I built Termitty because I was tired of the hacky solutions I kept writing for SSH automation.

While building my workflow automation SaaS, I had this seemingly simple requirement: workflows needed to execute tasks on remote servers, not just in browsers. Selenium made the browser part elegant - you could wait for elements, handle dynamic content, maintain session state. But for SSH? I was stuck in the stone age.

I ended up writing this monster of a custom driver that:

- Maintained persistent SSH connections in a pool - Kept track of working directories between commands - Tried to parse ANSI escape codes with regex (don't do this) - Had a "wait_for_prompt" function that was just `time.sleep(2)` with extra steps

The breaking point was when I needed to automate a workflow that involved:

1. SSH into a production server 2. Start a database backup 3. Navigate an ncurses configuration menu 4. Detect when the backup finished (from a progress bar!) 5. Verify the backup file

My "solution" was embarrassing - screen scraping terminal output with regex, hoping the prompt detection worked, and praying the timing lined up. It was brittle, unmaintainable, and I knew there had to be a better way.

So with Claude Opus 4's help, I built what I wished existed - Termitty brings Selenium's patterns to terminal automation

The key insight was that terminals are stateful UIs, just like web pages. So Termitty maintains a virtual terminal buffer - it actually understands what's on screen, tracks cursor position, handles ANSI codes properly. You can find text on the terminal screen just like finding DOM elements.

Some cool stuff that emerged:

- Full terminal emulation - It maintains a complete VT100/ANSI terminal state - Session recording - Records everything to JSON/asciinema format with a slick replay UI - Interactive shell - Persistent shell sessions that can control vim, top, installers - Real wait conditions - No more sleep()! Wait for specific output, prompts, or patterns

Once I had structured terminal state, AI integration became obvious - now LLMs can actually "see" the terminal and make intelligent decisions about what to do next.

What SSH automation pain have you encountered? I'm especially curious about edge cases I haven't hit yet!

P.S. - Check out the terminal player in the docs. You can record a session and share it with beautiful playback UI - it's like Loom for terminals!