What it does not do yet
Lodestone is for one-shot text and JSON calls today. These are the gaps, stated plainly.
Streaming
complete() and json() wait for the whole answer. Streaming needs a rule for a model that fails halfway; the likely one is to fail over only before the first token.
Tool calling, embeddings, speech
Lodestone is for one-shot text and JSON calls today. It does not let a model call functions, and it has no tier for embeddings or speech.
The Python port is alpha
It can't read provider model lists other than OpenRouter's, so Requesty's price, lab and data-policy details and discover are TypeScript-only, and a private call uses a fallback only if you vouch for it. A hung request is cut off at its slice and the next model is tried, but its thread only ends at its socket timeout. Images, streaming, tool calling and scorecards do not exist in Python. It has been tested against a local HTTP server and fake providers, and has made live calls to Groq and OpenRouter, which both answered; Requesty was refused for lack of credit and no other provider has been tried live from Python, which is the main reason it is alpha. Details.
Also worth knowing: only Groq and OpenRouter have been seen to answer a live call, so every other provider is checked against fakes. The Providers page says exactly how far each check went, and Where it runs says what has been deployed where.