Introduction

An OpenAI-compatible API that returns a hallucination probe score for every token it generates.

Telluvian serves language models with a hallucination probe attached. The probe reads the model's hidden states as it generates, so every token comes back with a score saying how likely that token is to be unsupported.

The API is OpenAI-compatible. If you can call OpenAI, you can call this — change the base URL and the key.

https://api.telluvian.ai/v1

What you get

Per-token scores. Not a single number for the whole response: one score per generated token, so you can point at the specific claim that looks wrong.
Native and routed models. Our own Gemma runs the probe on the same forward pass that produced the token. Models from other providers are replayed through our model so the probe can score them too.
Streaming or not. Scores arrive per chunk when streaming, or as arrays on the completed response.