Wayfinder Router Shapes Up as Deterministic Gate Between Local and Cloud LLMs

An open source CLI tool called Wayfinder Router reached 230 GitHub stars in its first week, giving developers a deterministic switchboard for sending LLM queries to either local or hosted models. The tool, published by developer itsthelore, arrived as the industry debate over local versus cloud AI intensifies.

Wayfinder Router uses a YAML configuration file to define routing rules. Instead of relying on a central proxy or a third party service, it sits on the developer's machine and inspects each query for keywords, model size, or prompt length before deciding where to execute it. The project's GitHub repository shows 188 commits as of early this week.

How deterministic routing works

Unlike inference gateways that use load balancing or latency heuristics, Wayfinder Router applies exact pattern matching. Developers write rules such as: if the prompt contains "passwords" or "source code," route to a local Ollama instance. Everything else goes to a hosted API like OpenAI or Anthropic. The tool supports multiple back ends simultaneously and can fall back to a remote provider when the local model is overloaded.

• Routes queries based on keyword, model size, and prompt length rules defined in YAML
• Supports local back ends including Ollama and llama.cpp
• Supports hosted back ends including OpenAI, Anthropic, and any OpenAI compatible API
• Performance benchmarks included in the repository show sub 10 millisecond routing overhead per decision
• No external dependencies beyond Go runtime; binary is roughly 12 MB

Developer response and use cases

The tool landed on Hacker News with three points and zero comments, but the GitHub star count grew steadily over the following days. Developers who clone the repo cite two primary use cases: slashing API costs for routine queries and keeping proprietary code out of third party inference engines. A SitePoint tutorial published concurrently walks through using Claude Code with Ollama through a similar local first approach, further signaling demand for local LLM workflows.

Wayfinder Router does not attempt any semantic understanding of the prompt. Critics argue that keyword based routing will miss nuanced contexts, while supporters counter that deterministic rules are auditable and predictable. The project's maintainer has not indicated whether future versions will add semantic classifiers or machine learned routing. For now, the approach mirrors the philosophy of tools like Teleport and Tailscale: explicit control beats black box decisions.