Agent swarms for everything.

The most capable agent systems today run inside a few frontier labs. We think large swarms of agents, working together, can exponentially accelerate the world's most long and complex work. That capability shouldn't stay with the people who train the models, nor should it be restricted to just some types of work.

Backed by Y Combinator. About the founders · Careers

01

Infrastructure

We build the tools for single agent swarms to efficiently coordinate and scale from just your laptop to an entire datacenter's worth of compute.

02

Data and models

We build traces and RL environments with long-horizon, many-agent work to bring models (both frontier and our own) to a place where they can productively scale to an unbounded amount of resources.

03

Evals

We measure the effectiveness of a model's swarming capabilities across domains and harnesses. See the benchmark.

Get in touch

For labs, partners, investors and researchers.

We only use this to reply to you.