Repo of the Day
taladb/taladb
Published: Nov 21, 2026
Open repository ↗TalaDB - An open-source embedded vector and document database for building local-first AI applications
Summary
TalaDB is an open-source embedded database that stores JSON-like documents, metadata, and vectors in the same Rust-backed engine. It gives browser, Node.js, React Native, Android, iOS, macOS, and Rust applications local document filtering, full-text search, vector similarity, and hybrid retrieval without requiring a database server.
What it is useful for
TalaDB is aimed at local-first applications where sending embeddings or search data to a hosted service is undesirable or impractical. A concrete example is an offline support-article app: keep articles in a collection, store an embedding beside each record, and find the five most relevant English support articles with one filtered vector query. The filter is applied before ranking, so the requested top five still satisfy the metadata constraints.
It also covers conventional document workloads: MongoDB-like filters and updates, secondary indexes, insert/update/delete operations, ACID transactions backed by redb, and live-query subscriptions. Retrieval is not limited to vectors. searchText uses BM25 for keyword ranking, while hybridSearch combines keyword and vector result ranks with reciprocal rank fusion—useful when one query contains both a rare SKU and a semantic paraphrase.
The same engine is exposed through a unified TypeScript API and native Kotlin and Swift packages. Exact k-nearest-neighbor search is the default; HNSW is optional for workloads that need approximate search at scale. A relevant limitation is that TalaDB does not replicate data. Its change webhook is an at-most-once notification, and the README says webhook delivery is implemented only by the TypeScript client, not Kotlin, Swift, or Rust.
How engineers can use it
For TypeScript applications, install taladb and the binding for the target runtime:
pnpm add taladb @taladb/web # browser
pnpm add taladb @taladb/react-native # React Native
pnpm add taladb @taladb/node # Node.js
React and React Native applications can optionally add @taladb/react for hooks such as useFind and useFindOne.
Then open a database and work with a typed collection:
import { openDB } from 'taladb'
const db = await openDB('myapp.db')
const articles = db.collection<Article>('articles')
In a browser this uses OPFS; Node.js and React Native use a file. The README demonstrates generating embeddings locally with Transformers.js, creating a vector index with articles.createVectorIndex('embedding', { dimensions: 384 }), inserting documents with their vectors, and querying with articles.findNearest('embedding', query, 5, { category: 'support', locale: 'en' }).
Rust projects can instead run cargo add taladb. The README also provides Android, iOS/macOS, and Rust guides with platform requirements and native examples. Engineers should follow the guide for their runtime before implementation, especially when adding encryption, migrations, live queries, or webhooks.