gm legends
Today: an AI that designs a real object and hands you the manufacturing file, a pet per coding agent walking along the bottom of your browser, and a small model whose whole job is deciding things quickly.
gm legends
Today: an AI that designs a real object and hands you the manufacturing file, a pet per coding agent walking along the bottom of your browser, and a small model whose whole job is deciding things quickly.
| • |
UseIt designs physical things and hands you a file a manufacturer can actually use. Tell it what you want and it asks the questions a designer would, builds the real structure underneath, and works inside the design software, so what comes out is a real object and not a picture of one. Circle any part to change only that, look at it in 3D in the material it would be made from, and you get back an editable file with real dimensions on it. Daniel's line is that everyone has something they meant to make, a gift box for a friend's wedding or stickers of a character they have doodled for years, and then sizes and materials and manufacturers got in the way and they quietly let it go. |
| • |
Your agent writes the code, and then something has to say whether it did what you asked. Usually that something is another model reading the diff and forming an opinion about it. Tiden captures your intent as requirements, versioned alongside the code, then checks every change against them using tests that actually ran in your environment, so what comes back is a pass or a fail rather than a second opinion. No model grades the diff. It works beside Claude Code, Cursor, Copilot and Codex, and it's built by the team behind Qase, the testing platform 2,000+ teams already run on. (sponsored) |
| • |
Agent Pets gives every coding agent you are running a small animal that walks along the bottom of whatever page you have open in Chrome. Above each one is a capsule saying what that session is doing, Thinking or Your Turn or Allow Bash, and when it wants something it bounces, so you can approve the command from the tab you were already in without going to hunt for the terminal it was hiding in. Aryan built it because he kept losing agents to permission prompts in windows he was not watching. You can also photograph your own cat, crop its face and put it in there with the others. |
| • |
Mouse-1 is a small model that exists to make the quick decisions inside an agent harness, the routing and classifying and choosing that currently goes to whatever big model you have wired in. Pete says that on his own evals, dropping it in made the harness about 190 times cheaper and 13 times faster than Opus 5.5 while producing the same output, which is his number and worth testing on your own work. You can fine-tune it to your harness through one endpoint, and he passes the LoRA training through from the H100 provider without marking it up. Zero data retention by default for everybody. |
