Noko AI - Chat for Qwen

A free, independent VS Code chat extension for local Qwen models. Built by Noko.

Ask coding questions, explain selected code, and work with project files from a chat panel inside your editor. Noko AI provides the chat interface and tool integration. Qwen provides the model responses.

Preview release: This build targets the existing Windows portable runner setup. It is not a standalone, install-and-chat product. A supported model and inference server must be installed separately. Development tools are experimental; model quality and speed depend on your model and hardware.

Features

• Local chat with workspace conversation history.
• Optional active-file or selected-code context.
• Screenshot attachments with a compatible vision model.
• Direct file edits, colored change previews, and native VS Code diffs.
• PHP and JavaScript syntax checks and focused runtime diagnostics.
• Local browser screenshots with a separate project login session.
• Frontend guidance for tables, dashboards and other interfaces.

Requirements and setup

This preview currently expects Windows, PowerShell, and the compatible portable runner containing start.ps1, model files and the llama.cpp runtime. The runner is not included in this extension.

1. Configure ediQwen.installDirectory to point to your portable runner directory.
2. Configure ediQwen.phpExecutable and ediQwen.nodeExecutable for checks you use.
3. Open Noko AI: Open Chat from the Command Palette.
4. Send a message. The extension connects to the local server at http://127.0.0.1:8089 and can start the configured runner when needed.

The current startup check expects Qwen3.5-4B with a 32K context window. Other models, configurable server endpoints, macOS and Linux are not supported by this release's setup workflow.

Settings retain the ediQwen.* prefix for compatibility with the earlier local prototype. Some defaults refer to a Laragon installation; configure them for your computer.

Working with a project

Open one trusted project folder. Ask for an explanation or a change. Active-file attachment is optional. Paste screenshots into the composer with Ctrl+V when using a vision-capable backend.

Editing and diagnostics: In a trusted workspace, requested edits save directly without a per-edit approval dialog. Runtime diagnostic code executes with your local user permissions. Review diffs and keep your project under version control. This preview does not yet provide a separate chat-only permission mode.

Browser inspection uses installed Edge or Chrome with a dedicated project profile. Laragon Apache site configuration can be discovered automatically; other setups may need ediQwen.previewUrl. If authentication is required, use Sign in, then Continue. Browser availability, syntax checks and visual design quality are different kinds of evidence.

Data and local operation

Chat messages, selected code, tool results and screenshots are sent to the configured local inference process. This extension does not require a cloud AI account. Browser checks load the requested site and its resources, which can make network requests.

Conversations and screenshots are stored in VS Code workspace storage. Project browser profiles and saved login cookies are stored in extension storage. Treat this storage as private. The local runner can remain active after you close the chat; use its stop.cmd to stop it.

Limitations

Small local models can misunderstand requests, produce incorrect code, or fail to complete complex tasks. Instructions and automated checks do not guarantee correctness. Review generated changes. Model weights are not included, and their licenses apply separately.

About Noko

Developed by Noko: noko.al.

Noko AI is an independent chat extension for Qwen models. It is not an official Qwen product and is not affiliated with or endorsed by the Qwen team.

License

This extension is licensed under the MIT License.

Bundled dependencies retain their own licenses.