wtf is relaywatch?

RelayWatch is a dashboard that keeps track of AI API relay stations, the third party services that resell access to models like ChatGPT, Claude, Gemini, and DeepSeek at their own prices. Instead of manually checking many different relay sites, RelayWatch discovers them across the web, collects their listed models and prices, and organizes everything into one searchable and comparable directory. The main feature is model price comparison: for any given AI model, you can see which relay site offers it, at what input, output, and caching price, along with each site's success rate, response delay, and throughput. Alongside pricing, RelayWatch aggregates each site's public announcements, such as maintenance notices, price changes, and promotions, and keeps a timeline of them. It also pulls in the official status pages of major AI providers, so you can tell whether a slowdown is caused by the relay site itself or by the upstream provider. A separate detection feature lets you use your own API key to test whether a specific site and model actually work as advertised. The backend is built with Python and FastAPI, and the frontend uses React with Vite. Data can be stored either as plain local JSON files for quick setup, or in a PostgreSQL database for larger, longer running deployments, with a safe switch over process so a failed import does not break the running data. To get started, you install the Python and frontend dependencies, normalize your collected site data into the data folder, and start the FastAPI server with Uvicorn. The repository only contains source code, example configuration, and documentation, real API keys, cookies, admin tokens, and database passwords are deliberately excluded and must be supplied through your own environment variables. The project is released under the MIT license.