Phoenix Router

The Phoenix Router is a kind of agent that helps with routing. It was made for the AMD Developer Hackathon Track 1 challenge. This agent does not send every request to a computer in the cloud. Instead it first figures out what kind of task it is dealing with. It uses rules to decide which of eight categories the task falls into: knowing facts, doing math understanding how someone feels, summarizing things identifying names fixing code mistakes using logic and generating code. Most of the time the Phoenix Router can handle requests on its own without needing the cloud. It uses a model called Qwen3.5-2B that runs on the computers processor. This model is very good at doing its job. The agent is smart. Can change how it works based on how hard the task is and how fast it can do it. This way it can make sure everything gets done within the time limit of 10 minutes. If it gets stuck or is running out of time the Phoenix Router will switch to a model from Fireworks. This helps save on using up tokens in the cloud without making the agent less reliable. The Phoenix Router was designed to work in a challenging environment with limited resources. It has to work with 2 virtual processors and 4 GB of memory. With these limits the project can always produce a valid output and handle errors nicely. It never fails during runtime. Always makes a results.json file. By being smart about how it routes tasks and manages resources and by using the cloud when necessary the Phoenix Router is very accurate and uses almost no tokens from Fireworks. This makes it a practical and efficient agent that is ready, for real-world use.