Grokked Bot
Open-source bot harness that runs AI teammates with their own containerized desktop, browser and shell on your own Mac or Linux computer.
At a Glance
Engagement
Available On
Listed Nov 2026
About Grokked Bot
Grokked Bot is an open-source bot harness published by GitHub user alnutile. It lets you message an AI bot like a colleague while the bot works in its own containerized computer, with a real desktop, a logged-in Chrome browser and a shell, on hardware you already own. The latest listed release is v0.6.0.
What It Is
Grokked Bot runs an agent loop in a local daemon and gives each bot its own Docker container containing Xvfb, openbox, Google Chrome, a terminal and a file manager. You pick the model through OpenRouter, a direct OpenAI key, or a local OpenAI-compatible server. The README says it is inspired by Grok Bot and Claude Cowork, but uses a container on your own machine instead of a rented cloud VM.
You watch the bot's live screen, can take over the mouse and keyboard at any moment, and hand control back so it continues from the page you left.
How a Bot Works
Each bot has its own browser profile, logins, files and shell. Per-bot tabs cover details such as instructions, an enforced site allowlist, model and per-message limits, plus environment variables, Slack, SSH keys, triggers and media. The agent drives the browser by accessibility refs rather than pixel coordinates, and uses tools for browsing, shell and file editing, and schedules. It can stop and ask a human when it hits a login wall or CAPTCHA.
Coding, Triggers and Automation
A bot can clone a repository, install runtimes on demand, edit code, run tests, commit, push a branch, open a pull request and respond to CI failures when given a GitHub token. Bots can be started by webhooks or cron schedules, including schedules the bot creates itself on request. An HTTP and WebSocket API and a CLI are available, along with an eval suite of twelve tasks for comparing models.
Safety and Setup
The domain allowlist is enforced inside the container, page text is fenced as untrusted, and saved passwords are encrypted and typed by the daemon without the model seeing them. Spending, step and time caps limit runs. Docker and about 5 GB of disk for the bot's computer are needed. The README notes that approvals, persistent memory and learned routines are not built yet.
Install Options
Installers exist for macOS (Apple Silicon and Intel) and Linux (.deb and AppImage), or you can run from source as a systemd user service. A Terraform setup for AWS is also described.
Community Discussions
Be the first to start a conversation about Grokked Bot
Share your experience with Grokked Bot, ask questions, or help others learn from your insights.
Pricing
Self-Hosted, Pricing Unpublished
Grokked Bot publishes self-hosting instructions and downloads. The project hasn’t published a software price or license. Your own hardware and model usage may carry separate costs.
- Self-host on Mac or Linux
- Use OpenRouter, OpenAI or a compatible local model server
Capabilities
Key Features
- Per-bot Docker container with headful Chrome, terminal and file manager
- Live screen view with human take-over and hand-back
- Model choice via OpenRouter, OpenAI or local OpenAI-compatible servers
- Per-bot site allowlist enforced in the container
- Saved logins encrypted and hidden from the model
- Per-bot environment variables, SSH keys and Slack integration
- Webhooks and cron schedules as triggers
- Coding workflow with git, pull requests and CI handling
- HTTP and WebSocket API and CLI
- Usage and cost tracking by bot, provider, model and role
- Eval suite of twelve tasks
- Spending, step and time caps
