Qwen Locally for Code: A Private Coding Assistant Instead of Copilot
How to run Qwen as a local coding assistant — free, private, and cloud-free. A local model that doesn't just write code, but edits files and runs terminal commands too.
Qwen is one of the strongest open models, and it's good at code. But on its own, a model only generates text: to get an actual coding assistant, it needs hands — access to files and the terminal. Here's how to put together a "Qwen + agent" combination locally, for free, and privately.
Why Qwen locally for code
- Privacy. Code doesn't leave for someone else's cloud — important for work and confidential projects.
- Free, no limits. No tokens, no subscription, unlike cloud assistants.
- No foreign card, no VPN needed. Everything runs on your own machine.
The model is only half of it
A bare Qwen in a terminal answers in text but won't touch your project. A real assistant needs to read files, edit them, and run commands — meaning it needs to be an agent. So the combination looks like this: a local Qwen model plus an agent that gives it access to the environment.
Doka already has this put together. The built-in Qwen3.5 downloads automatically for your available memory, and the agent around it works with files and the terminal.
Need a model tuned specifically for code? Load your own .gguf (Qwen Coder, for
instance) through the model manager — or connect it from Ollama / LM Studio over
API. Doka isn't locked to one model.
How to set it up, step by step
- Download Doka and install it. On first launch, it installs Qwen automatically.
- Want the code-specific model? Open the "Models" screen and load your Qwen
Coder
.gguffile, or connect a model from Ollama/LM Studio in the "API" tab. - Pick a working folder with your project above the chat box.
- Give it a task in plain language: "find out why this test is failing and fix it." The agent reads the files itself, makes edits, and runs terminal commands.
How to pick a model size
The model runs in RAM, so look at that: the more you have, the bigger and smarter a model you can run. More on picking a model is in the documentation. For code on a local machine, what matters more than "maximum size" is stability and speed on your own hardware.
The bottom line
Qwen on its own just writes text. Paired with an agent, it becomes a coding assistant that edits files in your project and runs tests on its own — free and cloud-free. You can put this combination together in a couple of minutes after installing Doka.