Brontosaurus is a browser-based generative AI canvas that creates interactive widgets from spoken commands in under one second. Independent developer Thomas D. Hughes published the demo on June 2, 2026 as a Show HN project exploring voice-first AI interfaces for the web.

What Happened

Hughes launched Brontosaurus as a Show HN experiment in real-time voice-driven UI generation. Open the page in Chrome, activate the mic, and describe what you want: "add a clock widget," "create a weather card," "move that to the right." The app routes your speech through a two-agent system: a Conductor agent that interprets and orchestrates the request, and a Builder agent that generates and renders each widget to the canvas.

Under the hood, Brontosaurus uses Chrome's built-in Web Speech API for transcription, then passes the interpreted command to OpenAI's gpt-oss-120b model served by Cerebras. Cerebras delivers around 3,000 tokens per second, enabling the sub-second widget generation that makes the interaction feel immediate. Multiple widgets can be generated in parallel from a single compound voice request.

Why It Matters

Voice-first creation tools have been technically feasible for years, but generation latency has always made them feel sluggish and impractical. Brontosaurus is a working proof of concept that inference speed, specifically thousands of tokens per second via purpose-built silicon, closes that gap. The interaction feels immediate because it is.

Hughes cites Ink and Switch research as a direct influence, specifically their work on generative canvas software and the goal of creation at the speed of thought. For UI designers and front-end creators, this points toward a workflow where describing a component verbally is faster than finding it in a component library.

Key Details

  • Browser support: Chrome only (requires Web Speech API)
  • Inference backend: OpenAI gpt-oss-120b via Cerebras at approximately 3,000 tokens per second
  • Architecture: Conductor agent decomposes and routes; Builder agent handles per-widget generation
  • Parallel generation: Multiple widgets render simultaneously from a single compound request
  • Status: Early development demo; creator is actively collecting feedback
  • Cost: Free to try, no account required

What to Do Next

Open Brontosaurus in Chrome and try asking for two or three different widget types in a single sentence. The parallel rendering is the most interesting behavior in the architecture. If you are building voice-driven or canvas-first AI tools, the creator is actively reading and responding to feedback on the Hacker News thread.