Hybrid Token Efficient Routing Agent

The Hybrid Token-Efficient Routing Agent is a command-line AI system designed to process a collection of natural-language tasks autonomously. Its primary purpose is to reduce remote token usage while maintaining the level of answer accuracy required by the evaluation process. For every task, the system extracts the prompt, identifies its category, estimates its complexity, and selects an appropriate language model. Simple tasks are handled by a bundled local model whenever possible, avoiding remote token costs. Tasks that require stronger mathematical, logical, factual, or programming capabilities are routed to an allowed model through the Fireworks AI service. The system operates inside a Docker container, receives tasks through a structured JSON input file, and records concise answers in a JSON output file. Its routing rules are deterministic and inexpensive, allowing model selection without consuming additional language-model tokens. The application is intended to run without interactive user input and relies on environment variables for model availability, API access, file locations, and local inference settings. This hybrid approach balances answer quality, execution time, and token efficiency.