Local models for speed and privacy. Cloud models for heavy reasoning. Automatic routing between them.
Pure cloud is expensive and creates privacy dependencies. Pure local lacks the reasoning power of frontier models. The hybrid approach gives you the best of both: fast, private processing for routine tasks and powerful cloud inference for complex reasoning.
Simple lookups, classification, summarization of short texts, and structured formatting run on local models. These tasks don't need GPT-4-class reasoning, and running them locally means faster responses, lower costs, and no data leaving your environment.
Complex analysis, multi-step reasoning, nuanced writing, and tasks requiring broad knowledge route to cloud models. The system selects the appropriate model based on task requirements — not every cloud task needs the most expensive model.
You don't need to decide where a task runs. The system evaluates complexity, data sensitivity, and cost to make the optimal routing decision for each request.