Winsen Labs · A research lab
Winsen Labs is a small research lab. We study how intelligence is structured, turn that research into products people actually love, and help companies put AI to work. Properly. The research, the products, and the runtime that carries them all live under one roof.
Who we are
We started Winsen Labs because most of the tools we used at work felt older than the people using them, and most of the research we cared about lived and died as a paper. So we built a place where research and product talk to each other every single day. Where a hard question is allowed to take a year if it needs one. We're small, we move slowly when slow is the right speed, and we answer the email ourselves.
What we do
A compact program can grow useful structure. A neural network. A network of proteins. A company that wants to work alongside AI. We chase that idea from the research all the way into the products, and we keep everything open where we can.
A self-evolving model of your company.
Winsen learns your processes, your voice, and your people, then trains AI employees that do real work the way you would. All supervised from one workspace. The model is part of the team, not a feature bolted on top.
The open-source agent runtime.
Build, ship, and operate AI agents on infrastructure you own. Model-agnostic, monitored by default, self-hostable in one docker compose up. No seat license. No lock-in. No drama.
A few hundred parameters that grow a whole network.
Our research on compact "genomes" that grow entire neural networks, then turn out to map onto the networks inside the body, from proteins to cancer pathways. Papers, code, and models: all open.
Idea to production, managed in the repo.
A project hub that lives in your codebase as one beautiful offline file, plus a Claude skill stack that walks an idea from spec to shipped. Project management, rebuilt for the age of AI coding agents.
npx @winsendotai/rocketmanHow we help
Each year we take on a few engagements with teams whose problems genuinely excite us. The kind where research is part of the answer, not an afterthought. We sit alongside your team, we ship something real, and we leave you owning every piece of it. No pitch deck. No black box. Just the work, done properly.
A private brain over your mail, calendar, tasks, docs, and data. Searchable, ask-able, and quietly compounding the longer you use it. The first step to weaving AI into how you actually work. Yours when we're done.
From a sketch on a whiteboard to an agent that lives in production and earns its keep. We design it, wire it into your stack, put guardrails where they matter, and stay until your team has every answer.
Customers and products. Proteins and pathways. Internal knowledge graphs. If your hardest problem looks like a network, we bring the compact priors and structural transfer from our research, and we ship something you can actually run.
Don't want to run the runtime, the monitoring, and the model layer yourself? We host it on Platos, manage observability and routing, keep models and tools current, and let your team focus on what the agents do, not how they stay alive.