Repolode
Submit
APUS-AI-Lab

APUS-AI-Lab/fast-browser-use

A local-first browser automation engine and Agent Skill that runs Qwen3.5 models on-device to pick page actions from visible elements.

View on GitHub
Stars
231
Forks
26
Open issues
0
Last commit
29 days ago

What is it?

Fast Browser Use is a browser automation tool that runs entirely on the local machine. Instead of asking a large cloud LLM to write selectors or Playwright code, it scans the rendered page for visible, interactable elements and presents them as a finite set of candidate actions such as CLICK, SELECT, TYPE_TEXT and DONE. A local Qwen3.5-9B or Qwen3.5-35B-A3B model scores these candidates in a single forward pass, using one token per option, and the engine executes the chosen action.

The project runs on Apple Silicon through MLX, or on NVIDIA GPUs and CPUs through PyTorch. It can be installed as a standard Agent Skill for Claude Code and Codex, and it also provides a standalone `fbu` CLI and a Python API. Runs can be checked with external assertions on final URL, title or page text, and each run writes a trace file.

The README reports timings for a few navigation and form tasks on an RTX PRO 6000 Blackwell GPU, with task times of roughly one to five seconds. Mutating actions are not retried automatically, and the included guards check visibility, DOM freshness and read-only state before acting.

Key features

  • 01Chooses browser actions from visible DOM elements, so the model never writes selectors.
  • 02Scores candidate actions in a single forward pass using local Qwen3.5 weights.
  • 03Runs on Apple Silicon via MLX or on NVIDIA GPU and CPU via PyTorch.
  • 04Installs as an Agent Skill for Claude Code and Codex, with a standalone fbu CLI.
  • 05Verifies outcomes with --expect-url, --expect-title and --expect-text assertions.
  • 06Provides a Python API for streaming task state.

Use cases

  • Running repeatable browser tasks from Claude Code or Codex without sending page content to a cloud API.
  • Adding assertion-checked browser steps to CI pipelines.
  • Automating form filling or navigation on sites where the data must stay on the local machine.

Quick start

  1. 1

    Install uv (Python package manager)

  2. 2

    Clone the repository

  3. 3

    Install Python dependencies with uv

    uv sync

Things to know

  • Needs Python 3.12+ and a local model download (Qwen3.5-9B or 35B-A3B), which is large.
  • Practical speed depends on hardware: Apple Silicon via MLX, or an NVIDIA GPU via PyTorch; CPU-only use is slower.
  • The README's benchmarks were measured on a single high-end workstation GPU, so results on other hardware may differ.

Topics

#browser-automation#local-llm#agent-skill#playwright#qwen#mlx#pytorch#web-automation

FAQ

What is fast-browser-use?
A local-first browser automation engine and Agent Skill that runs Qwen3.5 models on-device to pick page actions from visible elements. Fast Browser Use is a browser automation tool that runs entirely on the local machine. Instead of asking a large cloud LLM to write selectors or Playwright code, it scans the rendered page for visible, interactable elements and presents them as a finite set of candidate actions such as CLICK, SELECT, TYPE_TEXT and DONE. A local Qwen3.5-9B or Qwen3.5-35B-A3B model scores these candidates in a single forward pass, using one token per option, and the engine executes the chosen action.
Is fast-browser-use free to use? What license does it have?
fast-browser-use is open source under the MIT license. Check the license terms before using it in commercial products.
Is fast-browser-use still maintained?
Actively maintained. The last commit to fast-browser-use was on Sep 21, 2026.
What can I use fast-browser-use for?
Running repeatable browser tasks from Claude Code or Codex without sending page content to a cloud API. Adding assertion-checked browser steps to CI pipelines. Automating form filling or navigation on sites where the data must stay on the local machine.
How do I install fast-browser-use?
Install uv (Python package manager) Clone the repository Install Python dependencies with uv: uv sync
Where can I get fast-browser-use?
The source code and installation instructions are on GitHub: https://github.com/APUS-AI-Lab/fast-browser-use

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects