Installation¶

This guide will help you install AgenticPay and its dependencies.

Requirements¶

Python 3.10 or higher
Conda (recommended) or pip
GPU with CUDA support (recommended for local model inference)

Basic Installation¶

Using Conda (Recommended)¶

# Create conda environment
conda create -n agenticpay python=3.10 -y
conda activate agenticpay

# Clone the repository
git clone https://github.com/SafeRL-Lab/AgenticPay.git
cd AgenticPay

# Install dependencies
pip install -r requirements.txt

# Install package in editable mode
pip install -e .

Using pip¶

# Create virtual environment
python -m venv agenticpay-env
source agenticpay-env/bin/activate  # On Windows: agenticpay-env\Scripts\activate

# Clone and install
git clone https://github.com/SafeRL-Lab/AgenticPay.git
cd AgenticPay
pip install -r requirements.txt
pip install -e .

Model Setup¶

Local Models¶

For local model inference, download models from Hugging Face and save them to the agenticpay/models/download_models directory.

# Example: Download Qwen3-8B-Instruct
mkdir -p agenticpay/models/download_models
# Use huggingface-cli or git lfs to download the model

Supported local inference backends:

SGLang: High-performance serving framework
vLLM: Fast LLM inference with PagedAttention

OpenAI API¶

To use OpenAI models, set your API key:

export OPENAI_API_KEY="your-api-key-here"

Or in Python:

import os
os.environ["OPENAI_API_KEY"] = "your-api-key-here"

Verification¶

Verify your installation by running:

import agenticpay
from agenticpay import make
from agenticpay.agents.buyer_agent import BuyerAgent
from agenticpay.agents.seller_agent import SellerAgent

print("AgenticPay installed successfully!")

Dependencies¶

Core dependencies include:

torch - PyTorch for model inference
transformers - Hugging Face transformers
gymnasium - OpenAI Gymnasium for environment API
openai - OpenAI API client (optional)
sglang - SGLang inference backend (optional)
vllm - vLLM inference backend (optional)

For a complete list, see requirements.txt in the repository.