We built IMURG because we kept running into the same problem while working with AI agents: even when an agent has access to web search, the model can still confidently make things up. SIMURG is an open-source middleware that adds a verification layer to the generation process. It monitors the model, detects when a response is likely hallucinating, aborts the generation, searches the web for evidence, and retries with grounded information. The goal was to make this as simple as possible for developers. You can plug it into an OpenAI-compatible endpoint and get started with: pip install simurg SIMURG is completely free and open-source. I'd love to hear how other people are handling hallucinations in their AI agents, and what you'd like to see added next.

SIMURG

Hallucination detecting for AI models and free web-search.

Hallucination detecting for AI models and free web-search.