Local LLMs
Inference that can live with the organization — private, local, and under its own control. Not only a vendor cloud.
New Machinery 2.0 released — The Factory Is the Product: five
authorities, one truth, and a build system that learns. Read the announcement →
Research in organizational AI
Reflective Labs
We do deep research on AI that operates at the organizational level. Systems that learn from individuals to boost them — and collaborative, real-time intelligence that compounds over time, strengthening the team, the division, and the organization.
Raison d'être
The Physics of Organizations
Organizational behavior is structural. These are the laws every organization must obey.
The Organizational Capability
What determines capability? Not headcount, not tooling — the quality of the shared model.
The System 3 Age
What changes when AI produces fluent answers and deliberation must be deliberately protected.
What we study
From tools for one person to intelligence the organization can run.Learn from the individual. Compound at every scale.
Most AI is a personal assistant. We research systems that learn from how people actually work, then return that as capability — for the person first, then as collaborative, real-time intelligence for the team, the division, and the institution. Shared models. Traceable commitment. Learning that compounds instead of resetting.
Our partner, Reflective Group, builds the applications that put this research to work.
Research program
What we work on.
Post-training
Adapting open-source models to an organization’s language, evidence, and decision patterns. The model should learn the institution, not the other way around.
Agent loops
Closed loops that act, check, and revise. Organizational work is iterative; the intelligence that serves it has to be too.
Dynamic formations
Agents that reconfigure as the work does — around a person, a team, a division, the whole. Formation is a research problem, not a chat session.
Research stack
Four layers. One program.
Axioms™
The constraint layer: invariant rules and governing assumptions a shared organizational model must remain consistent with. Formal principles, not slogans.
Helms™
The authority layer: where humans exercise judgment over machine agency, ratify commitments, and keep accountable control of consequential decisions.
Converge™
The epistemic kernel: how distributed evidence is brought into a shared, revisable model, and how commitment is formed under uncertainty.
Organism™
The organizational intelligence runtime: structured intent, adversarial planning, coordinated execution, and dynamic formations of agents across distributed work.