@mikeo-ai/claude-context-local-core

The core indexing engine for claude-context-local - a privacy-first, 100% local code search and analysis tool using PostgreSQL + pgvector and Ollama embeddings.

🚀 Features

• 100% Local Processing: All embeddings and searches happen on your machine
• PostgreSQL + pgvector: Enterprise-grade vector storage without cloud dependencies
• Ollama Integration: Free, open-source embedding models (no API keys!)
• Privacy-First: Your code never leaves your computer
• Fast Search: 12ms average query time
• Multiple Splitters: AST-based and LangChain text splitters
• Incremental Indexing: Efficient updates with merkle tree synchronization

📦 Installation

npm install @mikeo-ai/claude-context-local-core

🔧 Usage

Basic Setup

import {
  Context,
  PostgresVectorDatabase,
  AstCodeSplitter
} from '@mikeo-ai/claude-context-local-core';

// Configure PostgreSQL with pgvector
const vectorDb = new PostgresVectorDatabase({
  host: 'localhost',
  port: 5432,
  database: 'embeddings',
  user: 'postgres',
  password: 'postgres'
});

// Initialize context with local embeddings
const context = new Context({
  vectorDatabase: vectorDb,
  embeddingProvider: 'Ollama',
  ollamaModel: 'DC1LEX/nomic-embed-text-v1.5-multimodal',
  codeSplitter: new AstCodeSplitter()
});

Index a Codebase

// Index your project
const stats = await context.indexCodebase('/path/to/your/project');
console.log(`Indexed ${stats.indexedFiles} files, ${stats.totalChunks} chunks`);

Search Code

// Semantic search through your code
const results = await context.semanticSearch(
  '/path/to/your/project',
  'authentication logic',
  10, // limit
  0.7  // similarity threshold
);

results.forEach(result => {
  console.log(`${result.relativePath}:${result.startLine}`);
  console.log(`Similarity: ${result.score}`);
  console.log(result.content);
});

🏗️ Architecture

Vector Databases

• PostgresVectorDatabase: Local PostgreSQL with pgvector extension (recommended)
• MilvusVectorDatabase: Original Milvus support (requires cloud/docker)
• MilvusRestfulVectorDatabase: RESTful API for Milvus

Embedding Providers

• OllamaEmbedding: Local Ollama models (recommended for privacy)
• OpenAIEmbedding: OpenAI API (requires API key)
• VoyageAIEmbedding: VoyageAI API (requires API key)
• GeminiEmbedding: Google Gemini API (requires API key)

Code Splitters

• AstCodeSplitter: Syntax-aware splitting using AST parsing
• LangChainCodeSplitter: Language-aware text splitting

⚙️ Configuration

Environment Variables

# Embedding Provider (use Ollama for local)
EMBEDDING_PROVIDER=Ollama
OLLAMA_MODEL=DC1LEX/nomic-embed-text-v1.5-multimodal
OLLAMA_HOST=http://localhost:11434

# PostgreSQL Configuration
POSTGRES_HOST=localhost
POSTGRES_PORT=5432
POSTGRES_DATABASE=embeddings
POSTGRES_USER=postgres
POSTGRES_PASSWORD=postgres

# Optional: Splitter Configuration
SPLITTER_TYPE=ast  # or 'langchain'
CHUNK_SIZE=2500
CHUNK_OVERLAP=300

🔄 Migration from Original

If migrating from the original @zilliz/claude-context-core:

// Old (cloud-based)
import { Context, MilvusVectorDatabase } from '@zilliz/claude-context-core';
const context = new Context({
  vectorDatabase: new MilvusVectorDatabase({ /* Milvus cloud config */ }),
  embeddingProvider: 'OpenAI',
  openAIApiKey: 'sk-...'
});

// New (100% local)
import { Context, PostgresVectorDatabase } from '@mikeo-ai/claude-context-local-core';
const context = new Context({
  vectorDatabase: new PostgresVectorDatabase({ /* local Postgres */ }),
  embeddingProvider: 'Ollama',
  ollamaModel: 'DC1LEX/nomic-embed-text-v1.5-multimodal'
});

📊 Performance

| Operation | Cloud Version | Local Version | |-----------|---------------|---------------| | Embedding Generation | 200-500ms | 50-100ms | | Vector Search | 60-250ms | 12ms | | Index 1000 files | ~5 min | ~2 min | | Storage per 1M chunks | Cloud costs | Free (local disk) |

🔗 Related Packages

• @mikeo-ai/claude-context-local-mcp - MCP server for Claude Code integration
• Main Repository - Full documentation and examples

🤝 Contributing

Contributions welcome! This is a community project focused on privacy and local-first development.

📄 License

MIT © 2025 MikeO-AI

Based on original work by Zilliz, also MIT licensed.

Privacy Guarantee: This package performs ALL operations locally. No telemetry, no external API calls (except optional cloud providers if explicitly configured), no data collection.