Lab
AGI-SAC
Aligned Group Integration — System Architecture Coexistence
A Python research instrument for observing alignment failure modes that only appear at the system level in stateful, multi-agent deployments with shared memory, tool access, and long interaction histories. It is research infrastructure for controlled experiments, not a production system, and it makes no claims about consciousness.
Visual toy, not the research model. The animation below is a simple flocking sketch I made to build intuition about cohesion versus noise. It is not AGI-SAC's experimental code, and its readouts are illustrative, not measurements.
Attraction toward coherent center
Random perturbation magnitude
Number of agents in the swarm
How the sketch works
Each dot follows local rules: drift toward the centre (cohesion, α), avoid crowding neighbours, and absorb random kicks (noise, Ω). Raise the noise past the cohesion and the group scatters; the label flips from STABLE to TURBULENT or CRITICAL.
That is the intuition AGI-SAC was built to test properly: whether coordination in a multi-agent system holds up under perturbation. The real instrument uses language-model agents, repeatable protocols, and measured coordination metrics. See the documentation.