Providers
Any service that speaks OpenAI's chat API works through openaiCompatible(), and presets exist for the ones Lodestone has been checked against. Here is exactly how far each check went.
| Preset | Service | What was checked |
|---|---|---|
groq() | Groq | checked a chat call answered live on 28 Oct 2026 (openai/gpt-oss-20b, about 1 s); the model list was read, 11 entries of which 4 are chat models the list also holds speech and safety models, which Lodestone now leaves out; the model Lodestone first listed, llama-3.3-70b-versatile, had been withdrawn (a 404, reported as withdrawn) |
requesty() | Requesty | partly key accepted and the model list read live (807 chat models, with price, lab and data policy) the chat call was refused with 402, payment required, because the account had no credit, so no answer has been seen from it |
llmapi() | LLM API | checked model list read live (152 chat of 393 models; image, video, speech and embedding ignored) chat call not made: no key |
cerebras() | Cerebras | partly endpoint answers (needs a key); free-tier terms not checked untested with a key |
openai(), mistral(), together() | the providers themselves | partly endpoint answers (needs a key) untested with a key |
ollama() | a local server | partly the standard local endpoint untested here |
(built in) openrouter | OpenRouter | checked used in production for the default path catalogue, calls, cost and expiry dates all seen live |
anthropic(), google() | direct | partly request shape tested against the documented format not called live |
A "model list read" means Lodestone's reader was run against the service's real list and parsed it correctly. A chat call has been seen to answer only on Groq (and on OpenRouter, in production); the others are tested against fakes of the documented request, not against the service itself. A live check on 28 October 2026 is how the withdrawn Groq model and the non-chat entries in its list were found, which is why those are in the notes. Requesty and LLM API have no free models; they add redundancy on the paid path. Free redundancy comes from providers with free tiers, such as Groq and Cerebras.
Checking a provider yourself: node scripts/live-check.mjs in the repository calls each one once with keys from your environment and prints what happened, without printing a key.