ZeroEntropy

ZeroEntropy is a San Francisco-based company that trains and serves small, specialized machine learning models for information retrieval, including rerankers (zerank-2, previously ze-rank-1/zerank-1), embedding models (zembed-1), and custom fine-tuned models such as query rewriters and context compressors. Its products are delivered as a developer API covering document ingestion, indexing, advanced querying, embedding, and reranking, with Python and TypeScript SDKs, an MCP server, VPC/self-hosted deployment (listed on AWS Marketplace and Azure), and access through partner inference providers. The company also publishes an eval harness ("MTEB Evals"), a training methodology it calls zELO (described as chess-Elo converted into a continuous relevance training signal), and technical catalogs (Concepts, Playbooks, Versus).

The company frames its thesis around "agentic retrieval" — a retrieval system that dynamically selects strategies based on query context and learns from feedback — and, more broadly, around a view that production AI will consist of many fine-tuned specialist models wrapped around frontier LLMs rather than a single general model. Its about page describes distilling a calibrated universal pairwise comparator over an axis of comparison into small pointwise scorers, with relevance as the first shipped axis and other axes (faithfulness, hedging, instruction adherence, tone, prompt-injection risk, code-bug probability, factuality, scientific rigor) as planned extensions. Enterprise posture includes SOC 2 Type II, HIPAA, GDPR, and CCPA compliance claims, plus model licensing, dedicated capacity, and SLAs.

On July 24, 2026, ZeroEntropy announced it had been acquired by Notion, open-sourced its models under Apache 2.0, and said its products would remain fully supported until September 4, 2026, publishing a migration guide for moving off the ZeroEntropy API (covering rerank/embed models, zSearch, OCR, and data backfill).