zilliztech/deep-searcher is an open-source project that searches and reasons over internal, non-public data to produce answers and reports. Its goal is to apply the deep research approach — bundling search and reasoning into a single flow — to enterprise data. It is written in Python and released under the Apache-2.0 license. The README describes the repository as an open-source deep research alternative.
The tool combines multiple LLMs such as OpenAI or DeepSeek with vector databases such as Milvus and Zilliz Cloud (databases that convert text into numbers so items with similar meaning can be found) to query, evaluate and reason over non-public data. You can choose among several types of embedding models and LLMs, and local files can be loaded as well, which makes it suitable for enterprise knowledge management, intelligent question answering and information retrieval. The repository is recorded as having 8350 stars, with a creation date of February 8, 2025.
That said, there is no official release yet, and no comparative performance figures or separate test materials are presented. The description of its supported scope is a claim grounded in the README. The momentum score also came out low, at 3.0 out of 10. So it reads as a candidate for enterprise document search scenarios, but it is hard to treat it as a definitive choice as long as performance evidence has not been published.