Rauto

The Hybrid Token-Efficient Routing Agent (Rauto) solves the "Dynamic Model Problem" by intelligently orchestrating user queries across a configurable pool of local and cloud-based Large Language Models. Instead of defaulting to expensive frontier models for every task, Rauto employs a fast, local helper model to instantly classify query domains and dynamically assess difficulty. It powers its Cost-Aware Decision Engine using a mathematical Bayesian Belief Fusion framework. This engine merges static empirical capability priors (derived from industry benchmarks like MMLU and HumanEval) with dynamic evidence retrieved from a local vector-based Experience Memory database. If the system encounters a highly novel query, it intelligently falls back to a 3-Layer Graph Memory structure to map abstract reasoning requirements to the best-suited model. Furthermore, Rauto ensures maximum accuracy via a rigorous verification layer. Utilizing an AST Code Sandbox and a local LLM Judge with majority-vote heuristics, it autonomously self-corrects failures before returning an answer. The result is a highly scalable, Docker-containerized orchestration pipeline that slashes API costs without sacrificing output quality.