Cline

Use the Cline extension in VS Code with Nebul models through the OpenAI Compatible provider.

Cline is an open-source AI coding agent for VS Code: plan and act modes, file edits, terminal use, and browser control. It supports any OpenAI-compatible endpoint as a provider, so it runs directly on your Nebul project's models.

Prerequisites

Visual Studio Code or a compatible editor
A Nebul AI Studio account with API access
An API key for your active project
A model ID from the Model Catalog. Pick one that supports tool calling; Cline needs it.

Installation

Install Cline from the VS Code Marketplace, then open it from the sidebar icon.

Connect it to Nebul

Open Cline and click the ⚙️ settings icon.
Set API Provider to OpenAI Compatible.
Fill in:

Setting	Value
Base URL	`https://api.inference.nebul.io/v1`
API Key	your Nebul key (`sk-...`)
Model ID	a model from the Model Catalog, for example `zai-org/GLM-5.3`

Under Model Configuration, set values that match the model's actual limits. Take them from /v1/model/info:

Setting	Where it comes from
Context Window	`max_input_tokens` (1048576 for `zai-org/GLM-5.3`)
Max Output Tokens	the model's output limit
Supports Images	`supports_vision`
Supports Computer Use	enable only for tool-calling models
Input/Output Price	`input_cost_per_1m_tokens` / `output_cost_per_1m_tokens`, which makes Cline's cost display accurate

Save, then switch to Act mode to start a task.

Cline verifies the connection when you save the provider. A failed check almost always means the base URL or the model ID is off.

Roo Code

Roo Code, a popular Cline fork, uses the same OpenAI Compatible provider flow with the same field names. The steps above apply unchanged.

Troubleshooting

Invalid API key: the key must come from the AI Studio project whose models you are using, and must not have been revoked.
Model not found: the model ID must exactly match the catalog ID, including vendor prefix and casing.
Connection errors: confirm the base URL is https://api.inference.nebul.io/v1 (with /v1) and that your network can reach api.inference.nebul.io.
Tasks stall or produce malformed edits: try a stronger model. Agent-style editing is sensitive to model capability; the reasoning models in the catalog handle Cline's prompts best.

Data and telemetry

Cline collects anonymous telemetry: feature usage, task metrics, and error events. It never includes code or conversation content. To turn it off, open Cline's settings and disable the Cline Telemetry toggle. See Cline's telemetry documentation for details.

Requests sent to Nebul are handled per Privacy & Security: no training on your data, EU-resident processing.