EdgeVectorDB: An AI-Native Database MVP (Edge-Optimized)

EdgeVectorDB is an AI-native database MVP built in Rust, optimized for edge computing use cases. It leverages SQLite as its base and provides capabilities for local vector search, low-latency inference data caching, and offline semantic retrieval.

Features

• Language: Rust
• Base: SQLite (via rusqlite)
• Vector Storage: Supports vector data column type with distance metrics (cosine, Euclidean, dot-product).
• Hybrid Query Engine: Designed for hybrid SQL + semantic queries.
• API: gRPC and REST API for basic operations (insert, search, delete).
• Edge Optimizations:
  • Offline embedding with ONNX inference.
  • SQLite WAL + async batching for edge write durability.
  • Optimized binary size with strip, lto, musl compile.
  • Static binaries for ARM64, x86_64.
  • Flat file export/import (.edgevec format).
• CLI Tool: Command-line interface for interacting with the database.

Project Structure

• core/: Contains the core logic for storage, vector indexing, embedding, and query parsing.
• api/: Implements the gRPC and HTTP API layers.
• bindings/: (Future) Optional Python bindings.
• cli/: Provides a local CLI tool for interacting with EdgeVectorDB.

Getting Started

Prerequisites

• Rust (latest stable version)
• build-essential (for SQLite compilation)
• protobuf-compiler (for gRPC)
• cmake and clang (for ONNX runtime)
• libssl-dev (for ONNX runtime)
• gcc-aarch64-linux-gnu and g++-aarch64-linux-gnu (for ARM64 cross-compilation)

Build

To build the project, navigate to the EdgeVectorDB directory and run:

cargo build --release

This will build the api and cli binaries in target/release/.

Running the API Server

To run the gRPC and REST API server:

./target/release/api

The gRPC server will listen on [::1]:50051 and the REST server on 127.0.0.1:3000.

Using the CLI Tool

Insert Data:

./target/release/cli insert --file <path_to_data_file> --db <database_file>

Example data file (sample_data.txt):

1.0,2.0,3.0
4.0,5.0,6.0
7.0,8.0,9.0

Query Data:

./target/release/cli query "<comma_separated_vector>" --k <num_results> --db <database_file>

Export Data:

./target/release/cli export --output <output_file.edgevec> --db <database_file>

Import Data:

./target/release/cli import --input <input_file.edgevec> --db <database_file>

Demo

A simple demo script demo.sh is provided to showcase the CLI tool's functionality:

./demo.sh

This script will:

1. Build the CLI tool.
2. Insert sample data from sample_data.txt into demo.db.
3. Query for similar vectors.
4. Export the database to demo.edgevec.

Docker

A Dockerfile is provided for building multi-architecture Docker images:

docker build -t edgevectordb .

Local Benchmarks

(To be added)