Open source · Self-hosted · Local-first

An AI worker for real computer tasks.

Tyrum is an open-source AI worker that can write code, run terminal commands, manage files, browse the web, and use desktop apps. It uses APIs when available and falls back to keyboard and mouse when needed. You choose the model, control the guardrails, and keep a full action trace.

Get started View on GitHub

Install

curl -fsSL https://get.tyrum.ai/install.sh | bash

How It Works

Three steps to useful automation.

1

Install and run

One command installs Tyrum locally. No hosted control plane required.

2

Connect your model

Use any OpenAI-compatible endpoint, including local models.

3

Hand off tasks, review traces

Give Tyrum work, approve risky actions, and inspect exactly what it did.

Features

Everything you need, nothing you don't.

🖥

Does everything you can

If you can do it on a computer, Tyrum can too. Code, apps, browsers, files, terminals — no limits on what it can work with.

🛡

Guardrails that learn

Tyrum asks before doing anything sketchy. But it also remembers your preferences — so it gets less annoying and more helpful over time.

📋

Receipts for everything

Full execution traces for every action. See exactly what happened, why, and reproduce it anytime.

🔌

Bring the best models

Use hosted or local models through OpenAI-compatible APIs. Switch providers anytime. Zero lock-in.

⚡

APIs first, desktop when needed

Prefers clean API calls for speed and reliability. Falls back to keyboard and mouse when there's no API — just like you would.

🔓

Open source, for real

Read every line. No hidden telemetry, no black-box binaries. Fork it, hack it, make it yours.

Who It's For

Built for people who build things.

Your personal AI worker

Hand off the tedious stuff. Tyrum handles repetitive tasks across any app while you focus on what matters.

  • Works with any application on your machine
  • Learns your preferences over time
  • Asks before anything risky

Autonomous coworkers

Add AI team members that have their own identity and their own accounts on your systems. They show up, do work, and report back.

  • Each worker gets its own profile and permissions
  • Works alongside your team naturally
  • Audit trails for everything they do

Tinker and extend

It's open source. Don't like something? Change it. Want a new capability? Build it. The code is yours.

  • Fork-friendly architecture
  • Pluggable model backends
  • Community-driven roadmap

FAQ

Got questions? We've got answers.

Anything that speaks OpenAI-compatible API format, including local backends. Bring your own key or endpoint.
Anything you can do on a computer. Write code, use apps, browse the web, manage files, run commands. It uses APIs when they exist and falls back to keyboard and mouse when they don't — just like a human would.
Only to whatever AI model provider you configure. Run a local model and nothing leaves your machine. Period.
Those tools help you write code. Tyrum can do anything on your computer — not just coding. It's also self-hosted, open source, and model-agnostic. No subscription, no vendor lock-in.
macOS and Linux. Needs Node.js 24+.
Tyrum is free and open source forever. You only pay for AI model API calls (or nothing if you run a local model).