Design and run supervised multi-agent inference cascades using idle local GPU capacity. Spawns parallel agents with read-wide write-narrow sandboxing, collects observations as append-only hash-chained deltas, and triangulates findings from model agreement and disagreement. Use when running autonomou
Design and run supervised multi-agent inference cascades using idle local GPU capacity. Spawns parallel agents with read-wide write-narrow sandboxing, collects observations as append-only hash-chained deltas, and triangulates findings from model agreement and disagreement. Use when running autonomous research loops, evaluating model performance against a live corpus, or exploring a problem space w