Developer releases Hallx, a risk-scoring tool to catch hallucinations and silent failures in LLM outputs across OpenAI, Anthropic, and Gemini.

Hallx validates three key aspects of LLM outputs: schema matching, consistency across multiple runs, and alignment with provided context

Returns a confidence score and risk level to help identify problematic outputs before they propagate through pipelines

Supports multiple LLM providers including OpenAI, Anthropic, Gemini, and Ollama with both synchronous and asynchronous operations

Uses heuristic-based scoring rather than guarantees; accuracy depends on the quality of the input context provided

Available as an open-source tool via pip install with star support on Hacker News (1 point, 1 comment at posting)