Skip to main content

VibeSys

Generating Bespoke Systems with AI Agents

VibeSys uses application requirements, workload characteristics, and hardware capabilities as inputs to an agentic search that generates a purpose-built system for each target.

Bespoke by target

Each target defines its own implementation contract, correctness checks, and performance benchmark, so VibeSys works across domains instead of assuming one runtime, language, or deployment shape.

Multi-agent optimization loop

An outer loop plans the search over designs; an inner loop implements and validates candidates. An independent judge checks correctness and reward-hacking risk before results are recorded.

Extend with skills, not forks

New model families, hardware platforms, and optimizations are added by writing an Agent Skill distilled from real serving engines and research, not by modifying the framework.

Architecture

An outer loop plans the search over designs; an inner loop implements and validates candidates; an independent judge checks correctness before results are recorded.

VibeSys architecture: an outer loop dispatches per-round tasks to an inner loop of Implementer, Accuracy Judge, and Performance Evaluator agents