ZeroEntropy and Zilliz Cloud Integration
Integrate Zilliz Cloud with ZeroEntropy to build high-accuracy semantic search using two-stage vector retrieval and reranking.
Use this integration for FreeWhat is ZeroEntropy
ZeroEntropy is an AI search infrastructure company that provides embedding and reranking models through a developer-friendly API.
It offers two core models:
zembed-1(Embedding model): Converts text into vector embeddings that capture semantic meaning. Supports configurable output dimensions — 40, 80, 160, 320, 640, 1,280, or 2,560 — so you can balance accuracy against storage cost. zembed-1 can reduce vector database storage costs by up to 10x compared to full-dimension alternatives.zerank-2(Reranking model): A cross-encoder that scores query-document pairs for relevance. Rather than relying on vector distance alone, it reads the query and document together to judge how well they match.zerank-2also supports instruction-following: you can pass natural-language instructions to guide ranking behavior without retraining the model.
Both models are open-weight and available through the ZeroEntropy API and Python SDK.
How ZeroEntropy Embeddings and Reranking Work with Zilliz Cloud
ZeroEntropy and Zilliz Cloud work together in a two-stage retrieval pipeline:
Stage 1 — Embed and retrieve. You use
zembed-1to generate vector embeddings for your documents, then store them in a Zilliz Cloud collection. When a query comes in, you embed it with the same model and run a vector similarity search in Zilliz Cloud to pull the top candidates. This step is fast, even across millions of documents.Stage 2 — Rerank for accuracy. You pass the candidates from Stage 1 to
zerank-2. The reranker evaluates each candidate against the query in detail and reorders them by relevance. This second pass catches results that are semantically close but not actually relevant, and surfaces ones that are.
The result: Zilliz Cloud handles scale and speed, ZeroEntropy handles precision.
Why Use ZeroEntropy with Zilliz Cloud to Build AI Apps
Higher search accuracy with two-stage retrieval. Vector similarity search finds documents that are close in embedding space, but close doesn't always mean relevant. Adding ZeroEntropy's reranker as a second stage improves result quality by scoring true relevance rather than relying on vector distance alone.
Lower storage costs with flexible embedding dimensions.
zembed-1supports output dimensions from 40 to 2,560. Smaller embeddings take up less space in your Zilliz Cloud collection, reducing storage and memory costs. When paired with reranking, you can use compact embeddings without losing search quality.Domain-specific ranking without model retraining.
zerank-2accepts natural-language instructions alongside the query — for example, "prioritize practical techniques over theory" or "rank by recency." This lets you adapt the same pipeline to different use cases (legal research, customer support, medical retrieval) without fine-tuning or retraining.Simple integration. The full pipeline — embed, store, search, rerank — takes a few lines of Python using the
zeroentropyandpymilvusSDKs. Zilliz Cloud handles indexing automatically once you insert your embeddings.Scalable vector storage and retrieval. Milvus is built to index and search billions of vectors. As your document collection grows, Zilliz Cloud scales automatically to handle the increased data volume and query load, so ZeroEntropy's reranker always has a fast first-stage retrieval to work with.
Managed infrastructure with automatic indexing. Zilliz Cloud handles index building, resource provisioning, and performance tuning out of the box. You insert your ZeroEntropy embeddings; Zilliz Cloud takes care of the rest — no manual index configuration or infrastructure management required.
How to Integrate ZeroEntropy with Zilliz Cloud
If you use open-source Milvus, just install the ZeroEntropy Python SDK (
zeroentropy) and the Milvus client library (pymilvus). If you use Zilliz Cloud (the managed Milvus), it's much easier. Embedding and reranking models are integrated internally; no external dependencies are required.Generate vector embeddings for your documents using
zembed-1, specifying your preferred dimension size.Insert the embeddings into a Zilliz Cloud or Milvus collection. Zilliz Cloud handles indexing automatically.
For each query, embed it with
zembed-1and search the collection in Zilliz Cloud to retrieve the top candidates.Pass the candidates to
zerank-2to rerank by relevance. Optionally, include a natural-language instruction to customize ranking behavior.
Learn How to Use ZeroEntropy with Zilliz Cloud
The best way to start is with a hands-on tutorial. This tutorial walks you through how to build a two-stage semantic search pipeline using ZeroEntropy's embedding and reranking models with Milvus.
Tutorial: Semantic Search with ZeroEntropy and Milvus
And here are a few more resources: