OmniAgent

## OmniAgent: A Hybrid Multi-Capability AI Agent OmniAgent is a production-ready hybrid AI agent that intelligently routes every request to the most appropriate AI model instead of sending all tasks to a single cloud LLM. The system combines a local vLLM server running on AMD GPU hardware with the Fireworks API to achieve lower costs, faster responses, and high-quality results. When a user submits a request, OmniAgent first identifies the task type, such as question answering, summarization, sentiment analysis, named entity recognition, mathematical reasoning, code generation, debugging, or general reasoning. Lightweight and deterministic tasks are executed locally through vLLM, while complex reasoning tasks are automatically routed to Fireworks. This hybrid routing strategy significantly reduces unnecessary cloud token usage and improves overall efficiency. To ensure reliable outputs, OmniAgent includes a verification layer with confidence scoring, provider-aware validation, structured JSON schema validation using Zod, and automatic fallback. If a local response has low confidence or the local server is unavailable, the system seamlessly retries the request using the cloud provider without requiring user intervention. The project is built with NestJS and TypeScript using a modular, scalable architecture that makes it easy to add new AI providers or capabilities in the future. It is fully containerized with Docker, supports batch execution for the AMD AI Hackathon Track 1 requirements, and includes more than 300 automated tests for reliability. By combining intelligent task routing, local-first inference, automatic verification, and cloud fallback, OmniAgent delivers a practical, scalable, and cost-efficient AI platform that reduces cloud dependency while maintaining high accuracy and production-ready performance.