¶
Cline's steps (read a file, write a file, run a command) are tool calls. They work across providers: an OpenAI-format tool call reaches Claude or Gemini, and the answer comes back in the shape Cline expects. Use a model that is good at tool use; small models tend to stall on long tasks.
Reading what the gateway did¶
Cline doesn't show response headers. Look the calls up in Request logs: filter on Key, Model or Path
(/v1/chat/completions or /v1/messages). One task usually shows as many calls, one per step.
Errors¶
Cline shows the gateway's error message on the step that failed and offers to retry. Common ones: 401
invalid_api_key (wrong key, or a provider's key), 403 model_not_allowed, 404 model_not_found (check the model
name), 403 content_guardrail_blocked (a guardrail stopped the content), 429 (a limit on the key or team). The full
list is in Headers and errors.
Model names¶
Use the names your organisation approved under Models in the Control room. With the Anthropic provider, Cline's model list shows Anthropic's own names; pick the one that matches a name your organisation set up. You can't type another name there: if your organisation's names differ from Anthropic's ids, use the OpenAI Compatible provider instead.
What goes through the gateway¶
| Cline feature | Through the gateway |
|---|---|
| Plan and Act modes, every step of a task | Yes |
| Prompt caching on Claude models (Anthropic provider) | Yes; on other providers the cache markers are removed and savings are still reported |
| MCP servers you add to Cline | The model calls that use their tools, yes; the MCP servers themselves are reached directly by Cline |
Check it works in a minute¶
Start a task with "Reply with exactly: ok", then open Request logs a few seconds later and filter on your key.
Good to know¶
- Autonomous tasks are chatty: one task can be dozens of calls. Set a limit on the key (or the team's budget) so a runaway task stops at a known cost.
- Issue one key per developer; give a shared automation its own key so it shows as its own line in cost reports.
- Cline's docs: OpenAI Compatible, Anthropic.