Zeltro › Qwen Code in Zeltro: low-cost and local AI coding on your own computer

Qwen Code in Zeltro

How Zeltro runs Qwen Code: an API key through OpenRouter or OpenAI, or a free local model in Ollama or LM Studio. The cheapest way we've built a whole app, set up for you.

Qwen Code is an open-source coding agent from the Qwen team at Alibaba, built around the Qwen Coder models. It has no subscription to sign in with: you give it a key and a model, and it's the low-cost and local choice in Zeltro.

Connect it with

Your planNo plan sign-in–

An API keyOpenRouter · OpenAI · any OpenAI-compatible server✓

A model on your computerOllama or LM Studio✓

Qwen Code Qwen team, Alibaba qwen Zeltro installs it github.com/QwenLM/qwen-code

Your project in Zeltro

• Runtime and database ready
• AGENTS.md to read
• Live preview

What Zeltro does for Qwen Code#

1. Installs it

   Choose Qwen Code when Zeltro first opens, or later in Settings → AI Profiles. Zeltro installs it for you.

2. Connects it

   Pick where the model comes from: OpenRouter, OpenAI, Ollama, LM Studio or another OpenAI-compatible server, plus a model name.

3. Hands it a ready project

   Every project already has its runtime, its database and a working .env, plus an AGENTS.md that tells the agent the project's address, database and commands.

4. Runs it beside your app

   Qwen Code works in a real terminal next to the live preview, and you write long requests in the message box. It doesn't tell the app when a turn is done yet, so keep an eye on its terminal.

The cheapest way we've built a whole app#

Qwen Code with Qwen3 Coder on OpenRouter built a whole Django app for about $0.87 in our test. Choose OpenRouter as the place the key is from, click Sign in with OpenRouter and the key is filled in for you, billed to your own OpenRouter account. You pay per request, with no subscription.

On a model on your own computer#

Nothing leaves your computer, and there's nothing to pay beyond your hardware. Install Ollama, pull a coding model, and choose Ollama in the app, or from a terminal:

ollama pull qwen3-coder:30b
zeltro ai-set --agent qwen --model qwen3-coder:30b \
  --api-base http://localhost:11434/v1 --api-key ollama

Two things matter. Raise Ollama's context length to at least 64k first (OLLAMA_CONTEXT_LENGTH=64000 ollama serve, or the slider in the Ollama app), because its default is far too short for an agent. And model size matters: about 30B on 24 GB of graphics memory is roughly the floor for building an app (qwen3-coder:30b is a 19 GB download). Smaller models answer confidently and get it wrong. LM Studio works the same way, at http://localhost:1234/v1. See cheap and local models.

From the terminal#

zeltro ai-set --agent qwen --model qwen/qwen3-coder-30b-a3b-instruct \
  --api-base https://openrouter.ai/api/v1 --api-key sk-or-...

Asking before each change. By default, the AI in each project works without stopping to ask first: the Works without asking switch in the project's panel is on. Switch it off and you approve each step in Qwen Code's terminal. Zeltro passes the choice to Qwen Code for each run (--yolo), so Qwen Code's own settings stay as they are. To change Qwen Code's own setting for every run, outside Zeltro too, run zeltro ai-unattended, which writes "autoAccept": true into ~/.qwen/settings.json.

Questions

Checked 2026-11-09. Other products change often — if something here is out of date, tell us and it gets fixed.