Otto controls real Chrome tabs from your CLI or your LLM over a secure relay — no headless farm, no cloud-browser rental. Here's how it works and why interaction should be code, not tokens:
Every time I needed an agent (or a test, or a monitor) to touch a real website, I hit the same wall: the browser. Not the automation logic — the infrastructure under it. You end up choosing between two bad options, and I got tired of both, so I built a third. It's open source and it's called Otto.
The two options that annoyed me
Option one: run a headless farm. Spin up Docker, manage a pool of Puppeteer/Playwright instances, keep them patched, scale them, and pray your IPs don't get flagged. It works, but it's a standing operational cost, and headless Chrome is not the same as the browser you actually use. Different fingerprint, different behavior, no logged-in session. Plenty of sites quietly serve headless traffic a worse — or broken — experience.
Option two: rent cloud browsers. Pay a per-session or per-minute fee to a hosted-browser service. Convenient until the bill scales with your usage, and you're still not running in your real session with your cookies.
