Cline + JuiceFactory
Autonomous coding agent for VS Code (formerly Claude Dev)
Cline is an autonomous coding agent that can read, write, and execute code in your workspace, with explicit human approval for every action. It supports OpenAI-compatible providers, so JuiceFactory drops in as the inference backend — keeping the agent's view of your codebase entirely in EU infrastructure.
Why route Cline at JuiceFactory
Cline reads many files at full content — its context windows on a real task can hit 100k+ tokens of source code. Sending all of that to a US provider on every iteration is the worst kind of GDPR exposure. JuiceFactory keeps the agent's entire reasoning loop in the EU, with stateless inference so nothing is retained between sessions.
Before you start
- Cline extension installed (VS Code Marketplace, search "Cline")
- A JuiceFactory API key
- A model with strong tool-use support — qwen3-vl or qwen3-coder-32b recommended
Setup
- 1
Open Cline's settings
Click the Cline icon in the VS Code sidebar, then the gear icon at the top of the panel.
- 2
Choose API Provider: OpenAI Compatible
In the provider dropdown, select "OpenAI Compatible". This unlocks the apiBase field — providers like Anthropic and OpenAI proper are hardcoded and would not allow a custom URL.
- 3
Fill in JuiceFactory base URL, key, and model ID
Base URL: https://api.juicefactory.ai/v1 — API Key: your jf-... key — Model ID: qwen3-vl (or another model that supports tool use).
- 4
Verify with a small task
Ask Cline something low-stakes first ("read this file and summarise it"). Cline shows the request/response in its log — confirm requests go to api.juicefactory.ai before giving it write/execute access.
Configuration
VS Code settings.json (optional explicit override)
{
"cline.apiProvider": "openai-compatible",
"cline.openAiBaseUrl": "https://api.juicefactory.ai/v1",
"cline.apiModelId": "qwen3-vl",
"cline.openAiApiKey": "jf-..."
}Recommended models
| Use case | Model | Why |
|---|---|---|
| General agent work (read/write/execute) | qwen3-vl | Strong tool-use, large context window, handles multi-file edits reliably. |
| Heavy code generation | qwen3-coder-32b | Coder-tuned variant — fewer hallucinated APIs in less common languages. |
Troubleshooting
Cline keeps asking the same clarifying question
Cline's loop assumes the model emits a tool_calls field on certain messages. Some open-weight serving stacks return assistant text instead. Make sure you're on a JuiceFactory model marked tool-use-supported (the model list in the dashboard shows this).
Long tasks fail mid-way with rate-limit errors
Cline runs many small calls per task. Bump your plan limits or enable burst credits — the dashboard shows your current per-minute and per-day caps.
Token costs are higher than expected
Cline re-sends the entire conversation each step. Use Cline's "checkpoint" feature to fork off shorter contexts, and prefer qwen3-coder-32b over qwen3-vl when vision isn't needed.
FAQ
Is Cline's file access restricted to my workspace?▾
Yes. Cline only sees files inside the open VS Code workspace (and asks you before reading new ones). The data sent to JuiceFactory is whatever Cline includes in its context — it does not exfiltrate beyond that.
Can I run Cline against a self-hosted JuiceFactory instance?▾
Yes. The "OpenAI Compatible" provider takes any URL. Point it at your private endpoint and Cline will route through your own EU GPU pool.
How does Cline differ from Continue's agent mode?▾
Cline is autonomous-first: it plans multi-step actions and only stops at approval gates. Continue is editor-first with optional agent mode. For greenfield code generation, Cline; for inline edits and chat, Continue.
Set up in 2 minutes
Create an account and use the included starter credit to verify Cline end-to-end.