Zilliz & Milvus

1. Company Identity

2. Founder & Original Creators

Charles Xie — founder, CEO and publicly identified inventor of Milvus

• Classification: Founder of Zilliz; original creator/inventor of Milvus according to Zilliz's current executive biography.
• Current role: Founder & CEO, Zilliz.
• Education: Master's degree in Computer Science, University of Wisconsin–Madison.
• Earlier career: Software engineering at Oracle; Zilliz describes him as a founding engineer of the Oracle 12c cloud database project.
• Expertise: Database systems, distributed systems, search, vector databases, AI infrastructure and cloud databases.
• Milvus contribution: Zilliz's official biography explicitly says Xie invented Milvus. The 2021 SIGMOD paper lists him as a co-author alongside the engineering team.
• Public DOB, birthplace, nationality: Not publicly available in the authoritative sources reviewed.
• Public net worth: Not publicly available; no reliable figure is included.

Milvus visual supplied for the profile.

Original technical contributors

The evidence does not support a claim that Milvus was the work of a single programmer. The landmark 2021 SIGMOD paper lists a large Zilliz engineering/research author group: Jianguo Wang, Xiaomeng Yi, Rentong Guo, Hai Jin, Peng Xu, Shengjun Li, Xiangyu Wang, Xiangzhou Guo, Chengming Li, Xiaohai Xu, Kun Yu, Yuxing Yuan, Yinghao Zou, Jiquan Long, Yudong Cai, Zhenxiang Li, Zhifeng Zhang, Yihua Mo, Jun Gu, Ruiyi Jiang, Yi Wei and Charles Xie. The paper describes the system's architecture and implementation and therefore provides strong evidence for the collective technical contribution, but it does not establish that every author was a founder or inventor.

3. Milvus Origin Story

Zilliz was founded in 2017. The team began developing what became Milvus in 2018, before the phrase "vector database" had become a mainstream product category. The underlying problem was not simply storing data: machine-learning systems increasingly converted text, images, audio, video and other unstructured information into high-dimensional numerical representations. Existing databases and standalone approximate-nearest-neighbor libraries did not provide a production-oriented system for managing that data at scale while supporting updates, metadata filtering, distributed execution and operational reliability.

According to Zilliz's historical engineering account, early users asked whether the technology could accelerate image search. That feedback helped the team recognize that vectors were a more fundamental abstraction than any one AI model or GPU workload. The project evolved into an open-source database centered on vector similarity search.

Milvus 0.10 was open-sourced in November 2019. In April 2020 it entered the LF AI & Data Foundation as an incubation-stage project. It graduated in 2021. Milvus 1.0 followed in 2021, while the team then made a difficult architectural decision: instead of continuing to patch 1.x, it would rebuild the system around cloud-native principles. Milvus 2.0 became that major redesign and reached general availability in January 2022.

4. Company & Milvus Timeline

5. What Is Milvus?

A vector database stores numerical representations of data and makes it efficient to find items that are similar to a query. A machine-learning model can turn a sentence, image, product, sound clip or document into an embedding — a list of numbers that captures useful semantic information.

For example, the sentences "a red sports car" and "a fast crimson automobile" may be represented by vectors that are close together even though the words are not identical. Milvus can search those vectors and return the nearest matches.

Milvus is not an LLM and does not normally generate the final natural-language answer. In a RAG system, an embedding model creates vectors, Milvus retrieves relevant context, and an LLM can then use that context to generate an answer.

6. Technology Architecture

High-level architecture

Current Milvus documentation describes a cloud-native, disaggregated architecture with an access layer, coordinator, worker nodes and storage. The current architecture separates streaming processing from historical/batch processing and uses shared storage so compute components can scale independently.

Vector indexing and search

Milvus integrates vector search libraries and algorithms including Faiss, HNSW, DiskANN and SCANN. Its Knowhere execution layer provides a common vector-index execution interface and hardware-aware selection between CPU and GPU execution. Current documentation also covers sparse and dense vectors, hybrid retrieval, metadata filtering, multi-vector data, and text search.

