Precompiled Agents Workflows Network (P.A.W.N. ♟️)

Precompiled Agents Workflows Network is a multi-package project that provides prebuilt agent workflows for blockchain and crypto agents. PAWN workflows are implemented using Langgraph Supervisor, ensuring maximum composability for agent systems. PAWN also allows seamless integration with various agent frameworks, such as GOAT.

How to Install?

You can install the base package along with only the dependencies you need. For example:

• Basic installation:

pip install pawn-ai

• Installation with a specific submodule (e.g., llamafeed_worflow):

  pip install pawn-ai[llamafeed_worflow]

• Installation with multiple submodules:

  pip install pawn-ai[goat_evm_workflow,goat_solana_workflow]

The extras allow you to install only the dependencies relevant to the submodules you intend to use.

Integration

PAWN offers several integration options for incorporating agent workflows into your system:

1. Integrating with an Existing Supervisor-Based System

If you already have a supervisor-based agent system, you can attach PAWN's precompiled workflows directly.

For example:

from pawn.llamafeed_workflow import LlamaFeedWorkflow

team1 = ...
team2 = LlamaFeedWorkflow()

custom_workflow = create_supervisor(
    agents=[team1, team2.workflow],
    model=model,
    prompt=(
        "You are a team supervisor managing three teams:"
        ...
    )
)

This approach allows you to seamlessly integrate PAWN's workflows as additional tools in your existing system. See more here.

2. Using the Workflow as a Tool (Agent as a Tool Pattern)

PAWN workflows can also be used as tools within your agent's framework. This approach follows the "agent as a tool" pattern described in the LangGraph documentation. For example using OpenAI Agents SDK:

import asyncio

from agents import Agent, Runner, function_tool
from pawn.hyperliquid_trader_workflow import HyperliquidWorkflow


@function_tool
def get_price(symbol: str) -> float:
    workflow: HyperliquidWorkflow = HyperliquidWorkflow()
    response = workflow.invoke(f'What is the price of {symbol}?')
    ...
    return result


agent = Agent(
    name="Price Fetcher",
    instructions="You are a helpful agent that can fetch price of crypto.",
    tools=[get_price],
)


async def main():
    result = await Runner.run(agent, input="What's the price of bitcoin?")
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())

This integration leverages PAWN's workflows as modular tools that can be called by your agent system when needed.

3. Running the Workflow as a Standalone FastAPI or fastMCP Server

You can also run a PAWN workflow as an independent service using FastAPI (or a similar framework like fastMCP). This allows you to deploy the workflow as a RESTful API. For example, using FastAPI:

from fastapi import FastAPI
from pawn.goat_evm_workflow import EVMAgenticWorkflow

app = FastAPI()
workflow = EVMAgenticWorkflow()

@app.post("/invoke")
def invoke_workflow(payload: dict):
    return workflow.invoke(payload)

This will launch a standalone server where you can access the workflow via HTTP.

Contributing

Contributions are welcome! Please ensure that your changes are well-documented and include examples.

License

This project is licensed under the MIT License. See the LICENSE file for details.