taladb

Stores document embeddings and vector data directly on the client device using WebAssembly-based indexing, eliminating the need for cloud vector database infrastructure. Implements in-process vector storage with support for semantic search without external API calls, using a hybrid approach that combines dense vector indices with document metadata storage in a single local database instance.

Combines traditional full-text document search with vector similarity matching, using a two-stage ranking pipeline that first filters by keyword relevance then re-ranks by semantic similarity. Implements hybrid search by maintaining parallel indices — a text inverted index for keyword matching and a vector index for semantic queries — with configurable weighting between both signals.

Provides a fluent TypeScript query builder API with full type inference for document schemas, catching query errors at compile time rather than runtime. Implements generic type parameters to ensure filter predicates, sort fields, and projections match the document schema, with IDE autocomplete for all query operations.

Supports adding, updating, and removing documents from the vector index without full re-indexing, using delta tracking to identify changed documents and update only affected index entries. Implements incremental index maintenance with optional background compaction to reclaim space from deleted documents.

Provides an abstraction layer for embedding models that supports multiple providers (OpenAI, Hugging Face, local ONNX models) with a unified API, allowing applications to switch embedding providers without changing database code. Implements caching of computed embeddings to avoid redundant API calls and supports batch embedding requests for efficiency.

Provides unified storage API that abstracts over browser IndexedDB, React Native AsyncStorage, and Node.js file system, with automatic schema versioning and migration support. Implements a storage adapter pattern that detects the runtime environment and selects the appropriate backend, while maintaining a consistent query interface across all platforms and handling schema evolution through versioned migrations.

Implements operational transformation or CRDT-based synchronization to keep local document state in sync across multiple clients and tabs, with automatic conflict resolution using configurable merge strategies. Detects concurrent edits, applies transformations to maintain consistency, and provides hooks for custom conflict resolution logic when automatic merging fails.

Enables filtering and querying documents based on semantic similarity to a query embedding, supporting range queries on vector distance and multi-field filtering combined with vector similarity. Implements vector distance calculations (cosine, euclidean) with optional metadata filtering, allowing developers to find documents semantically similar to a query without full-text matching.

Supports bulk insertion and updating of documents with vector embeddings, providing progress callbacks and error handling for large-scale indexing operations. Implements batched writes to optimize storage performance, with checkpointing to resume interrupted indexing and detailed progress reporting for long-running operations.

Implements full-text search across multiple document fields using configurable tokenization and stemming strategies, building inverted indices for fast keyword matching. Supports field-specific boosting to weight matches in title differently than body text, and provides query operators (AND, OR, NOT) for complex boolean queries.

Provides React hooks (useQuery, useDocument, useSearch) that automatically re-render components when queried documents change, implementing reactive data binding between the database and React component state. Hooks handle subscription management, cleanup, and memoization to prevent unnecessary re-renders, integrating with React's concurrent rendering and Suspense.

Implements a React Native storage adapter using AsyncStorage with automatic JSON serialization and deserialization, handling the async-only nature of AsyncStorage while providing a synchronous-looking API through batching and caching. Manages storage quota limits and provides utilities for data migration between storage backends.

Implements a server-side storage backend using Node.js file system with write-ahead logging (WAL) to provide ACID guarantees for document persistence. Uses atomic file operations and transaction logs to ensure data consistency even in case of process crashes, with optional compression for reducing disk usage.