Welcome to AgenticPay’s Documentation!¶

AgenticPay is a multi-agent LLM negotiation framework for simulating buyer-seller transactions. It provides a Gymnasium-like API for building and testing intelligent negotiation agents powered by Large Language Models.

Framework Overview¶

*Figure 2: Scenario and Task Categories*¶

Key Features¶

🤖 LLM-based Agents: Buyer and Seller agents powered by LLMs (OpenAI, vLLM, SGLang)
💬 Multi-turn Conversations: Extended negotiation dialogues with context preservation
🧠 Memory System: Conversation history management for context-aware negotiations
📊 State Tracking: Comprehensive tracking of prices, rounds, and negotiation status
🎯 Flexible Configuration: Customizable negotiation parameters and agent behaviors
🔌 Extensible Design: Easy to add new agent types or LLM providers
🏪 Environment Registration: Gymnasium-like environment registration system
🛍️ Multi-Product Negotiations: Support for multiple products with context preservation
👥 Multi-Agent Scenarios: Multiple buyers, sellers, and products combinations
🔄 Parallel & Sequential Modes: Both parallel and sequential negotiation support

Quick Example¶

from agenticpay import make
from agenticpay.agents.buyer_agent import BuyerAgent
from agenticpay.agents.seller_agent import SellerAgent
from agenticpay.models.sglang_lm import SGLangLM

# Initialize model
model = SGLangLM(model_path="path/to/your/model")

# Create agents
buyer = BuyerAgent(model=model, buyer_max_price=120.0)
seller = SellerAgent(model=model, seller_min_price=80.0)

# Create environment
env = make(
    "Task1_basic_price_negotiation-v0",
    buyer_agent=buyer,
    seller_agent=seller,
    max_rounds=20,
    initial_seller_price=150.0,
)

# Run negotiation
observation, info = env.reset(
    user_requirement="I need a winter jacket",
    product_info={"name": "Premium Winter Jacket", "price": 180.0}
)

done = False
while not done:
    buyer_action = buyer.respond(observation["conversation_history"], observation)
    seller_action = seller.respond(observation["conversation_history"], observation)
    observation, reward, terminated, truncated, info = env.step(buyer_action, seller_action)
    done = terminated or truncated

Note

This project is under active development.

Contents¶

Getting Started

Examples

Examples