Run local desktop commands through the ChatGPT website
An HN commenter built a desktop app plus Chrome extension so the ChatGPT website can act as a local coding agent running real desktop commands; it “sorta” works and is not generally released. Task: The author investigated using the ChatGPT website as a local coding agent to execute desktop commands. They implemented a bridge using a custom desktop application and a Chrome extension, requiring significant workarounds against website restrictions. The human role involved building this infrastructure and supervising command execution to prevent destructive actions.; tools: ChatGPT. Human role: Not stated in the source; limitations: This workflow is not generally available and requires maintaining custom software to bypass website limitations. Security risks of allowing a web UI to execute local commands are inherent and not fully detailed in the source. The approach is fragile, relying on "fighting" the website's interface..
Tools
Published on Agent Work Atlas:
Overview
An HN commenter built a desktop app plus Chrome extension so the ChatGPT website can act as a local coding agent running real desktop commands; it “sorta” works and is not generally released.
Workflow
The author investigated using the ChatGPT website as a local coding agent to execute desktop commands. They implemented a bridge using a custom desktop application and a Chrome extension, requiring significant workarounds against website restrictions. The human role involved building this infrastructure and supervising command execution to prevent destructive actions.
Reported result
The author describes the setup as "sorta" working, indicating partial functionality without full reliability. No benchmarks, task success rates, or performance metrics were provided. The solution was not released for general use due to technical and security constraints.
Human role
Not stated in the source
Limitations
This workflow is not generally available and requires maintaining custom software to bypass website limitations. Security risks of allowing a web UI to execute local commands are inherent and not fully detailed in the source. The approach is fragile, relying on "fighting" the website's interface.
Try it
Act as a local coding agent for the project in [project_folder]. First, analyze the directory structure and propose a [build/test_command] to verify the environment. Outline necessary fixes for any failures in [language/framework]. Display all proposed file changes and command executions for my review before applying them. Explicitly ask for confirmation before any file deletions or package installations. Stop and summarize the plan once the proposed commands are expected to pass. Do not execute any commands until I explicitly approve each step. If inputs are missing, ask clarifying questions about the project stack or specific goals before proceeding.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.