Compose a task
Use the local dashboard with an OpenAI API model or a compatible Ollama model.
A browser workspace with action review
You decide the next move.
Run a visible Chromium browser locally, describe a task, and review the agent’s proposed action before approving it.
Auto ShiftPROJECT OVERVIEW
01Describe the task
02Review the action
03Approve or stop
Illustrated workflow · not a live application session
01 / Inside the project
Built around the work.
Start with the implemented workflow, then follow the source to understand how it works.
Review the action
Inspect the snapshot and approve or reject each planned action.
Take over
Stop the agent and use the separate browser window yourself.
02 / Explore the workflow
Three steps to understand it.
Compose a task
Use the local dashboard with an OpenAI API model or a compatible Ollama model.
Full setup and implementation ↗03 / Run it yourself
Start from the source.
Python 3.12, Git, and a graphical desktop. Setup downloads dependencies and Chromium.
Clone the repository and run these commands from its root. Review the README before connecting external services.
Open repository ↗SOURCE CHECKOUT
git clone https://github.com/honeyamn10-source/auto-shift.git
cd auto-shift
python setup.py
python launch.py04 / Know the scope
Clear expectations.
First implementation. Real account sign-in and a complete model-driven task still require acceptance testing. This does not connect to ChatGPT’s browser or reuse its subscription.
Check the latest results in GitHub Actions. A passing automated check covers its tested paths, not every real-world integration.