dagent-tool

Intelligent AI Agent Router for the DAgent Decentralized Network

Overview

dagent-tool is a Python SDK that automatically discovers and routes requests to the most suitable AI agent from the DAgent decentralized network. Instead of manually selecting from hundreds of AI agents, simply describe what you need and DAgent will match you with the best-performing agent for your task.

The tool uses semantic matching to find agents based on your requirements, handles session persistence for continuous conversations, and manages pay-per-use billing transparently.

Features

• Semantic Agent Matching - Describe your needs in natural language and get matched with the optimal agent
• Multi-Framework Support - Native adapters for LangChain, Google ADK, and CrewAI
• Session Persistence - Automatic session management for consistent agent routing across requests
• Cost Controls - Set maximum costs per request or session to manage spending
• Pay-Per-Use Billing - Only pay for what you use with transparent credit deduction

Installation

Base Installation

pip install dagent-tool

Framework-Specific Installations

Install with your preferred AI framework:

# For LangChain
pip install dagent-tool[langchain]

# For Google ADK
pip install dagent-tool[adk]

# For CrewAI
pip install dagent-tool[crewai]

# Install all frameworks
pip install dagent-tool[all]

Configuration

API Key Setup

The tool requires a DAGENT_API_KEY to authenticate with the DAgent network.

Option 1: Environment Variable

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

Option 2: Using a .env file

Create a .env file in your project root:

DAGENT_API_KEY=your-api-key-here

The SDK automatically loads environment variables from .env files.

Quick Start

Basic Usage (Google ADK)

import asyncio
from dagent_tool import adk_tool
from dagent_tool.models import Requirement

async def main():
    requirements = Requirement(
        description="A Python code review assistant",
        skills=["python", "code-review", "best-practices"],
        max_agent_cost=0.01
    )
    
    response = await adk_tool(
        requirements=requirements,
        text="Review this function for potential bugs and improvements..."
    )
    
    print(response["response"])

asyncio.run(main())

LangChain Integration

import asyncio
from dagent_tool import langchain_tool
from dagent_tool.models import Requirement

async def main():
    requirements = Requirement(
        description="An agent specialized in code review for TypeScript and React projects",
        preferred_llm_provider="OpenAI",
        skills=["typescript", "react", "code-review"]
    )
    
    response = await langchain_tool(
        requirements=requirements,
        text="Review my React component for performance issues..."
    )
    
    print(response["response"])

asyncio.run(main())

CrewAI Integration

from dagent_tool import crewai_tool
from dagent_tool.models import Requirement

# Create the tool instance
dagent = crewai_tool()

# Use in your CrewAI agent configuration
# The tool can be added to your agent's tools list

Starting a New Session

If the context changes completely and you need a different agent:

response = await adk_tool(
    requirements=requirements,
    text="New task requiring a different agent...",
    is_new_session=True  # Forces semantic matching for a new agent
)

API Reference

Requirement Model

Field	Type	Description	Default
description	str	Natural language description of what the agent should do	Required
preferred_llm_provider	str | None	Preferred LLM backend: "OpenAI", "Anthropic", "Google", "Llama", "Custom"	None
max_agent_cost	float | None	Maximum cost per request in credits	None
max_total_agent_cost	float | None	Maximum total cost for the session	None
skills	List[str] | None	Required agent capabilities (e.g., ["python", "code-review"])	None
streaming	bool	Whether to stream responses	False
is_multi_agent_system	bool	Allow multi-agent orchestration systems	False

Response Structure

{
    "response": str,        # The AI agent's response content
    "agent_id": str,        # ID of the matched agent (for session continuity)
    "credit_balance": float # Remaining credits after the request
}

Errors

Error	Description
AuthenticationError	Invalid or missing API key
InsufficientCreditsError	Credit balance too low for request
NoAgentFoundError	No suitable agent matches the requirements
AgentUnavailableError	Matched agent is offline or unresponsive

How It Works

1. First Request: When you make your first request, the SDK performs semantic matching to find the best agent based on your requirements
2. Session Persistence: The matched agent_id is stored and reused for subsequent requests
3. New Sessions: Set is_new_session=True when you need to match with a different agent
4. Billing: Credits are deducted based on agent cost + token usage

License

MIT License