Cast Gorilla
Making video streaming to the TV feel easy.
The question I was chasing
Could casting to today's TVs feel easy, even with the restrictions that come with Chromecast, Android TV and Apple TV?
Living-room streaming is supposed to be convenient. In practice there are a lot of format and device rules in the way. I wanted a system that reshapes the file on the fly when the TV needs it, so pointing a Mac at a TV feels like pressing play rather than solving a format puzzle. Cast Gorilla is not an AI product. I set out to build it with AI engineering and modern tooling, frontier models including Claude Code and Codex, so the idea could move at the pace of the vision: real encoding work, tested against real hardware.
The constraints
Video encoding against real external devices is a difficult challenge. There are a lot of profiles and codecs to work through, and each TV decides what it will actually play.
The work had to stay simple, elegant and fun to use, even while the pipeline underneath was doing something hard on real hardware. The agents had to stay honest too: a plan that only looked right in a chat window was not good enough.
The decisions that mattered
Build a smart pipeline that respects people's hardware. Probe the file, pick the cheapest path the TV will accept, and only change what stands in the way. Leave the original alone when you can. Explain the plan in plain English.
Test against real TVs, not green logs. An NVIDIA Shield and a Chromecast HD were the gates from day one. A custom command-line harness drove stress tests of the streaming engine on those devices, so Claude Code and Codex were working against physical behaviour, not a simulated stack.
Keep the app calm enough for mid-evening. The heavy work stays under the hood so casting still feels like pressing play.
Built with: TypeScript, Electron, ffmpeg (VideoToolbox + AudioToolbox), CASTV2, Claude Code, Codex, Cursor
Where it landed
castgorilla.com is live, the source is public under MIT, and the signed Mac build is on sale on Gumroad. It works well: it looks great, it feels great, and it is powerful under the hood.
It is also a study in AI engineering: vision first, then frontier models and a CLI harness that kept the work honest against real TVs.
Part of the Rolling Waves work archive.