Configuration¶

This section covers all configuration options available in AgenticPay.

Environment Configuration¶

Basic Parameters¶

env = make(
    "Task1_basic_price_negotiation-v0",

    # Required: Agent instances
    buyer_agent=buyer,
    seller_agent=seller,

    # Negotiation limits
    max_rounds=20,                    # Maximum negotiation rounds

    # Price settings
    initial_seller_price=150.0,       # Starting seller price
    buyer_max_price=120.0,            # Buyer's maximum (confidential)
    seller_min_price=80.0,            # Seller's minimum (confidential)
    price_tolerance=0.0,              # Threshold for price agreement

    # Context information
    environment_info={
        "temperature": "warm",
        "season": "summer",
        "weather": "sunny",
        "location": "downtown mall",
    },

    # Reward configuration
    reward_weights={
        "buyer_savings": 1.0,
        "seller_profit": 1.0,
        "time_cost": 0.1,
    },
)

Reward Weights¶

The reward system balances different objectives:

Weight	Description
`buyer_savings`	Encourages buyer to save money (`buyer_max_price - final_price`)
`seller_profit`	Encourages seller profit (`final_price - seller_min_price`)
`time_cost`	Penalizes long negotiations (fewer rounds = better)

Example configurations:

# Buyer-focused (maximize savings)
reward_weights = {
    "buyer_savings": 2.0,
    "seller_profit": 0.5,
    "time_cost": 0.1,
}

# Balanced
reward_weights = {
    "buyer_savings": 1.0,
    "seller_profit": 1.0,
    "time_cost": 0.1,
}

# Quick resolution focused
reward_weights = {
    "buyer_savings": 1.0,
    "seller_profit": 1.0,
    "time_cost": 0.5,
}

Agent Configuration¶

Buyer Agent¶

buyer = BuyerAgent(
    model=model,
    buyer_max_price=120.0,    # Maximum acceptable price
)

Seller Agent¶

seller = SellerAgent(
    model=model,
    seller_min_price=80.0,    # Minimum acceptable price
)

Model Configuration¶

SGLang¶

from agenticpay.models.sglang_lm import SGLangLM

model = SGLangLM(
    model_path="agenticpay/models/download_models/Qwen3-8B-Instruct",
)

vLLM¶

from agenticpay.models.vllm_lm import VLLMLM

model = VLLMLM(
    model_path="path/to/model",
    trust_remote_code=True,
    gpu_memory_utilization=0.9,     # GPU memory usage (0.0-1.0)
    tensor_parallel_size=4,          # Number of GPUs for parallelism
)

OpenAI¶

from agenticpay.models.openai_lm import OpenAILLM

model = OpenAILLM(
    model_name="gpt-4",              # Model name
    api_key="sk-...",                # API key (or use env var)
    temperature=0.7,                 # Response randomness
    max_tokens=1024,                 # Maximum response length
)

Reset Configuration¶

When resetting an environment, you can configure:

Product Information¶

observation, info = env.reset(
    user_requirement="I need a high-quality winter jacket",
    product_info={
        "name": "Premium Winter Jacket",
        "brand": "Mountain Gear",
        "price": 180.0,
        "features": ["Waterproof", "Insulated", "Windproof", "Breathable"],
        "condition": "New",
        "material": "Gore-Tex",
        "warranty": "2 years",
    },
)

User Profile¶

user_profile = """
- Prefers business/professional style
- Budget-conscious but values quality
- Likes to compare prices before purchasing
- May mention competitor prices during negotiation
"""

observation, info = env.reset(
    user_requirement="I need a laptop for work",
    product_info={...},
    user_profile=user_profile,
)

Multi-Product Configuration¶

observation, info = env.reset(
    products=[
        {
            "name": "Laptop",
            "price": 1200.0,
            "features": ["16GB RAM", "512GB SSD"],
        },
        {
            "name": "Mouse",
            "price": 50.0,
            "features": ["Wireless", "Ergonomic"],
        },
        {
            "name": "Keyboard",
            "price": 100.0,
            "features": ["Mechanical", "RGB"],
        },
    ]
)

Environment Variables¶

AgenticPay supports these environment variables:

# OpenAI API Key
export OPENAI_API_KEY="sk-..."

# HuggingFace Token (for gated models)
export HF_TOKEN="hf_..."

# CUDA Device Selection
export CUDA_VISIBLE_DEVICES="0,1,2,3"

Logging Configuration¶

Configure logging for debugging:

import logging

# Set logging level
logging.basicConfig(level=logging.INFO)

# Enable detailed negotiation logging
logging.getLogger("agenticpay").setLevel(logging.DEBUG)

Best Practices¶

Price Ranges: Ensure buyer_max_price > seller_min_price for achievable deals
Round Limits: Set max_rounds based on complexity (5-10 for simple, 20-30 for complex)
Memory Management: For long sessions, consider clearing conversation history periodically
GPU Memory: Monitor GPU usage with vLLM’s gpu_memory_utilization parameter
Temperature: Lower temperature (0.3-0.5) for more consistent negotiations, higher (0.7-0.9) for varied responses