GitHub - alicoding/semantic-search-service

 ███████ ███████ ███    ███  █████  ███    ██ ████████ ██  ██████ 
 ██      ██      ████  ████ ██   ██ ████   ██    ██    ██ ██      
 ███████ █████   ██ ████ ██ ███████ ██ ██  ██    ██    ██ ██      
      ██ ██      ██  ██  ██ ██   ██ ██  ██ ██    ██    ██ ██      
 ███████ ███████ ██      ██ ██   ██ ██   ████    ██    ██  ██████ 
                                                                   
 ███████ ███████  █████  ██████   ██████ ██   ██                  
 ██      ██      ██   ██ ██   ██ ██      ██   ██                  
 ███████ █████   ███████ ██████  ██      ███████                  
      ██ ██      ██   ██ ██   ██ ██      ██   ██                  
 ███████ ███████ ██   ██ ██   ██  ██████ ██   ██

Enterprise-grade semantic search powered by LlamaIndex PropertyGraphIndex with TRUE 95/5 architecture.

837 lines of code. 25 Python modules. One unified intelligence layer.

🎯 What This Actually Does

Semantic Search Service is a complete intelligence layer for your development workflow:

🔍 Semantic Code Search - Search your codebase semantically, not just text matching
🧠 Conversation Memory - Index and search your Claude/AI conversations
📊 Knowledge Graphs - PropertyGraphIndex creates entity relationships from your code
🔄 Business Logic Extraction - Automatically extract business rules and workflows
⚡ Real-time Integrations - Sub-100ms responses for tools like temporal-hooks and task-enforcer
🎨 Auto-documentation - Generate API docs and diagrams automatically
🌐 Multiple Interfaces - FastAPI REST, CLI, and MCP (Model Context Protocol) for Claude

🚀 Quick Start

git clone https://github.com/alicoding/semantic-search-service.git
cd semantic-search-service

# Copy and configure
cp .env.example .env
# Add your OPENAI_API_KEY or ELECTRONHUB_API_KEY

# Start everything
./setup.sh

# Index your project
./semantic-search index . my-project

# Search semantically
./semantic-search search "authentication logic" my-project

✨ Core Features

🔍 Semantic Search

# Index any codebase
./semantic-search index /path/to/project project-name

# Semantic search (not just text matching)
./semantic-search search "error handling patterns" project-name

# Check if components exist (for task-enforcer integration)
./semantic-search exists "AuthService" project-name

# Find SOLID/DRY violations (for temporal-hooks integration)
./semantic-search violations project-name

🧠 Conversation Memory

# Index your Claude conversations
curl -X POST "http://localhost:8000/index/conversations" \
  -H "Content-Type: application/json" \
  -d '{"path": "/path/to/conversations", "collection": "my-conversations"}'

# Search your conversation history
curl "http://localhost:8000/search/memory?query=authentication&limit=5"

📊 Knowledge Graphs

# Generate knowledge graph from codebase
curl "http://localhost:8000/graph/my-project"

# Export to NetworkX format
curl "http://localhost:8000/graph/my-project/export"

# Visualize relationships
curl "http://localhost:8000/graph/my-project/visualize"

🎨 Auto-documentation

# Generate API documentation
python src/core/auto_docs.py generate

# Generate sequence diagrams
curl -X POST "http://localhost:8000/diagram/sequence?project=my-project"

# Extract business logic
curl -X POST "http://localhost:8000/extract/business-logic?project=my-project"

🛠️ Configuration

Everything is configured via config.yaml:

# LLM Configuration - Works with OpenAI, ElectronHub, or Ollama
llm_provider: openai  # or ollama for offline
embed_provider: openai
openai_model: claude-opus-4-1-20250805  # ElectronHub models supported!