Supported similarity measures include Euclidean/L2, inner product and cosine for floating-point vectors, plus binary metrics. Index strategies vary by workload: graph-based methods can favor high recall and speed, while quantization methods reduce memory at some accuracy cost.

GPU acceleration

Milvus supports GPU indexes such as GPU_CAGRA, GPU_IVF_PQ, GPU_IVF_FLAT and GPU_BRUTE_FORCE. The documentation notes that GPU acceleration is particularly useful for high-throughput or high-recall workloads and that GPU is not automatically faster for every latency-sensitive query pattern.

APIs and languages

Milvus exposes REST and gRPC APIs and provides SDKs including Python, Java, Go, C# and Node.js. PyMilvus is the primary Python client and also provides the Milvus Lite path.

Cloud-native dependencies

Milvus can use Kubernetes for distributed deployment. Its documented storage/dependency ecosystem includes etcd for metadata, object storage such as S3/MinIO/Azure Blob, and WAL implementations including Woodpecker, Kafka and Pulsar.

7. Milvus Products & Ecosystem

8. Milvus 3.0 — From Vector Database to Lake-Native Retrieval

Milvus 3.0 is the most important recent architectural milestone. The 3.0 beta arrived May 9, 2026; Milvus 3.0.0 was released July 29, 2026.

Why this matters

Earlier vector databases were often positioned as a serving layer sitting beside a data lake. Milvus 3.0 attempts to reduce that separation: data can stay in the lake while retrieval infrastructure builds the indexes and serves semantic queries over it. The strategic implication is a move from "vector database as another copy of data" toward "retrieval engine operating over the data platform."

Milvus visual supplied by the publisher. The 3.0 discussion above is based on Milvus release notes and official announcement materials.

9. AI Use Cases

10. Customers & Adoption

Public customer counts have changed over time and should not be mixed across years. In 2022, TechCrunch reported that Zilliz said Milvus downloads had passed one million and production users had grown 300% year-over-year; named customers included eBay, Tencent, Walmart, IKEA, Intuit and Compass. Current Milvus materials describe deployments at more than 300 major enterprises and list organizations such as Salesforce, PayPal, Shopee, Airbnb, eBay, NVIDIA, IBM, AT&T, LINE, Roblox and Inflection.

11. Business Model

The business model separates the open-source project from the commercial service:

• Milvus OSS: free software under Apache 2.0. Users can operate it themselves on local machines, Docker or Kubernetes.
• Zilliz Cloud: managed Milvus service. Zilliz monetizes the operational layer: managed compute, storage, reliability, scaling, security, enterprise controls and support.
• BYOC: Zilliz Cloud can deploy the data plane inside a customer's cloud account/VPC for sovereignty and governance requirements.
• Usage-based economics: Current Zilliz Cloud documentation describes pay-as-you-go and usage-based charging for compute, storage, data transfer and additional services. Enterprise plans add higher reliability and governance features.
• Developer-led distribution: Open source functions as a product-discovery and trust channel. A recent Zilliz executive interview says roughly 80% of Zilliz Cloud customers began as Milvus users; this is a company-reported figure.

Why the model is strategically attractive

Open source lowers adoption friction and lets developers validate the technology independently. The commercial service then monetizes the difficult operational work that enterprises may not want to own: upgrades, scaling, failure recovery, tuning, compliance and support.

12. Funding & Investors

Total reported funding: $113M was announced by Zilliz in 2022. Some commercial databases report slightly different totals due to unattributed or database-recorded rounds; the company-announced $113M figure is preferred.

13. Competitive Landscape

There is no single universal winner. Milvus is strongest when vector retrieval is a core infrastructure workload and the organization needs distributed scaling, multiple index types, hybrid search and deep operational control. A simpler embedded database or PostgreSQL extension can be more rational when the dataset and operational requirements are modest.

14. Technical Research & Publications

The 2021 SIGMOD publication is particularly important because it provides a technical, peer-reviewed description of the system and names the engineering contributors rather than presenting Milvus as the invention of a single person.

