Builds & use cases

A chat bot with no LLM: an agents & browsers build posted by CJ (Coding Garden) (@CodingGarden) that puts Jev, TypeSafe AI's System One decision model, to work, with a demo video on the post.

Antonio Leiva explains in Spanish how he used his always-on dot since the Friday before launch, connecting email and calendar and letting it propose work unprompted. A practical walkthrough of a real setup.

An engineer's personal homepage rebuilt with original music, film clips and Blender-modelled textures.

OpenAI researcher Noam Brown tested dots over a weekend and had his dot hunt down subscriptions to cancel. It then connected to a customer service channel and handled an entire texting conversation on his behalf.

A Blender scene produced by the model from a short prompt, shared as a video clip of the result.

Gasty gave Opus 5.5 an email from a travel company and got an animated promotional clip.

Meng To had Astra build a polished Three.js game where rocks, trees and characters are all code, then recreate each scene in Blender and cut the trailer, noting a big jump over Sol in UI and 3D design.

Wenli Xiao showed Astra doing physical in-context learning: give it a recording of a human doing a new task and it drives a robot arm to repeat it on the first pass.

Opus 5.5 produced a complete product launch video using the HyperFrames toolkit.

Show HN for stillwet.art, an experiment that hands Opus 5.5 a simulated wet-paint canvas so it paints with brushes and pigment rather than emitting pixels or SVG directly.

The builder says a single prompt produced design, models, map, UI, scripts and thumbnails for a complete Roblox game, and claims a pipeline of several games per day.

Video demo of one 'Chief of Staff' bot that sends each request to engineer bots running different coding agents, with shared persistent memory and real-time X search, so nothing gets copy-pasted between chats.

Vox (@Voxyz_ai) switched to dots as his primary agent, praising the iMessage-style interface and desktop voice conversation that feels like talking to a person, and details the bugs he hit. A hands-on usage report.

A full analytics dashboard produced by one prompt to Opus 5.5 and deployed live.