🤖 Open-Agent-DB — The Universal AI Agent & MCP Hub

The Universal Open Registry and Semantic Vector Search Engine for AI Agent Skills, MCP Servers, Project Rules, and Autonomous Directives.

• GitHub Repository: https://github.com/nsozturk/open-agent-db
• Total Ecosystem Assets: 3,480,000+ indexed assets
• Ecosystems: Claude Code, Cursor, OpenAI Codex, Google Gemini, GitHub Copilot, OpenCode, and Universal Agent Standards.

---

📂 Repository Contents

This repository follows a Unified Hub Model, containing both vector embeddings and the complete offline database:

1. 🧠 Semantic Vector DB (open_agent_db_vectors.parquet)

• Dense 128-dimensional float32 vector embeddings over Agent Skills, MCP Servers, and Directives.
• Standard Parquet format, previewable directly in the Hugging Face Dataset Viewer.
• Ready for instant semantic similarity search via FAISS, cosine distance, or LanceDB.

2. 🗄️ Universal SQL Database (open_agent_db_sql.sqlite.zst)

• The full 30+ GB offline SQLite database with full-text search (FTS5), schema indices, and raw BLOB files for 3.48M+ assets.
• Highly compressed with Zstandard (~12 GB download, expands to 31 GB in ~10 seconds).
• Zero dependencies needed to query: use SQLite, DuckDB, or Python.

---

🚀 Quick Usage in Python

1. Semantic Vector Search with Hugging Face Datasets & FAISS

from datasets import load_dataset
import numpy as np

# Load pre-computed vector dataset from Hugging Face
dataset = load_dataset('ns0bj/open-agent-db', data_files='open_agent_db_vectors.parquet', split='train')

print(f"Total vector records: {len(dataset):,}")
print("Sample record:", dataset[0])

# Add native FAISS index for sub-millisecond similarity search
dataset.add_faiss_index(column="embedding")

# Example query vector (128-dimensional)
query_vector = np.zeros(128, dtype=np.float32)
scores, samples = dataset.get_nearest_examples("embedding", query_vector, k=5)

2. Downloading & Using the Complete SQL Database

# Download compressed database via huggingface-cli
hf download ns0bj/open-agent-db open_agent_db_sql.sqlite.zst --repo-type dataset --local-dir ./data

# Decompress in seconds
unzstd ./data/open_agent_db_sql.sqlite.zst -o ./data/open_agent_db.sqlite

# Query immediately with SQLite
sqlite3 ./data/open_agent_db.sqlite "SELECT count(*) FROM assets;"

import sqlite3

conn = sqlite3.connect("data/open_agent_db.sqlite")
cursor = conn.cursor()
cursor.execute("SELECT name, ecosystem, stars, github_url FROM assets WHERE stars > 1000 ORDER BY stars DESC LIMIT 10")
for row in cursor.fetchall():
    print(row)

---

📄 License

MIT License. Created and maintained by Enes Öztürk (@nsozturk).