Sloop

Sloop is a New York-based generative AI company that develops "hypertokens," which it describes as a subset of standard tokens used as information landmarks that act simultaneously as reference points and error-correcting codes for AI models. The company packages this as a context builder that injects hypertokens into prompts to construct what it calls a context map, with the stated goal of improving in-context recall, semantic search, fine-tuning alignment, and generative reasoning across any large language model.

The company positions hypertokens as applicable across three stages of an AI workflow: embedding (index and search quality), fine-tuning (aligning a model to a specific domain), and inference (prompt optimization for chat-style models). It frames the problem it targets as the family of AI failure modes commonly called hallucination, including in-context recall mistakes, foundational knowledge errors, and reasoning errors, and states that hypertokens can reduce such errors to a desired level of precision. Stated use cases include due diligence document review, precision medicine insights, and review of construction blueprints.

Sloop describes itself as in stealth, with plans to publish an arXiv paper presenting proof of its method at the time it exits stealth, and offers case studies on request in the interim. Its website promotes a free trial and beta sign-up.