AgentDB Vector Search

AgentDB 向量搜索

此技能的功能

使用 AgentDB 高性能向量数据库实现基于向量的语义搜索，操作速度比传统方案快 150 倍至 12,500 倍。支持 HNSW 索引、量化和亚毫秒级搜索（<100µs）。

前提条件

• Node.js 18+
• AgentDB v1.0.7+（通过 agentic-flow 或独立安装）
• OpenAI API 密钥（用于生成嵌入向量）或自定义嵌入模型

使用 CLI 快速入门

初始化向量数据库

# Initialize with default dimensions (1536 for OpenAI ada-002)
npx agentdb@latest init .$vectors.db

# Custom dimensions for different embedding models
npx agentdb@latest init .$vectors.db --dimension 768  # sentence-transformers
npx agentdb@latest init .$vectors.db --dimension 384  # all-MiniLM-L6-v2

# Use preset configurations
npx agentdb@latest init .$vectors.db --preset small   # <10K vectors
npx agentdb@latest init .$vectors.db --preset medium  # 10K-100K vectors
npx agentdb@latest init .$vectors.db --preset large   # >100K vectors

# In-memory database for testing
npx agentdb@latest init .$vectors.db --in-memory

查询向量数据库

# Basic similarity search
npx agentdb@latest query .$vectors.db "[0.1,0.2,0.3,...]"

# Top-k results
npx agentdb@latest query .$vectors.db "[0.1,0.2,0.3]" -k 10

# With similarity threshold (cosine similarity)
npx agentdb@latest query .$vectors.db "0.1 0.2 0.3" -t 0.75 -m cosine

# Different distance metrics
npx agentdb@latest query .$vectors.db "[...]" -m euclidean  # L2 distance
npx agentdb@latest query .$vectors.db "[...]" -m dot        # Dot product

# JSON output for automation
npx agentdb@latest query .$vectors.db "[...]" -f json -k 5

# Verbose output with distances
npx agentdb@latest query .$vectors.db "[...]" -v

导入/导出向量

# Export vectors to JSON
npx agentdb@latest export .$vectors.db .$backup.json

# Import vectors from JSON
npx agentdb@latest import .$backup.json

# Get database statistics
npx agentdb@latest stats .$vectors.db

使用 API 快速入门

import { createAgentDBAdapter, computeEmbedding } from 'agentic-flow$reasoningbank';

// Initialize with vector search optimizations
const adapter = await createAgentDBAdapter({
  dbPath: '.agentdb$vectors.db',
  enableLearning: false,       // Vector search only
  enableReasoning: true,       // Enable semantic matching
  quantizationType: 'binary',  // 32x memory reduction
  cacheSize: 1000,             // Fast retrieval
});

// Store document with embedding
const text = "The quantum computer achieved 100 qubits";
const embedding = await computeEmbedding(text);

await adapter.insertPattern({
  id: '',
  type: 'document',
  domain: 'technology',
  pattern_data: JSON.stringify({
    embedding,
    text,
    metadata: { category: "quantum", date: "2025-01-15" }
  }),
  confidence: 1.0,
  usage_count: 0,
  success_count: 0,
  created_at: Date.now(),
  last_used: Date.now(),
});

// Semantic search with MMR (Maximal Marginal Relevance)
const queryEmbedding = await computeEmbedding("quantum computing advances");
const results = await adapter.retrieveWithReasoning(queryEmbedding, {
  domain: 'technology',
  k: 10,
  useMMR: true,              // Diverse results
  synthesizeContext: true,    // Rich context
});

核心功能

1. 向量存储

// Store with automatic embedding
await db.storeWithEmbedding({
  content: "Your document text",
  metadata: { source: "docs", page: 42 }
});

2. 相似度搜索

// Find similar documents
const similar = await db.findSimilar("quantum computing", {
  limit: 5,
  minScore: 0.75
});

3. 混合搜索（向量 + 元数据）

// Combine vector similarity with metadata filtering
const results = await db.hybridSearch({
  query: "machine learning models",
  filters: {
    category: "research",
    date: { $gte: "2024-01-01" }
  },
  limit: 20
});

高级用法

RAG（检索增强生成）

// Build RAG pipeline
async function ragQuery(question: string) {
  // 1. Get relevant context
  const context = await db.searchSimilar(
    await embed(question),
    { limit: 5, threshold: 0.7 }
  );

  // 2. Generate answer with context
  const prompt = `Context: ${context.map(c => c.text).join('\n')}
Question: ${question}`;

  return await llm.generate(prompt);
}

批量操作

// Efficient batch storage
await db.batchStore(documents.map(doc => ({
  text: doc.content,
  embedding: doc.vector,
  metadata: doc.meta
})));

