OmniAgent
AMD AI Agent Hackathon

OmniAgent

Hybrid Multi-Capability AI Agent. Local-first inference, confidence-aware fallback to Fireworks AI, and self-verification before every answer.

Hybrid Routing

A task router classifies every request and dispatches it to the right capability automatically.

Local vLLM

Generation runs local-first on a vLLM server for low-latency, on-prem inference.

Fireworks AI

Production-grade hosted models serve as the always-available cloud inference tier.

Confidence-aware Fallback

Low-confidence or failed local responses transparently retry on Fireworks — no request is lost.

Provider-aware Verification

Every answer is verified before it returns, by the same provider family that generated it.

Docker Ready

Ships as a container with server and batch modes — one image for the API and offline runs.

How it works

One request, routed, generated locally when possible, verified before it returns.

JSON TaskPOST /agent/tasksTask Routercapability matchCapabilityqa · math · codegen …Hybrid Providerlocal-firstLocal vLLMprimaryFireworks AIconfidence fallbackVerificationprovider-awareverified response