Build retrieval augmented generation search that combines vector similarity with exact filters.
Detect near duplicate documents or records as new data is ingested.
Give an AI agent memory retrieval by meaning, recency, and category without a separate vector database.
Run structured SQL queries alongside AI style similarity search on the same dataset.
| visorcraft/mongreldb | adsharma/ast-grep | codeitlikemiley/antigravity-sdk-rust | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | Rust | Rust | Rust |
| Setup difficulty | moderate | easy | hard |
| Complexity | 4/5 | 3/5 | 4/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires understanding of its indexing and query configuration options to use the AI search features correctly.
MongrelDB is a database engine, written in Rust, built to store and search large amounts of structured data efficiently while also supporting the kind of search that modern AI applications need. It can run embedded inside another program or as its own server, and it stores data in a custom column based file format designed to make writes fast and reliable, with a write ahead log so that data is not lost if something crashes mid write. What makes it interesting for AI work is that it keeps several very different kinds of search side by side on the same data. It supports normal SQL queries, exact text search for finding words or phrases, and several forms of what is called vector search, which is a way of finding items that are similar in meaning rather than identical in wording. This includes searching by dense number based embeddings, a sparse style of search suited to rare or specific terms, and a way to detect near duplicate items using set comparison techniques. All of these can be combined and filtered together, for example limiting search results to a specific user's data before ranking them by similarity. The README lists concrete uses such as retrieval augmented generation, where an AI system pulls relevant documents before answering a question, detecting duplicate content as new data comes in, and letting an AI agent recall past memories by meaning, recency, or type. It is meant to let a developer build these AI search features locally, without needing a separate specialized vector database service running elsewhere. The project sets hard limits on things like how many results or candidates a single search can consider, and lets an operator configure lower limits, timeouts, and how many searches can run at once, which suggests it is meant to be used in real production systems rather than as an experiment. It supports the C programming interface, Node.js through a native addon, and multiple other languages, and it is dual licensed under the MIT license and the Apache 2.0 license. The full README is longer than what was shown.
An embedded and server database written in Rust that combines SQL, full-text search, and multiple kinds of AI style similarity search on the same data.
Mainly Rust. The stack also includes Rust, SQL, DataFusion.
Use freely for any purpose, including commercial use, under either the MIT or Apache 2.0 license, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.