00 · Why we exist
Intelligence is becoming infrastructure.
For decades, we taught computers to store, calculate and execute. Now AI systems interpret language, recognize patterns, create, plan, research, use tools and coordinate actions. What once depended exclusively on human processes is becoming a programmable computational capability.
This changes more than software. It changes who can create knowledge, operate systems, build companies, conduct research, learn and transform the world around them.
The next great transformation in AI will not only be bigger models. It will be distributing the capability to build intelligence.
Polybee exists to work at this foundational layer: researching, building and democratizing the technologies behind intelligence—models, data, agentic systems, tools, evaluation, compute and knowledge.
We were born in Brazil. We think globally. We believe the future of AI will be richer when more people, organizations, languages and communities can genuinely participate in building it.
01 · The age of programmable intelligence
We are building a new layer of computing.
The internet made information distributable. The cloud made computing available on demand. AI is beginning to make perception, reasoning, generation, planning and action components of digital systems.
We are still at the beginning. Paradigms will shift. Models will be replaced. Architectures will be reinvented. What seems definitive today may look rudimentary in a few years.
Our mission is therefore not to defend a stack, an API or a trend. It is to develop the capability to understand and build the foundations beneath them.
Models change. Tools change. The capability to build endures.
02 · Access is not capability
We have never had access to so much intelligence—and never relied on so few to produce it.
Much of the AI frontier is developed by a small number of organizations concentrating models, compute infrastructure, platforms, capital and talent. They have accelerated the field tremendously, and we use, study and collaborate with technologies they produce.
But a healthy ecosystem cannot exist solely at the consumption layer. Access to an interface does not mean having the technical capability to adapt a model, evaluate its behavior, host inference, create quality data, replace components or operate independently.
Access matters. Capability goes deeper.
03 · Brazil as our origin, the world as our horizon
Brazil should help build AI, not only use it.
Every major technological transformation offers a choice: participate in building new capabilities or limit ourselves to consuming what others define. That choice matters to us.
We want Brazil to be able to create frameworks, models, datasets, tools, infrastructure, intellectual property, research and AI companies. We can collaborate with the best in the world while developing our own expertise.
That includes treating Portuguese as a first-class language in AI, rather than adapting it after fundamental decisions have already been made elsewhere. We want to contribute to models, evaluations, data, voice, multimodality and systems that understand our language and contexts from the outset.
Portuguese should not be a translation of AI. It should be one of its native languages.
Our Brazilian origins do not limit our ambition. They define it. We build from here to participate in a global frontier.
04 · Technological sovereignty
Sovereignty is the ability to choose.
Technological sovereignty does not mean isolation, rebuilding every component alone, or rejecting global platforms. It means having enough knowledge, alternatives and infrastructure to decide which model to use, where to run it, where to store data, which components to replace, which risks to accept, what technology to develop and whom to depend on.
Interdependence by choice is collaboration. Dependence without alternatives is vulnerability.
Polybee researches practical independence: model-agnostic architectures, open standards, local execution when needed, interoperability, portability and knowledge that does not disappear when an API changes or a provider is no longer suitable.
05 · Democratization
We do not only want more AI users. We want more AI builders.
Democratizing AI is more than offering a simple interface. It means distributing the knowledge, tools, data, compute and methods people need to understand, experiment, adapt and create.
A child should be able to understand. A student, to experiment. A developer, to build. A startup, to compete. A university, to research. A company, to control its infrastructure. A community, to create intelligence for its own problems.
That requires education beyond prompting: labs, documentation, open source, benchmarks, spaces for experimentation and accessible paths from curiosity to real technical capability.
Democratizing AI means narrowing the distance between using intelligence and being able to build it.
06 · The layer beneath the interface
Behind the scenes, intelligence becomes infrastructure.
Much of the market sees AI through the interface. We want to understand and build the layers that make that interface possible, reliable and portable.
