Our manifesto
Intelligence should make more things possible for more people.
We want useful technology to feel close at hand. It should work where people live and build, respect the limits of their machines, and earn its place through what it enables.
01. Start with what people need
The point of a model is not the model itself. The point is what someone can do with it. We begin with real tasks, then choose the smallest system that can do the work well.
Classification, search, routing, and extraction may look modest beside a general assistant. They also solve concrete problems, fit into existing software, and can be measured against clear outcomes.
02. Bring intelligence closer
Every unnecessary handoff adds delay, cost, and dependence. We explore local execution because useful software should be able to work near its user, on hardware they can access.
Local does not mean effortless or universal. Devices have limits, and some tasks need remote resources. We will describe those tradeoffs plainly and make the boundary visible.
03. Make small systems dependable
Software grows through parts that people can understand and combine. Focused models can become useful building blocks when they behave consistently, expose their limits, and fit into the tools people already use.
We value clear interfaces, reproducible evaluation, and components that can be inspected. Good foundations leave room for others to make things we never imagined.
04. Let evidence do the talking
Fast is a measurement, not a mood. Better means useful quality at a cost people can accept, with latency, memory, energy, and reliability accounted for.
We will distinguish measured results from targets and experiments. When a system falls short, that is part of the record. Honest limits help builders decide what to trust and what to improve.
05. Earn trust in the details
People should be able to understand what a system does with their information and what it cannot do. Privacy, safety, and access are design questions from the beginning, not decorations added at launch.
Trust grows through careful defaults, visible behavior, independent scrutiny, and the freedom to choose another path.
What we are here to do
Build practical intelligence. Keep it understandable. Measure what matters. Share the work so others can build on it.
We are a small independent lab in Brazil. We start with focused research and stay accountable to the people who use what we make.
Build closer. Prove each step. Leave room for others.