Client examples for querying codebase indexes stored in SQLite-vec and Qdrant.

These scripts demonstrate how to:

• Connect to SQLite-vec or Qdrant databases
• Generate embeddings for search queries
• Search for code chunks or git commits
• Display results with similarity scores

• search-demo.js - JavaScript/Node.js interactive search
• search-demo.py - Python interactive search
• search-test.js - Non-interactive quick test

• search-commits.js - Search git commits analyzed by LLM

# No dependencies needed - uses native fetch!

# Get all commits (max 5)
node search-commits.js codebase-908e5cbf73d44edcbc

# Search commits semantically
node search-commits.js codebase-908e5cbf73d44edcbc "cleanup and refactoring"

# Install dependencies
npm install better-sqlite3 sqlite-vec dotenv

# Run the demo
node search-demo.js

# Or make it executable
chmod +x search-demo.js
./search-demo.js

# Install dependencies
pip install sqlite-vec requests python-dotenv

# Run the demo
python search-demo.py

# Or make it executable
chmod +x search-demo.py
./search-demo.py

Both scripts read configuration from environment variables (.env file):

# Embedding Provider
EMBED_PROVIDER=openai-compatible

# Model Configuration
EMBED_MODEL=Qwen/Qwen3-Embedding-8B
EMBED_BASE_URL=https://api.studio.nebius.com/v1/
EMBED_API_KEY=your-api-key-here

# Vector Dimension (must match your indexed data)
EMBED_DIMENSION=4096

• openai - Official OpenAI API
• openai-compatible - OpenAI-compatible APIs (Nebius, Together, etc.)
• ollama - Local Ollama instance

1. Start the script:

   node search-demo.js
   # or
   python search-demo.py

2. Enter database path when prompted:

   Enter the path to the SQLite database: .codebase/vectors.db
   

3. Enter search queries:

   Enter your search query: authentication function
   

4. View results:

   Found 10 results:
   ================================================================================
   
   1. src/auth/login.ts (lines 15-45)
      Score: 87.32% | Distance: 0.1268
      ------------------------------------------------------------------------------
      export async function authenticateUser(username: string, password: string) {
        const user = await db.users.findOne({ username });
        if (!user) {
          throw new Error('User not found');
        }
   

5. Exit:

   Enter your search query: exit
   

Both scripts read embedding configuration from environment variables.

// JavaScript
const db = new Database(dbPath);
sqlite_vec.load(db);

# Python
conn = sqlite3.connect(db_path)
sqlite_vec.load(conn)

// JavaScript
const queryVector = await generateEmbedding(query);

# Python
query_vector = generate_embedding(query)

SELECT 
  file_path,
  code_chunk,
  start_line,
  end_line,
  distance
FROM code_vectors
WHERE embedding MATCH ?
ORDER BY distance
LIMIT 10

Results are sorted by similarity score (1 - distance) and displayed with:

• File path and line numbers
• Similarity score (0-100%)
• Code preview (first 5 lines)

• 90-100% - Excellent match (almost identical)
• 70-90% - Good match (semantically similar)
• 50-70% - Moderate match (related concepts)
• <50% - Weak match (loosely related)

• Lower distance = Higher similarity
• Distance is converted to score: score = 1 - distance
• For cosine distance: 0 = identical, 2 = opposite

JavaScript:

const results = searchDatabase(db, tableName, queryVector, 20); // 20 results

Python:

results = search_database(conn, table_name, query_vector, limit=20)  # 20 results

JavaScript:

const query = `
  SELECT * FROM ${tableName}
  WHERE embedding MATCH ?
    AND file_path LIKE 'src/auth/%'
  ORDER BY distance
  LIMIT ?
`;

Python:

query = f"""
  SELECT * FROM {table_name}
  WHERE embedding MATCH ?
    AND file_path LIKE 'src/auth/%'
  ORDER BY distance
  LIMIT ?
"""

// JavaScript
const results = searchResults.filter(r => r.score >= 0.7); // 70% minimum

# Python
results = [r for r in results if r['score'] >= 0.7]  # 70% minimum

• Check the path you entered
• Make sure you've indexed your codebase first:

  codesql -start .

• The database might not be a SQLite-vec database
• Try indexing your codebase first

• Check your EMBED_API_KEY is correct
• Verify EMBED_BASE_URL is accessible
• Ensure EMBED_MODEL is valid for your provider

• EMBED_DIMENSION must match the dimension used during indexing
• Check your .env file
• Common dimensions: 1536 (OpenAI small), 4096 (Qwen)

Query: user authentication login

Query: SQL query database connection

Query: try catch error handling

Query: REST API endpoint route handler

• SQLite-vec Documentation
• Main Project README
• Installation Guide

Same as the main project.