We started Seleya Labs because we think the next decade of AI is going to look very different from the one we're in.
The current story is simple. A few labs sit at the frontier. Everyone else rents intelligence from them by the token. The labs get smarter, the API gets better, and that's how AI happens.
The trouble is, this story assumes compute is abundant. It isn't. Memory is sold out through 2026. Power, not GPUs, is now the binding constraint on new data centers. The supply chain has hard chokepoints at every layer, from rare earths to ASML, and none of them resolve in under five years. The labs at the top of the stack will get their compute. Most of us won't, at least not at any price we want to pay.
Our bet is that what most people actually need from AI doesn't require a frontier model. A correctly-sized model, fine-tuned on your own work, with proper memory and grounding, will outperform a frontier model at your work. The hard part isn't the model. It's everything around the model: the runtime, the context, the memory, the tooling, the way it deploys on hardware you control. That's what we're building.
Spock is our first product, and it's running in production today. A longer thesis is publishing in 2026.
Spock starts with a product people can actually use: chat, files, Spaces, approvals, scheduled work, and history in one place. The UI is not just a wrapper around the Agent. It gives people a clear way to see what Spock knows, what it is doing, what it has produced, and where its answers came from.
Spock connects to the systems where work and life already happen: Microsoft, Google, SharePoint, email, calendar, WhatsApp, files, and more. That layer keeps expanding, including MCP, so Spock can work across the tools people and teams already rely on.
The Agent does the work. It can search, reason, write, edit, create files, execute code, call tools, ask for approval, run on a schedule, and pick work back up later. It is built for tasks that need context, judgment, and follow-through.
Spaces give Spock context. They hold documents, conversations, files, outputs, citations, and history, then help the Agent understand how that context relates. For a person, a team, or a business, Spaces become the memory layer that makes Spock more useful over time.
The long-term direction is control. Spock should become a system people and organizations can trust, govern, deploy, and shape around their own data, tools, models, and infrastructure.
We're four people. We build AI for customers who can't afford for it to be wrong. We care about what the system can explain, not just what it can say.
That's Louis, Ty, Tertius, and Jordan.