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 gets its own contract, checks, and benchmark, not a fixed runtime, language, or deployment shape

Multi-agent optimization loop

An outer loop searches over designs while an inner loop implements, validates, and judges each candidate

Extend with skills

New model families, hardware, and optimizations arrive as Agent Skills, not framework changes

Architecture

Click a use case on the left to see its input, system, and result on the right.

An outer loop searches over designs; an inner loop implements and validates each one.
USE CASES
USER INPUTStarting pointModel, weights,or a checkoutHardwareDevice specNVIDIAAMDIntelAppleAmazonWorkloadBenchmark metricsVibeSysOUTER LOOPSearch PolicyDesign decisionsBranch / backtrackSearch StateGit stateFeedback historyBase Commit + TaskCommit + FeedbackINNER LOOPErrorImplementerAccuracy JudgeProfilerSkills LibraryExecution EnvironmentDockerModalBESPOKE SYSTEMserving_system/├─ api_server.py├─ scheduler.py├─ router.py├─ cache.py├─ model.py└─ backend.pytests/├─ test_accuracy.py└─ test_router.pybench/├─ microbench.py├─ latency.csv└─ tpt.csv