Hashimoto on Ghostty: why terminals need a standards body, not a browser rebuild

Mitchell Hashimoto, the HashiCorp founder behind Terraform, Vault, Consul and Nomad, explains why he pivoted from distributed-systems tooling to building Ghostty, a terminal emulator written in Zig. After 15 years shipping CLIs, he realized he didn’t actually understand how a terminal worked, and wanted an excuse to sharpen dormant skills in GPU programming and single-node systems work where cache locality matters. No existing terminal hit his target of fast, feature-rich, and natively cross-platform, so he built one. It started as a private beta shared in a friends’ Discord — he deliberately delayed a public release to avoid the disproportionate attention his name would draw — until those friends were using it daily.

Hashimoto pushes back on the idea of turning terminals into general application platforms like the browser. Text-grid apps have their own strengths — quick to build, easy to script, with a clear security model — and compose far better than GUI paradigms. The real bottleneck, he argues, is the PTY’s in-band signaling: an unstructured byte stream riddled with escape sequences. He credits PowerShell’s structured-data model as pointing in the right direction. His design method is to study decades of prior art (AppKit, Win32, GTK, the DOM) before inventing anything, and he says he hasn’t shipped a single custom protocol yet. Two ideas he’s serious about: an n-screen API that would let programs populate unlimited background screens and render them as native window tabs, and a button protocol extending OSC 8 hyperlinks so clickable elements survive being scrolled into history — a gap that bites main-screen tools like Claude Code.

The deeper problem, in his telling, is governance: terminals have no standards body, so the last 20 years of “standardization” have amounted to copying whatever the popular emulators do, producing a feature hodge-podge with no coherent vision. He floats building an entirely new home for text-based applications with a legacy-terminal compatibility layer, but admits he doesn’t know the right path. On maintainership, he’s blunt that open-source authors owe users nothing beyond the “as-is, no warranty” license — yet he feels a self-imposed pull to make the software genuinely good, balancing days spent grinding through issues against days spent chasing the larger vision.