15. Open-Source Community

• Milvus is an LF AI & Data project and its source repository is under the milvus-io GitHub organization.
• The project uses the Apache 2.0 license.
• The repository has accumulated more than 43,000 GitHub stars according to a recent Zilliz retrospective, reflecting substantial developer interest.
• Milvus uses GitHub, documentation, Discord/community channels, events and office hours as major community interfaces.
• Milvus has moved through LF AI & Data incubation and graduation, giving the project a governance home separate from Zilliz's commercial interests.

16. Leadership

For other requested executive titles such as CFO, COO, VP Engineering and Head of Research, a current authoritative public roster was not sufficiently available in the sources reviewed. They are therefore not guessed or inferred.

17. Company Culture & Philosophy

• Engineering-first: The historical Milvus story emphasizes solving difficult database and systems problems before a mature "vector database" category existed.
• Open source as product development: Zilliz uses community feedback and real workloads as a feedback loop.
• Developer-first distribution: GitHub and technical education have historically been important acquisition channels.
• Cloud-native thinking: Milvus 2.0 and 3.0 both emphasize separation of storage and compute, elastic scaling and cloud object storage.
• Hardware awareness: CPU SIMD and GPU acceleration are treated as first-class performance considerations.
• Production over demos: The company increasingly frames vector search around reliability, governance, cost and long-lived enterprise workloads.

Detailed internal hiring processes, benefits, diversity statistics and employee satisfaction metrics are not sufficiently disclosed in the authoritative sources reviewed.

18. Awards & Recognition

• LF AI & Data graduation: Milvus graduated from the foundation's project lifecycle in 2021.
• BigANN challenge: Zilliz's project retrospective says Milvus won the billion-scale vector search challenge in 2021.
• Industry recognition: Zilliz says it was named a leader by Forrester during 2024–2025. This is a vendor-reported recognition and should not be confused with a government or academic certification.
• Community scale: More than 43,000 GitHub stars are cited by Zilliz's recent project retrospective.

19. Challenges & Controversies

Documented engineering challenges

• Milvus 1.x had scalability, availability, storage-cost and operational limitations that helped motivate the 2.0 rewrite.
• Distributed vector search is difficult because vector data is memory-intensive and search, indexing and ingestion can have very different resource profiles.
• GPU indexes have workload-specific constraints; GPU is not automatically the best option for every query.
• Running distributed Milvus requires database and Kubernetes expertise, while Milvus Lite exists specifically to reduce this barrier for smaller applications.

Business challenges

• The vector database market has become crowded with purpose-built systems, PostgreSQL extensions and search engines adding vector capabilities.
• Open-source monetization requires a commercial layer that creates enough value beyond self-hosting to justify subscription/usage fees.
• The rapid evolution of embeddings, rerankers, LLMs and retrieval methods means the database must continuously adapt.

Legal / controversy review

No major public legal case, regulatory enforcement action or sustained controversy involving Milvus/Zilliz was identified in the authoritative sources reviewed for this profile. This is not a claim that no dispute has ever existed; it means no material case was found that met the verification threshold for inclusion.

20. SWOT Analysis

21. Milvus vs Traditional Databases

22. Key Innovations

23. Founder / Inventor Profile: Charles Xie

Charles Xie is the central founder figure in the Zilliz/Milvus story. His background combines database engineering and startup building. Zilliz's current biography identifies him as founder and CEO, says he previously worked at Oracle as a founding engineer of the Oracle 12c cloud database project, and states that he invented Milvus. He holds a master's degree in computer science from the University of Wisconsin–Madison.

The significance of his contribution is best understood as a systems thesis: AI would increasingly create high-dimensional representations that conventional databases were not optimized to store and search. Instead of treating vector search as an isolated algorithmic feature, the Milvus team treated it as a database-management problem involving persistence, indexing, dynamic updates, distributed execution, hardware utilization and operational reliability.

His public commentary in 2026 continues this systems-oriented approach. Xie argues that vector databases will expand beyond simple nearest-neighbor retrieval into broader vector computing workloads such as clustering and classification, while emphasizing lower infrastructure costs as a prerequisite for broader AI adoption.

