Wayfinder is an open-source, deterministic routing tool that analyzes prompt complexity offline to automatically direct tasks to either local or cloud-based LLMs, optimizing performance and reducing costs.

Key Points

• Routes prompts based on structural complexity scores (0.0–1.0) without requiring additional model calls or network latency.
• Supports tiered routing, allowing users to assign different prompt difficulty levels to specific local or hosted model endpoints.
• Compatible with any OpenAI-style API client, enabling integration with tools like LibreChat, Open WebUI, and various agent frameworks.
• Includes calibration features to tune routing thresholds against user-provided datasets or automated feedback logs.
• Operates entirely offline for decision-making, ensuring sensitive API keys and prompt data remain secure and local.
• Provides a transparent "explain" mode to visualize why specific prompts were routed to a particular model tier.

Why it Matters

By eliminating the need for an LLM to decide where to route a prompt, Wayfinder removes the latency and costs typically associated with intelligent model selection. This approach allows developers to maintain high-quality outputs for complex tasks while significantly lowering expenses by offloading simple queries to cheaper, local infrastructure.