Lists that keep themselves alive
A model list that nobody edits goes stale; one that fills itself can grow until it is slow, or admit something you would not choose. Lodestone does the first and bounds the second.
Outdated models drop out
A daily checkModels() saves each provider's catalogue. A model it no longer lists is dropped. One with an expiry date within a week is left out (OpenRouter publishes the date; 14 models carried one the day this was written). One failing right now is skipped by the breaker.
A pool of 8, three tried
The ranked pool lives in the background. Models that are down are filtered out, and a request tries at most three. When one fails, the next best takes its place for later requests, so adding models adds choice, not delay.
Trust is checked, not assumed
A discovered model must come from a trusted lab. You can block labs or models, refuse any a provider says is trained on or keeps prompts too long, and mark a single call private.
Illustrative: a pool of eight. The two that are down are skipped; the next three live ones are tried.
What the trust policy decides
| Model found by discovery | Default policy | Strict: no training, prompts kept ≤ 30 days |
|---|---|---|
google/gemma-4-31b-it:freeOpenRouter, publishes nothing on data use | used | refused: provider does not say what happens to prompts |
shady-lab/free-llm:freeOpenRouter, a lab nobody has heard of | refused: lab not on the trusted list | refused: provider does not say what happens to prompts |
poolside/laguna-s-2.1:freeOpenRouter, retiring on 31 Oct 2026 | refused: retiring within a week | refused: retiring within a week |
bedrock/claude-haiku-5-5Requesty: not trained on, prompts kept 0 days | used | used |
openai/gpt-6.1-terra:flexRequesty: marked as trained on | used | refused: trained on |
gemini-3.8-flashLLM API, family "gemini-flash", publishes nothing on data use | used | refused: provider does not say what happens to prompts |
Verdicts are computed by the library's own vet() as this page is built, on rows shaped like real lists. On 28 Oct 2026, of Requesty's 807 chat models, 205 came from a lab not on the default list and 9 were marked as trained on. LLM API publishes nothing about what happens to prompts, so under a strict policy none of its models can be discovered; models you name yourself still work.
What this is not. There is no vulnerability database for hosted AI models, so Lodestone does not claim to scan one. It checks what providers actually publish, and enforces your rules on it. Those facts are self-reported by providers. The security page lists exactly what is and is not checked.