We build agents that reason about the world rather than memorize it.

AgentBull is a research lab and AI-agent company grounded in complex systems theory. The proprietary end of our work lands in finance: alternative quantitative strategies of our own design, trading live with real capital and verifiable returns. The open end is general: infrastructure for AI colleagues, useful in any trade.

Classical physics predicts well because it names causes: find what drives change, write it down as dynamics, and the future follows. Traditional quant kept only the statistics — measure the past, then bet the future will resemble it.

Complexity economics starts from the opposite picture. Standard theory imagines a system at rest, disturbed now and then and settling back. But an economy is never at rest. Non-equilibrium is its normal state; equilibrium is the anomaly — brief, local, a pause between changes. Agents adapt, their adaptations remake the world they adapt to, and the act of predicting a market changes the market. A pattern, once found, is already dying.

So we model the causes of change, and we build cognition on causal logic rather than correlation. We infer probabilities forward instead of counting frequencies backward.

In the open. Most of what you will find here has nothing to do with finance. Ankole, our main open-source project, is a self-hosted AgentOS for shared AI colleagues: one installation, many agents, each with its own identity, memory, permissions, and work it answers for. It is built for work that needs an owner, not just an answer — sessions that run for days, fail alone, recover with context, and leave a trail you can audit. Nothing in it knows what a price is. And it runs on your own infrastructure, because a colleague you cannot inspect is not one.

In markets. Our proprietary work is an AI Associate for Finance — an associate, not a research assistant. Grounded in system dynamics, it runs one closed loop — perceive, recognize, decide, execute, learn from what follows — and what comes out is a trading signal, not an opinion. Research tools are judged by how convincing they sound; ours is judged by what it earns. It holds real trading capability and produces returns you can verify. It does not comment on the market; it answers to it.

• Causes over correlations.
• Dynamics over curve fits.
• Open code over white papers.
• Live capital over backtests — the market keeps score in numbers no one can argue with.

In 1962, Douglas Engelbart described the mission of augmenting human intellect: raising our collective capability to face complex problems, understand them, and solve them. Sixty years on, that work is unfinished.

Markets are our proving ground because they grade every idea, every day. But an agent that holds a seat on a human team — perceiving, deciding, answering for its results — is augmentation in the oldest sense. We intend to carry a part of that program.