The FAIRouter library, designed for routing requests between various language models, is now available as a NuGet package. It automatically selects a model based on task requirements, evaluates responses, and learns from user feedback.
The library supports .NET 10 and allows developers to integrate models via OpenRouter or OpenAI-compatible APIs. Users can define specific model lists or choose from 54 pre-configured options. The system stores weights and interaction logs in a local database file.
Feedback is provided on a scale from 0 to 1, enabling the router to refine its performance over time through training epochs. Developers can select from three predefined profiles—Quality, Balance, and Price—or set a sufficiency threshold to optimize for cost and speed.
Beyond simple text generation, FAIRouter supports multi-turn dialogues and custom tools. Users can define agents with specific parameters for input/output costs and token processing speeds, allowing the router to manage complex workflows alongside standard LLM queries.
The project is distributed under the Apache-2.0 license. Source code and documentation are hosted on GitHub, with the package version 0.1.0 currently available for implementation.
