Open Dot is an open-source Mac app for OpenAI's personal-agent idea, and it reached 221 GitHub stars in two days: Reveneau

Open Dot is an open-source Mac app for OpenAI's personal-agent idea, and it reached 221 GitHub stars in two days

An open-source Mac app called Open Dot runs the same kind of personal background agent OpenAI announced on 29 September for its Pro and Business Premium tiers. The project uses your own OpenAI key or open models through OpenRouter, connects to 1,500 apps through Composio, and reached 221 stars on GitHub in under two days.

Why it mattersA team that wants to try the Dots pattern without paying for a Pro seat per person can now self-host the whole workflow on a Mac, pick its own models and tool list, and keep the data in a folder on the laptop.

The agent that keeps checking your email and calendar on its own, between the times you open the app, is a paid ChatGPT feature right now. OpenAI announced Dots on 29 September for its Pro and Business Premium plans. A new open-source project called Open Dot does the same job on your own Mac, with your own OpenAI key or open models through OpenRouter. The repository reached 221 stars on GitHub in under two days.

What the app does

Each dot has its own browser that stays signed in to the sites it uses, and the user can watch it work and take over when it needs a login or a captcha. Rules decide what a dot may do on its own and what it must ask about first. A small review model checks each risky action against the rules, and the user approves or denies it from a card in the chat. Passwords are kept in an encrypted vault that uses a key from the macOS Keychain, so the model never sees them.

Connections to Gmail, Google Calendar, Slack, Notion, GitHub and 1,500 other apps go through Composio, which the project's own README names as the tool integration layer. A dot can read on its own, but has to ask before it sends, posts, pays for or changes anything. Routines put a dot on a schedule, and triggers wake it when something happens in one of its connected apps.

The models, the compute, and the terms

The default model is the best one the user's key can run, and the model picker falls back to Kimi, DeepSeek, Qwen and GLM through an OpenRouter key. Voice calls run on OpenAI's Realtime API, so that part needs an OpenAI key either way. For long-running work the app can start a cloud computer on E2B, a local Docker container, or a folder on the Mac.

The app runs on one Mac at a time. There is no account system, no login screen, and no cloud backend of its own. The README says this openly, and warns against putting the app on a public server as it is. The GitHub repository ships without a LICENSE file as of 1 October, and package.json carries "private": true, so a team adopting it should check the terms with the author before shipping anything on top of it.

What the traction says

The repository was created on 29 September at 16:34 UTC by the composio-community organisation, a 41-repository GitHub organisation separate from Composio's main ComposioHQ account. From the GitHub API stargazers_count field as of 1 October 06:20 UTC, the count is 221 stars with 20 forks and 1 open issue, over the 37 hours and 45 minutes since the repository was created. That is 5.9 stars an hour. The README states that the project is not affiliated with OpenAI; it does not state whether it is an official Composio release or a community effort using the Composio platform.

For a product team that wants the Dots pattern without a Pro seat for every user, this is a working alternative to try on one laptop first. The three choices a team was going to have to make for an in-house version (model, tool integrations, where the computer runs) are each wired to a config knob here, with the open-model path kept open through OpenRouter.

Source

Primary: composio-community/open-dot README and GitHub API, stargazers_count 221 as of 2026-10-01 06:20 UTC.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.