@smartpointstech/spt-sptdb-onnx

SPT SptDB with optimized ONNX embeddings - 100% local, GPU-accelerated AI agent memory

Overview

This package provides a high-performance AI agent memory system that combines ONNX Runtime for local embedding generation with SPT's vector storage via sptflo-memory. It supports GPU acceleration, batch embedding, and includes ReflexionMemory and ReasoningBank for agent learning workflows. A CLI is included for database management.

Installation

npm install @smartpointstech/spt-sptdb-onnx

Key Features

• Local ONNX Embeddings -- generates embeddings on-device using models like all-MiniLM-L6-v2
• GPU Acceleration -- optional GPU-backed inference via ONNX Runtime
• ReflexionMemory -- episodic memory with critique, reward, and success tracking
• ReasoningBank Integration -- pattern storage via sptflo-memory
• CLI Tool -- spt-sptdb-onnx command for database operations and benchmarks

Dependencies

Key internal dependencies:

• @smartpointstech/sptflo-memory -- database creation, ReasoningBank, ReflexionMemory
• onnxruntime-node -- ONNX Runtime for local model inference
• @xenova/transformers -- transformer model loading and tokenization

Quick Start

import { createONNXSptDB, ONNXEmbeddingService } from '@smartpointstech/spt-sptdb-onnx';

const sptDB = await createONNXSptDB({
  dbPath: './agent-memory.db',
  modelName: 'Xenova/all-MiniLM-L6-v2',
  useGPU: true,
});

API Reference

Exports

Export	Type	Description
createONNXSptDB	function	Create an SptDB instance with ONNX embeddings
ONNXEmbeddingService	class	ONNX-powered embedding generation service
ONNXConfig	type	Configuration for model, GPU, batch size, cache
EmbeddingResult	type	Single embedding result
BatchEmbeddingResult	type	Batch embedding result