Finalist
Zeppelin
Zeppelin is a production-grade, S3-native vector search engine built entirely by Claude Opus 4.6 in long-running agentic sessions.
Core trick
Keeps vectors directly in S3 so stateless nodes scale horizontally with no inter-node coordination.
Tags
Description
Zeppelin is a production-grade, S3-native vector search engine built entirely by Claude Opus 4.6 in long-running agentic sessions.
21,000 lines of Rust, 451 tests, 26 TLA+ formal proofs for a complete distributed systems architecture.
Opus autonomously planned, implemented, debugged, and iterated for hours at a stretch, discovering invisible concurrency bugs through probabilistic reasoning (computing exact failure distributions across namespaces), writing TLA+ specs that caught bugs and races no test could find and uncovered framework-level precision bugs in Rust's serde library.
In the end we have a working database that stores vectors in S3, scales horizontally with zero coordination with performance in the same order of magnitude as that of mature database companies.
Check it out at https://zepdb.github.io/
I have hosted a Zeppelin instance with an embedding endpoint for everyone to play around with.
Install it as a skill by doing: "curl -sL https://zepdb.github.io/install.sh | sh"
README excerpt
<p align="center"> <img src="assets/icon.svg" alt="Zeppelin" width="128" height="128"> </p> <h1 align="center">Zeppelin</h1> <p align="center"> <a href="https://www.rust-lang.org"><img src="https://img.shields.io/badge/Rust-1.85+-DEA584?logo=rust&logoColor=white" alt="Rust: 1.85+"></a> <a href="https://opensource.org/licenses/Apache-2.0"><img src="https://img.shields.io/badge/License-Apache--2.0-blue.svg" alt="License: Apache-2.0"></a> <a href="https://github.com/zepdb/zeppelin/actions/workflows/ci.yml"><img src="https://github.com/zepdb/zeppelin/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://codecov.io/gh/zepdb/zeppelin"><img src="https://codecov.io/gh/zepdb/zeppelin/branch/main/graph/badge.svg" alt="codecov"></a> </p> <p align="center"> An S3-native vector search engine. Object storage is the source of truth. Nodes are stateless. </p> --- ## Why Zeppelin Most vector databases replicate your data across SSD-backed nodes and keep indexes memory-resident — you pay for that fleet whether you query it or not. Zeppelin stores everything (vectors, indexes, WAL, manifest) in object storage and keeps nodes stateless. A warm dense query is one conditional manifest GET plus two bounded waves of parallel range reads — a quantized coarse scan, then an exact rerank — so compute scales with query volume, storage scales at object-store prices, and any node, including one that just booted, can serve any namespace. There is no rebalancing, no replica fleet,