MCP 服务集成

# Start AgentDB MCP server for Claude Code
npx agentdb@latest mcp

# Add to Claude Code (one-time setup)
claude mcp add agentdb npx agentdb@latest mcp

# Now use MCP tools in Claude Code:
# - agentdb_query: Semantic vector search
# - agentdb_store: Store documents with embeddings
# - agentdb_stats: Database statistics

性能基准测试

# Run comprehensive benchmarks
npx agentdb@latest benchmark

# Results:
# ✅ Pattern Search: 150x faster (100µs vs 15ms)
# ✅ Batch Insert: 500x faster (2ms vs 1s for 100 vectors)
# ✅ Large-scale Query: 12,500x faster (8ms vs 100s at 1M vectors)
# ✅ Memory Efficiency: 4-32x reduction with quantization

量化选项

AgentDB 提供多种量化策略，以提高内存使用效率：

二值量化（缩减 32 倍）

const adapter = await createAgentDBAdapter({
  quantizationType: 'binary',  // 768-dim → 96 bytes
});

标量量化（缩减 4 倍）

const adapter = await createAgentDBAdapter({
  quantizationType: 'scalar',  // 768-dim → 768 bytes
});

乘积量化（缩减 8 至 16 倍）

const adapter = await createAgentDBAdapter({
  quantizationType: 'product',  // 768-dim → 48-96 bytes
});

距离度量

# Cosine similarity (default, best for most use cases)
npx agentdb@latest query .$db.sqlite "[...]" -m cosine

# Euclidean distance (L2 norm)
npx agentdb@latest query .$db.sqlite "[...]" -m euclidean

# Dot product (for normalized vectors)
npx agentdb@latest query .$db.sqlite "[...]" -m dot

高级功能

HNSW 索引

• O(log n) 搜索复杂度
• 亚毫秒级检索（<100µs）
• 自动构建索引

缓存

• 1000 个模式的内存缓存
• <1ms 模式检索
• 自动缓存失效

MMR（最大边际相关性）

• 结果集更加多样
• 避免冗余
• 兼顾相关性与多样性

性能提示

1. 启用 HNSW 索引：AgentDB 会自动启用，速度快 10 至 100 倍
2. 使用量化：二值量化（32 倍）、标量量化（4 倍）、乘积量化（8 至 16 倍）可减少内存占用
3. 批量操作：批量插入速度快 500 倍
4. 匹配维度：1536（OpenAI）、768（sentence-transformers）、384（MiniLM）
5. 相似度阈值：从 0.7 开始以保证质量，并根据使用场景调整
6. 启用缓存：为频繁查询启用 1000 个模式的缓存

故障排除

问题：搜索性能较慢

# Check if HNSW indexing is enabled (automatic)
npx agentdb@latest stats .$vectors.db

# Expected: <100µs search time

问题：内存使用量过高

# Enable binary quantization (32x reduction)
# Use in adapter: quantizationType: 'binary'

问题：相关性较差

# Adjust similarity threshold
npx agentdb@latest query .$db.sqlite "[...]" -t 0.8  # Higher threshold

# Or use MMR for diverse results
# Use in adapter: useMMR: true

问题：维度不匹配

# Check embedding model dimensions:
# - OpenAI ada-002: 1536
# - sentence-transformers: 768
# - all-MiniLM-L6-v2: 384

npx agentdb@latest init .$db.sqlite --dimension 768

数据库统计信息

# Get comprehensive stats
npx agentdb@latest stats .$vectors.db

# Shows:
# - Total patterns$vectors
# - Database size
# - Average confidence
# - Domains distribution
# - Index status

性能特征

• 向量搜索：<100µs（HNSW 索引）
• 模式检索：<1ms（使用缓存时）
• 批量插入：插入 100 个向量耗时 2ms
• 内存效率：量化后内存占用减少 4 至 32 倍
• 可扩展性：可高效处理 1M+ 个向量
• 延迟：大多数操作均为亚毫秒级

了解更多

• GitHub：https:/$github.com$ruvnet$agentic-flow$tree$main$packages$agentdb
• 文档：node_modules$agentic-flow/docs/AGENTDB_INTEGRATION.md
• MCP 集成：使用 npx agentdb@latest mcp 集成 Claude Code
• 网站：https:/$agentdb.ruv.io
• CLI 帮助：npx agentdb@latest --help
• 命令帮助：npx agentdb@latest help <command>