The cognitive architecture, open to inspection.
IRIS is not a black box. Every decision passes through a layered architecture with explicit purpose: reasoning, memory, verification, synthesis, language and action. Every layer observes, records and can be audited.
This is what makes it possible to build enterprise systems on AI without losing traceability or control.
Six layers with explicit purpose.
The six cognitive modes
Cognitive
Structured reasoning over the problem. Identifies entities, relations, constraints and goals.
Reactive
Immediate response when context is clear and verified. No over-deliberation.
Proactive
Detection of relevant events before the user requests them. Continuous watch.
Deliberative
Activation when the problem has ambiguity or risk. Multiple agents confront hypotheses.
Adaptive
Strategy adjustment based on feedback and memory. Does not repeat known errors.
Iterative
Progressive improvement over previous conclusions when new information appears.
Specialized agents working in parallel.
The swarm is not a metaphor: it is architecture. Reasoner, critic, verifier, director and memory operate in parallel, confront hypotheses and reach consensus before emitting any output.
External LLMs as tools, not as brain.
IRIS orchestrates models like Claude, GPT or Gemini as technical infrastructure. Deliberation, memory and decisions belong to the engine. External models are swappable; the cognitive architecture is not.