We built Mirasim because we kept losing track of our own coding agents.
Past three or four terminal tabs of Claude Code and Codex, nobody on our team knew which one was done, which one was waiting on a question, and which two were editing the same file.
So now we just tell Mira Bot what needs doing, in a chat. What it fixes for us:
1. One message instead of five tabs. Say "fix the checkout timeout, and add a CSV export", and Mira Bot hands each task to an agent and starts it. Claude Code, Codex and 8 other agent CLIs work.
2. You can see who is doing what. Every task shows its current step in the chat, and the result lands in the same place when it's done.
3. Nothing is hidden. Open a task and it's a normal Claude Code or Codex session underneath. You can keep talking to that agent directly.
4. No second bill. It runs on the Claude or ChatGPT subscription you're already signed into.
It's a desktop app for Mac, Windows and Linux, free with your own accounts. It's still alpha and it isn't open source.
15 seconds of it below (demo workspace, made-up data). If you run more than one agent at a time, what breaks first for you?
https://t.co/iuWKwkBHs0
You need a periodic table to pick a Gemini 4 model now, so here is one.
118 elements in glass. Click any of them. Try heating the gold.
Built by Gemini 4 Carbon and Claude Fable 5.2.
our computer use just cleared a 16x16 minesweeper board in 3.9 seconds.
one tool call. 78 steps. no screenshot in between.
it's open source:
https://t.co/cEoJGnFyU9
We just open-sourced our computer use framework. and yeah, we're calling it sota.
same model, same 60 tasks, vs codex:
26% faster
45% fewer model calls
50% fewer input tokens
Mirasim Computer Use is now SOTA & open source.
Same model, same tasks
Head-to-head with codex Computer Use.
Mirasim leads on every metric:
> 50% fewer input tokens
> 45% fewer model requests
> 26% less time
> half the failed runs (4 vs 8)
one living room, six interior styles, switching live in the browser.
modern minimal → scandi → wabi-sabi → industrial → cream → new chinese
webgpu, a single 264kb html file. no textures, no 3d models, every chair and lamp is generated in code.
vibed it with fable 5.2 in one afternoon.
1/4
@cursor_ai made Projects multiplayer yesterday: teammates share one Project with a unified context layer and a task board, while their coordinator agents sync across the team.
As a PM at @Mirasim_ADE, I mapped one everyday spec workflow across their post and our setup.
the biggest change from vibe coding isn't speed. it's that i'm embarrassed to write bad specs now.
be vague with a person and they'll ask what you meant. be vague with an agent and it'll hand you something vague right back.