24. 50 Important Facts

25. Final Summary

Lessons for Entrepreneurs & AI Infrastructure Teams

1. Build around a fundamental abstraction. Milvus was built around vectors as a durable data abstraction rather than a temporary LLM feature.
2. Open source can be a distribution strategy. The community can become a product-testing, education and trust engine when the software solves a real problem.
3. Do not confuse adoption with monetization. A large open-source community does not automatically create a commercial business; the paid layer must remove meaningful operational pain.
4. Design for the next scale, but provide a simple starting point. Milvus Lite, Standalone and Distributed create a migration path from experiment to production.
5. Rebuild when the architecture is fundamentally wrong. The Milvus 2.0 rewrite shows the cost of preserving an architecture that no longer matches customer requirements.
6. Retrieval quality is a systems problem. Embedding models, indexing, filtering, sparse retrieval, reranking, storage and hardware all affect the final AI experience.
7. Storage economics matter. Milvus 3.0's lake-native direction is partly a response to the cost and governance problems created by copying large data sets.
8. Infrastructure products need operational maturity. Performance benchmarks matter, but reliability, recovery, upgrades, security and cost control determine whether an AI database becomes mission-critical.
9. Know when not to use a specialized database. For small datasets, a PostgreSQL extension or embedded vector store may be simpler and more economical.
10. Keep the project and business identities clear. Governance through an open-source foundation helps preserve trust while the commercial company builds managed products around the project.

Key Research Conclusions

• Milvus is best understood as a database-engineering project that anticipated the AI retrieval infrastructure market rather than as a product created specifically for today's RAG boom.
• Charles Xie is the primary founder figure and is explicitly identified by Zilliz as Milvus's inventor, but the technical system was created by a substantial engineering team.
• The 2020 LF AI & Data relationship and 2021 graduation gave Milvus an independent open-source governance context.
• The 2022 Milvus 2.0 redesign was a major technical turning point: compute/storage disaggregation and independently scalable components made the system much more suitable for cloud production.
• Milvus 3.0 changes the strategic category again by allowing retrieval over lake-resident data and by moving more ranking, filtering and aggregation work into the retrieval engine.
• Zilliz's commercial strategy is a classic open-core/managed-service pattern: open source builds adoption and trust; managed infrastructure monetizes operational complexity.
• The strongest long-term competitive question is not whether Milvus can perform vector search, but whether it can remain economically superior, reliable and easy enough as databases, search engines and cloud platforms absorb vector capabilities.

Sources & Verification

The report was researched against primary and high-quality sources available through August 22, 2026. Claims that are company-reported are identified as such. Unknown personal details and private-company financial metrics are explicitly left undisclosed.

1. Milvus Release Notes — v3.0.0 and v3.0-beta
2. Milvus: Announcing Milvus 3.0
3. Milvus Architecture Overview
4. Milvus Deployment Options
5. Milvus Lite Documentation
6. What is Milvus / current product overview
7. Milvus GPU Index Documentation
8. Milvus Glossary / Knowhere
9. Milvus 2.0 architecture engineering article
10. Milvus 2.0 redesign article
11. SIGMOD 2021 — Milvus: A Purpose-Built Vector Data Management System
12. Zilliz CEO Charles Xie — vector database evolution and philosophy
13. Zilliz CEO Charles Xie — AI and VectorDB future
14. Zilliz — how the company built for production AI
15. TechCrunch — Zilliz $43M Series B, 2020
16. TechCrunch — Zilliz $60M and Silicon Valley move, 2022
17. Business Wire — Zilliz $60M Series B extension
18. Linux Foundation 2020 Annual Report — LF AI & Data / Milvus
19. AWS — Zilliz strategic cloud provider announcement, 2025
20. Zilliz Cloud deployment and plan comparison
21. Zilliz Cloud cost model
22. Zilliz Cloud global/enterprise capabilities
23. Milvus use cases and public customer references
24. Milvus GitHub repository
25. Milvus Lite GitHub repository