Wanted to show what I've been working on for the past few months, an agentic super computer that can take in any fathomable goal, recursively decompose that goal until sub-goals are proximate to existing ai agent capabilities, and then execute it by spawning up to 10k agents in parallel.

There has been a lot of new research around the space, from Kimi-2.5 swarms to Claude's Agent teams, but no system yet has what's needed in order to actually be able to accomplish most people's everyday goals.

1. Out-of-the box data integrations 2. A stable enough task decomposer that doesn't break under compounding errors 3. A harness that can persist runs stably for weeks on end

Would love to get any feedback & answer any questions, the product is open & open to free to use rn!

The vision here is build a system that allows anyone to be a compute-bender; if you can name what you want to see in reality, then Rose Labs orchestrates the stream of agents to accomplish it in the cheapest, most efficient way.