# Ollama for offline/enterprise
ollama_model: llama3.1:latest
ollama_base_url: http://localhost:11434

# Performance
num_workers: 4
cache_ttl: 3600

# Vector Store
qdrant_url: http://localhost:6333
redis_host: localhost
redis_enabled: true

📡 API Endpoints

Core Search:

POST /search/{project} - Semantic search in project
POST /index - Index new project
GET /exists - Check component existence
GET /violations/{project} - Find code violations

Conversation Memory:

POST /index/conversations - Index Claude/AI conversations
GET /search/memory - Search conversation history

Knowledge Graphs:

GET /graph/{project} - Get project knowledge graph
GET /graph/{project}/visualize - Generate visualizations

Business Intelligence:

POST /extract/business-logic - Extract business rules
POST /diagram/sequence - Generate sequence diagrams

Integrations:

GET /check/violation - Real-time violation check (temporal-hooks)
GET /context/project - Project context (AI agents)

🔌 Integrations

temporal-hooks Integration

Real-time violation detection during development:

GET /check/violation?action=create-new-service&context=my-project
# Returns violations instantly (<100ms cached)

task-enforcer Integration

Check if components exist before creating tasks:

GET /exists?component=UserService&project=my-app
# Returns: {"exists": true, "confidence": 0.92, "file": "user_service.py"}

Claude MCP Integration

Works directly in Claude Code sessions as MCP tools:

search_code - Semantic search in projects
check_exists - Component existence checking
find_violations - SOLID/DRY violation detection

🏗️ Architecture

TRUE 95/5 Pattern:

95% LlamaIndex Native - PropertyGraphIndex, StorageContext, Settings
5% Glue Code - Thin wrappers and configuration

Tech Stack:

LlamaIndex - PropertyGraphIndex, VectorStoreIndex, query engines
Qdrant - Vector database (enterprise-grade)
Redis - Sub-100ms caching
FastAPI - REST API with auto-generated docs
Typer - Rich CLI interface

📊 Performance

Feature	Target	Status
Search Response	<500ms	✅ <200ms
Violation Check	<100ms	✅ Cached
Component Exists	<200ms	✅ <100ms
Conversation Search	<500ms	✅
Knowledge Graph	<3s	✅

🔧 Installation

Prerequisites

Python 3.8+
Docker (for Qdrant)
OpenAI API key OR Ollama (offline mode)

Full Setup

# Clone repository
git clone https://github.com/alicoding/semantic-search-service.git
cd semantic-search-service

# Configure environment
cp .env.example .env
# Edit .env with your API keys

# Automated setup
./setup.sh

# Test installation
curl http://localhost:8000/docs

Docker Setup

# Start services
docker-compose up -d

# Index sample project
./semantic-search index . sample-project

# Test search
./semantic-search search "FastAPI endpoints" sample-project

🧪 Testing Your Conversations

Got Claude conversations? Index and search them:

# If you have claude-parser installed
curl -X POST "http://localhost:8000/index/conversations" \
  -H "Content-Type: application/json" \
  -d '{"path": "/path/to/conversations", "collection": "my-chats"}'

# Search your conversation history
curl "http://localhost:8000/search/memory?query=how to implement caching&limit=3"

🐛 Known Issues

We track all issues publicly with proper DoR/DoD:

#23 🚨 CRITICAL: Missing health endpoint
#24 🚨 CRITICAL: API doesn't initialize LlamaIndex Settings
#25 ⚠️ HIGH: Rate limiting crashes git hooks

See PRODUCTION_READINESS_AUDIT.md for complete analysis.

🤝 Contributing

Found a bug? Create an issue with:

Clear reproduction steps
Expected vs actual behavior
Your environment (Ollama/OpenAI, OS, etc.)

We fix real bugs that real users encounter.

📚 Documentation

PRODUCTION_READINESS_AUDIT.md - Current production issues
CAPABILITY_MAPPING.md - What actually works vs. claimed
docs/ - Detailed guides and API references

🎯 Real-World Example

Before: Grep for "authentication" returns 500 text matches
After: Semantic search finds actual auth patterns, business logic, and related components with confidence scores

./semantic-search search "user authentication flow" my-project
# Returns: AuthService.authenticate() method, login flow, JWT handling, etc.
# With semantic understanding, not just text matching

Built with LlamaIndex native patterns. Stop searching. Start finding. 🎯