Our research and engineering agenda moves across connected areas:
- Agentic systems: harnesses, orchestration, memory, tools, multi-agent architectures and specification-driven development.
- Models & intelligence: fine-tuning, distillation, SLMs, specialized models, multimodality and Portuguese-language model research.
- Data & knowledge: datasets, pipelines, retrieval, knowledge graphs, ontologies, memory and knowledge infrastructure.
- Evaluation & governance: evals, observability, traceability, safety, policies and reliability criteria.
- Compute, edge & robotics: local inference, clusters, mini data centers, embedded systems and models for machines and robots.
We turn these areas into reusable artifacts: frameworks, harnesses, components, protocols, models, tools and infrastructure that increase autonomy, reliability and speed for AI builders.
We work beneath the interface because real capability begins with understanding the foundations.
07 · Open-core
Open where it multiplies. Proprietary where it sustains.
Shared knowledge accelerates ecosystems. Papers, code, models, datasets, benchmarks, documentation and tools can allow thousands of people to move forward where only a few organizations once could.
Research must also be sustainable. Compute, good researchers, infrastructure and time to investigate hard problems all cost money.
We open what can accelerate the ecosystem. We protect what makes the mission sustainable.
Polybee is a commercial research company with an open-core philosophy. We will publish and open components when doing so expands collective capability. We will build intellectual property, products, licensing, infrastructure and services when these finance continuing research and help it scale.
Profit and purpose need not be adversaries. Properly designed, the first can sustain the second.
08 · Beyond the cloud
Intelligence must exist where it is needed.
The cloud will remain a foundation of modern computing. But intelligence will not be confined to large data centers, browsers or remote APIs.
It will live in devices, sensors, machines, vehicles, factories, equipment, cities and robots. Some applications demand low latency; others privacy or autonomy. Others must simply work where connectivity or cost makes the cloud insufficient.
We therefore study smaller models, local inference, edge AI, private clusters, hybrid infrastructure, mini data centers and robotics.
We do not only want to research AI software. We want to research intelligent systems.
The frontier of intelligence will be distributed. Our research must be too.
09 · What we believe
Ten principles guiding what we build.
01
Intelligence is infrastructure. We want to understand the foundations of a new layer of computing.
02
Intelligence should be accessible. Technical capability should not remain confined to large organizations and research centers.
03
Intelligence should be understandable. People who rely on intelligent systems must be able to evaluate, trace and control them.
04
Independence comes from the ability to build. Sovereignty means having knowledge and alternatives to choose.
05
Open source accelerates ecosystems. Open knowledge lowers barriers and multiplies experimentation and collaboration.
06
Models are only part of the system. Data, agents, tools, memory, evals, protocols and compute are part of applied intelligence.
07
Research should reach the real world. We aim to turn investigation into code, models, tools, infrastructure and practical capability.
08
Context matters. Language, culture, data, infrastructure and economic conditions shape what intelligent systems must be.
09
Intelligence will be everywhere. Cloud, edge, devices, machines and robots will coexist in distributed architectures.
10
AI should expand human capability. We build technology to enlarge what people and communities can understand, create and transform.
10 · Our promise
Research that becomes technology. Technology that becomes capability.
We will investigate hard problems even when no clear product awaits at the end. We will build prototypes, tools, frameworks, models and infrastructure. We will measure, publish, open, license, teach, document, experiment, fail, correct and try again.
We will collaborate with researchers, developers, universities, companies, communities and other labs. No organization can build alone a technology reshaping the whole world.
We do not know which architecture will dominate in ten years, which models or chips will survive, or the final shape of the systems we now call AI.
But we know the capability we want to expand: the ability to understand, build, control and benefit from intelligence.
From Brazil to the world. Build intelligence. Distribute capability.
Polybee exists to broaden who can participate in building intelligence—as researchers, engineers, creators, entrepreneurs, students and communities, not only users.
The future of AI should be shaped not only by the largest models and data centers, but also by the diversity of people and places able to build on, around and beyond them.