agentnexus-tools

Agent Nexus

Automatically convert any API into AI-ready tools in 30 seconds

⚠️ Experimental: This project is in active development. Review all generated code before use in production. Contributions welcome to improve safety and reliability.

Agent Nexus eliminates the need for manual API integration code. Point it at any API URL and get production-ready Python tools that AI agents can use immediately.

What It Does

Developers spend hours writing integration code for every API their AI agents need to use. Agent Nexus automates this entirely.

Give it an API URL → Get working Python code + searchable catalog entry in under 30 seconds.

How It Works

Agent Nexus uses 4 specialized agents working together:

1. API Introspector - Discovers API endpoints automatically (OpenAPI spec or intelligent probing)
2. Tool Generator - Creates clean Python integration code with authentication
3. Catalog Search - Indexes tools with AI embeddings for semantic search
4. Tool Orchestrator - Coordinates multi-tool workflows

Tech Stack

• Python 3.11+
• Elasticsearch 8.15+ (storage, vector search, ES|QL)
• sentence-transformers (AI embeddings)
• Click (CLI)
• Docker & Docker Compose

Installation

# Install from PyPI
pip install agentnexus-tools

# Or install from source
git clone https://github.com/lcgani/agent-nexus
cd agent-nexus
pip install -e .

Quick Start

# 1. Start Elasticsearch (optional - use --skip-index for faster generation)
docker-compose up -d

# 2. Generate your first tool
agent-nexus generate https://api.github.com

# 3. Use the generated tool
python your_script.py

Usage Examples

Generate Tool from Any API

# GitHub
agent-nexus generate https://api.github.com

# Stripe
agent-nexus generate https://api.stripe.com

# Fast mode (skip Elasticsearch indexing)
agent-nexus generate https://api.github.com --skip-index

Search with Natural Language

# Find payment APIs
python -m src.cli search "payment processing credit cards"

# Find weather APIs
python -m src.cli search "weather forecast temperature"

# Find code hosting APIs
python -m src.cli search "git repositories version control"

Use Generated Tools

# Import auto-generated tool
exec(open('generated_tools/api.github.com.py').read())
github = ApiGithubCom()

# Make API calls
import requests
response = requests.get(
    f"{github.base_url}/users/octocat",
    headers=github._headers()
)
print(response.json())

Architecture

API URL Input
     ↓
Agent 1: Introspector (discovers endpoints)
     ↓
Elasticsearch (stores discovery data)
     ↓
Agent 2: Generator (creates Python code)
     ↓
Elasticsearch (stores tool + embeddings)
     ↓
Agent 3: Search (semantic search)
     ↓
Agent 4: Orchestrator (multi-tool workflows)

Project Structure

agent-nexus/
├── src/
│   ├── agents/
│   │   ├── introspector.py    # API discovery
│   │   ├── generator.py       # Code generation
│   │   ├── search.py          # Vector search
│   │   └── orchestrator.py    # Workflow coordination
│   ├── elasticsearch/
│   │   ├── client.py          # ES connection
│   │   └── schemas.py         # Index mappings
│   ├── cli.py                 # CLI interface
│   └── config.py              # Configuration
├── generated_tools/           # Auto-generated tools
├── docker-compose.yml         # Elasticsearch setup
├── requirements.txt
└── README.md

Performance Targets

• Tool generation: <30 seconds per API
• Search relevance: >90% accuracy
• Catalog size: 50+ tools tested
• End-to-end: <1 minute from URL to working tool

Commands

# Setup Elasticsearch indexes
python -m src.cli setup

# Generate tool from API
python -m src.cli generate <API_URL>

# Search tool catalog
python -m src.cli search "<query>"

License

MIT