Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

totec448-spec/chat-on-steroids

AI Engineering

A local MCP workspace that lets ChatGPT read files, run commands and delegate tasks across multiple agents.

5,183
675
TypeScript
MIT
2 months old

What it does

Chat On Steroids (CoS) is a desktop app that connects your ChatGPT conversation to local tools through MCP, plus a browser extension that loads into the ChatGPT web UI. Instead of calling Codex directly, it lets ChatGPT read and edit files in folders you approve, run shell commands and tests, keep terminals open, and use desktop capabilities where enabled. It also records conversation content locally and supports phone notifications through ntfy or your own Telegram bot.

Long-running work is organized around three ideas: Goal follows unfinished tasks, Loop continues within a brief, and Compact & Resume carries chat and worker history into a fresh conversation. Independent jobs can be split across workers that retain their context between tasks. When a step needs a human, CoS presents it as a step for you with the command, folder and expected result; you run it and mark it done, and the app does not execute it.

It is aimed at developers who want a Codex-style local coding loop while staying inside their existing ChatGPT account and plan.

Get started

  1. Install CoS and open Settings → Setup; the wizard walks through seven steps.
  2. Connect the browser: load the extension folder shown by Open extension folder into Chrome, Edge or Brave via Developer mode → Load unpacked. Pairing is automatic.
  3. Choose the folder ChatGPT may work in and the capabilities it may use there.
  4. Connect Core: create a tunnel and a restricted API key at OpenAI, press Connect, then add the app in ChatGPT under Plugins → Add → Add custom MCP server. See the tunnel setup guide.
  5. Pick a model per chat, write a task and send it.

Downloads are available for Windows x64, macOS Apple silicon and Linux x64, with all releases on the releases page.

Requirements and constraints

  • Desktop support covers Windows 10/11, macOS 13 Ventura or newer, and current desktop Linux. Browsers: Chrome 125+, current Edge or Brave. Custom MCP apps in ChatGPT depend on your plan and workspace policy; check account availability.
  • Windows builds are not publisher-signed and macOS builds are unsigned and unnotarized. Verify packages against the release checksums. On macOS, after each update the app asks once for your login password to reopen saved keys; chats and browser pairing wait until answered.
  • Linux requires a Secret Service keyring. The DEB is preferred; the AppImage launcher may fall back to --no-sandbox when unprivileged user namespaces are disabled.
  • Shell commands run with your normal user privileges. Fresh installs enable Core capabilities and two workers, and Windows additionally enables Desktop permissions, so review approved folders and capabilities before connecting.
  • Worker, Goal/Loop and Compact & Resume features organize work but do not add quota or model access. The project states it is not intended to bypass usage limits or safety controls, and it is not affiliated with or endorsed by OpenAI. It is MIT-licensed, with a separate security document.

HotRepo's take

The repository's single recorded TypeScript daily snapshot on 2026-11-14 shows 5,182 stars and 675 forks with 12 stars gained and a best position of 12 on that list. That is one data point, not a trend, and no all-language daily rank is recorded. The notable technical angle is the deliberate use of the ChatGPT conversation and MCP tunnel rather than direct Codex invocation, alongside local recording and browser-UI automation — a design that raises provider-policy questions the README addresses at length rather than treating as invisible implementation detail.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. nightly-labs/openbotA desktop workspace that keeps multiple coding agents in separate sessions, with queues, file sharing and agent-to-agent messaging.TypeScript903
  2. eduwxyz/tucanoA desktop app that runs a team of coding agents in shared channels, using harnesses you already have installed and signed in.TypeScript106
  3. elie222/rakazoA self-hosted multi-channel AI agent platform for persistent bots with sandboxed computers, aimed at people who want to own the stack.TypeScript5,573
  4. makecindy/cindyA desktop and mobile client that wraps Claude Code and Codex into one local agent with shared memory, skills and automation.TypeScript4,011
  5. agentes-chat/agentes-chatSelf-hosted team chat where people and AI agents share channels, DMs and approvals.TypeScript183