Hey, Xuan here. I am building Komos.

The idea is simple. You teach the computer what to do the same way you'd teach a coworker. Record your screen and narrate what you're doing. Komos watches, learns the steps, and turns it into an automation you can run again.

Here's how it works:

- AI agent in a VM. Each task gets its own virtual machine with a full browser. The agent can browse websites, run code, process data, send emails, and more. Think of it as a computer that works for you.

- Moss, your AI Automation Engineer. Chat with Moss to build, run, and debug automations. One user last night went from signing up to a working task in 45 minutes, all through conversation.

- 100+ integrations built in. The AI decides the best route. If there's a direct integration available, it uses that instead of the browser. Faster and more reliable, and you don't have to think about it.

- Production grade reliability. This is where most automation tools fall apart. We use a combination of visual recognition and direct page code operations to make sure every click, every input, every extraction is accurate. When a website changes its layout, the AI figures out what changed and adjusts. Each run also learns from the ones before it, so future executions get faster and more accurate over time.

- Self hosted runners. Need to automate internal tools behind a VPN? There are runners for Windows and Mac that connect to your own machines.

Curious what people would use this for. Happy to answer questions.

- Moss, your AI Automation Engineer. Chat with Moss to build, run, and debug automations. One user last night went from signing up to a working task in 45 minutes, all through conversation.

Komos

AI agents that learn from watching you work

AI agents that learn from watching you work