Curious to see what OpenAI's "o" is tomorrow, and also to see Grok Bot's multiplayer announcement today. Wondering how this compares to our Operator and multiplayer agents our team has been using internally since Feb 2025.

How its going: LatchLoop made over 4800 commits and 443,000+ lines of code changed to itself in the last 30 days.

Astra is amazing. It built this game in 2 prompts. One for the base game and one that mostly said "make the graphics better". This is built with three.js running in the browser at 120fps.

Astra is available in LatchLoop. Use it with your existing ChatGPT subscription. Its impressive at putting 3D games together quickly and building complex visuals in code from references.

The Ox Alpha (stealth model) is free for the next week in LatchLoop inference Gateway. 1M token context, zero data retention, and impressive performance. Give it a try for free in LatchLoop!

Kimi K3 is live in LatchLoop and available at an incredible 460 tok/s via @modal This level of intelligence at this speed is impressive. Give it a try!

xAI's new Grok 4.5 is available in LatchLoop! We are seeing impressive results from it so far at an excellent price.

Jensen Huang: "Nobody writes prompts anymore. The new job is to write and handle loops."

That thing that Google and OpenAI announced today about being able to build apps directly in AI Studio/ChatGPT... Yeah, that's already in LatchLoop for desktop. We call them agent apps. Build apps that live next to your docs and code, run without hosting, share with your team:

LatchLoop can now attach demos and screenshots of its work to PRs it opens. Your team can review artifacts created by cloud agents directly in LatchLoop because LatchLoop is multiplayer! Watch as LatchLoop builds a color theme for itself, iterates until it has proved the feature works, and then shows you a video of it testing the result.

Cursor 3 today looks similar to how LatchLoop did almost a year ago. LatchLoop is built by one developer with no funding.

Want to know where your context is being used in realtime during a coding agent's run? Now you can. Hover the context percentage and you can see a breakdown of what's in the model's context window.

If you are serious about managing coding agents, you need new ways to be able to visualize progress across a number of tasks at once. Kanban boards don't cut it. LatchLoop's radar view puts tasks closer to the center as they near completion, cubes are larger based on complexity of task, and tasks can drift any direction depending on if they are getting stale or not.

Claude Sonnet 4.5 is now available in LatchLoop. Give it a try and let us know what you think! Our first impressions are that it has impressive performance, but GPT-5 beats it in price.

How much are you delegating to AI coding agents? We've made it easy to track the requests you make from your dashboard in LatchLoop.