asdecided/WayfinderRouter

Simple CLI tool for deterministic routing of queries between local and hosted LLM models

What it solves

Wayfinder provides a unified local endpoint for developers to access multiple AI models. It eliminates the need to manually switch between different local runtimes, account-backed models, and API-based models by automatically routing requests to the most eligible model based on a deterministic scoring system.

How it works

At its core is a portable Router written in Rust that acts as a gateway. It discovers available local models via loopback catalogs and validates them with inference probes. When a request comes in, the Router uses an offline, keyless scoring path to decide which model to use. Credentials are only resolved at the time of delivery to the chosen provider. It exposes OpenAI-compatible and Anthropic-compatible endpoints, allowing tools like Codex, Claude Code, and Aider to connect to a single loopback policy.

Who it’s for

It is primarily designed for developers using the Omarchy ecosystem, as well as users of coding agents and those who want a centralized, privacy-focused way to manage local and hosted AI models on macOS, Linux, or via Docker.

Highlights

  • Deterministic Routing: Uses a scored decision path that remains offline and keyless.
  • Multi-Provider Support: Integrates local runtimes, Apple Foundation Models, Anthropic, and OpenAI.
  • Broad Compatibility: Provides OpenAI and Anthropic compatible gateways for easy integration with existing AI clients.
  • Cross-Platform: Available as an Omarchy plugin, a native macOS app, a standalone Rust binary, and a Docker container.

Related

  • Project
  • Project
  • Project
  • Project
  • Project