folkbench/folkbench is a project that provides both a tool comparing the availability, response time and price of AI API relay services — that is, services delivering model APIs through an alternative route — ranking them, and a tool for switching Claude Code and Codex settings. The repository was created in October 2026, has 2,202 stars, and its primary language is Rust. There are no public releases, so the only things that can be checked this time are the repository itself and the Tauri-based desktop client Folkbench Switch code inside it.
The key thing in this repository is not the measurement engine but the public dashboard and the settings-switch client. Folkbench is a service that continuously tests and publishes the availability, response time and price of relay services, and the client included in the repository swaps the Base URL and API Key between several relay services and the official connection, then applies those values to the settings of tools such as Claude Code, Codex and Gemini CLI. Because the API keys and settings are stored on the local device, it is used to manage frequently changing relay routes in one place. Since it presents rankings and switching as a single bundle, it reads as reference material for evaluating tools for people who use agent development tools.
That said, the grounds for describing it as actual measurement cannot be confirmed within the public repository. There is no code that actually calls the relay servers to measure availability and latency, and the figures that appear on the dashboard are values obtained from the Folkbench service side. Quality-per-price rankings depend on the interests of relay service providers and on the measurement methodology, so they are hard to verify independently. The license is marked NOASSERTION, so the usage conditions cannot be known. This repository's score record is momentum 9.3, novelty 5.0, quality 5.0 and tech 4.0, which puts its technical depth on the lower side. It is useful as a settings-switching tool, but it is not at the stage where the ranking table can serve as the basis for judging quality.