qz / 轻舟
Install

OPEN SOURCE · LOCAL FIRST

An agent that
actually uses
the browser.

Read websites, automate actions, and save results as tables, files, or image collections. Connect your own model.

Chrome extension · Compatible with cloud and local model APIs

qz at work on Wikipedia’s Hangzhou page
Wikipedia · Page reading and JavaScript execution
Watch demo00:45
qz-browser-agentApache-2.0

Web research, saved as data.

qz Data view: a live research comparison of Playwright, Puppeteer, and Selenium

Compare tools using their official documentation

qz read the Playwright, Puppeteer, and Selenium docs and saved a six-criterion comparison. The result is a table you can edit and export.

qz image collection with Hangzhou landmark photographs

Collect images with their sources

Save photos from a page to a local collection. Browse thumbnails and open the source image. Shown here: Hangzhou landmarks from Wikipedia.

Task tree showing agents researching Hangzhou in parallel

Research with parallel subtasks

Delegate parts of a task and combine the findings. Here, agents research West Lake walks and Longjing tea experiences for a Hangzhou itinerary.

Scripts, tools,
and custom UI.

This Wikipedia Skill finds city photos and displays the results in an image grid. Its instructions, scripts, and UI can be reused on another city’s page.

JavaScript resource editor for the Wikipedia city photos Skill

Page scripts

Inspect and edit a Skill’s JavaScript resources directly in the panel.

Wikipedia Skill result with summary cards and an image grid

Custom result cards

Display tool results as summary cards, image grids, or tables inside the conversation.

  • Skills

    Install from a Git repository, edit in the panel, and activate on matching websites.

    Skill guide
  • TypeScript capabilities

    Add agent tools, panel views, settings pages, or providers as compiled TypeScript modules.

    Capability API
  • MCP bridge

    Give external agents access to browser tools and task controls through a local MCP bridge.

    MCP setup

Local storage. Your model.

Sessions, Skills, and collected data are stored in your browser. Connect a service with a compatible API and tool calling, including a local model server.

Task context is sent to your configured model service. Skills and MCP connections may exchange data with services you configure.

Permissions & data flow

Install qz

Download the ZIP for desktop Chrome. A model API key or local model endpoint is required.

Download latest releaseInstallation guide
  1. 01

    Download & load

    Extract the release ZIP. Open chrome://extensions, enable Developer mode, then load the extracted extension folder.

  2. 02

    Connect your model

    Open qz Settings. Choose a provider, enter your credentials, select a model, and save.

  3. 03

    Give it a task

    Open the side panel and describe what you want to do. You can adjust the task while it runs.

TRY A TASK

Open Wikipedia’s Hangzhou page. Make a table of its main sights, with a short description and a source link for each.

qz = qingzhou (轻舟), “light boat”.

The light boat has already passed ten thousand mountains.

qz demo

English demo · Captions available

Page reading, data extraction, scripting, Skills, and parallel tasks.

Screenshot

Open original