Let ChatGPT Read, Edit, and Run Your Project Locally

You've probably pasted code into ChatGPT, gotten a helpful answer, then spent the next ten minutes copying files back and forth. What if the model could just open your project, edit the files, and run the tests itself? That's the pitch behind Chat On Steroids, an open-source workspace that connects your local files and terminal to your existing ChatGPT conversation.

What It Does

Chat On Steroids (CoS) is a desktop application that lets ChatGPT read and edit files, run tests, keep terminals open, and use your desktop. It follows actual tool results as they arrive, so you're watching the work happen rather than waiting on a summary. The project connects local tools through MCP (Model Context Protocol), and a companion component observes and automates the ChatGPT browser UI.

What's notable is what it doesn't do. It uses your ChatGPT conversation rather than invoking Codex directly. Your account's model availability, usage, and context limits still apply—ChatGPT Work and Codex share usage limits. It ships as a download for Windows x64, macOS Apple silicon, and Linux x64, and it's explicitly labeled an independent beta to use at your own risk.

Beyond the basics, CoS adds a layer for managing work. You can split independent jobs across workers, and those workers keep their context so the next task picks up where they left off. For longer jobs, you can send a correction while work is running, use Goal to follow unfinished work, use Loop to keep working within your brief, or use Compact & Resume to carry the session and worker history into a fresh chat.

Why It's Cool

• It works on the real project, not a sandbox. The model reads and edits your actual files, runs your actual tests, and keeps terminals open. That's a meaningfully different loop than copy-pasting snippets into a chat window.

• Workers with persistent context are the interesting part. Splitting independent jobs across workers isn't novel on its own—plenty of agent tools do parallel execution. What matters here is that workers keep their context, so you're not re-explaining the project state every time you hand off a task. That's the detail that makes delegation actually useful instead of just flashy.

• Long tasks don't have to be all-or-nothing. Being able to send a correction mid-run, or use Compact & Resume to carry history into a fresh chat, addresses a real annoyance with agentic tools: once you kick off a long job, you're often stuck waiting or starting over. Here you can steer.

• It rides your existing ChatGPT plan. Because it uses your ChatGPT conversation rather than hitting the API directly, you're working within the plan you already have. The README is upfront that this means shared usage limits apply—no free lunch—but it also means no separate billing setup to think about.

• The responsible-use section is refreshingly direct. The README states plainly that CoS is not intended to bypass usage limits, account restrictions, or safety controls, and that workers, Goal/Loop, Compact & Resume, and finish checkpoints organize work but don't grant extra quota. It also spells out that a local permission or enabled connector isn't permission to override a provider refusal. That kind of honesty is worth noting—many tools in this space are vague about where the line is.

How to Try It

Getting started is a download away. Grab the build for your platform from the repository's releases:

• Windows x64: Chat-On-Steroids-Setup-x64.exe
• macOS Apple silicon: Chat-On-Steroids-macOS-arm64.dmg
• Linux x64: Chat-On-Steroids-Linux-x64.deb

There's also an all downloads page if you want to browse versions.

Before you connect anything, read the usage notice in the README. The project is an independent beta, and the README asks you to use it within your provider's rules. Concretely, that means following OpenAI's Terms of Use, Usage Policies, and Service Terms, plus your workspace's own rules and any connected service's terms. If you're in the EEA, Switzerland, or the UK, there's a separate Europe Terms link.

Once installed, the workflow centers on giving ChatGPT access to your project directory and letting it work. There's a demo video linked in the README if you want to see model selection, task plans, live tool results, and reusable workers in action before you commit.

The repo lives at github.com/totec448-spec/chat-on-steroids, and there's a CHANGELOG if you want to track what's new.

Final Thoughts

Chat On Steroids is best suited to developers who already pay for ChatGPT and want to close the gap between chatting about code and actually running it. The worker model and the mid-task correction tools suggest it's built for people juggling multi-step jobs, not just quick edits. The honest caveats—independent beta, shared usage limits, don't try to evade restrictions—are a good sign that the maintainers aren't overselling it. If you've been curious about giving an AI agent real access to your local environment, this is a low-friction way to try it within the plan you already have.

Follow @githubprojects for more developer tools and open source projects.