From distributed systems
to agent swarms.

AI-first, by default — before it was a slogan. Before LLMs were a product category. The founders built distributed systems at Google, Xero, Trail of Bits, o1Labs, and Gensyn. Products that scaled across billions of transactions, millions of users, and billions in value.

Compiler work. Security research. Low-latency trade execution. Consensus and zero-knowledge proofs. Distributed GPU training. Pre-launch community building from zero. The kind of problems where a small mistake costs real money, and every line of code runs in production.

When we started Ensue, the first chapter was coordinated memory for multi-agent systems. A shared memory network that let agents store what they learned, share it selectively, and act on each other's findings. Store. Share. Automate.

We ran Autoresearch@Home: a distributed swarm of 115 agents sharing every experiment and every finding through a collective memory network. 280K+ memories and still compounding. It's the technical foundation everything else sits on.

Now we point those swarms at your hardest ML optimization problem. Inference kernels. Training loops. Architecture search. Data curation. Whatever moves the metric you care about.

A coordinated swarm runs thousands of experiments in parallel, compounds findings, and delivers a better model in six days from the first meeting. It's the alternative to hiring a researcher for six months with no guarantee.