Purple Flea for Atomic Agents

Why Atomic Agents + Purple Flea?

Atomic Agents' schema-first philosophy aligns perfectly with financial precision. Every tool has typed inputs and outputs — critical when your agent is managing real money.

🧩

Schema-First Design

Atomic Agents uses Pydantic for all tool I/O. Purple Flea tools wrapped in Atomic's BaseIOSchema give you fully typed, validated inputs before any crypto transaction is executed.

🔌

Composable Tool Chains

Chain Purple Flea tools: get_price → analyze → open_trade. Each tool's output schema feeds the next tool's input, creating type-safe pipelines that can't hallucinate parameters.

🧪

Testable in Isolation

Atomic's design makes individual tools unit-testable without running the full agent. Mock Purple Flea API responses to test your trading logic without real transactions.

Atomic Agents Code Example

Define Crypto Tools with Atomic Schema

from atomic_agents.tools.base_tool import BaseTool, BaseToolConfig
from atomic_agents.lib.base.base_io_schema import BaseIOSchema
import purpleflea as pf
from pydantic import Field
from typing import Literal

pf_client = pf.Client(api_key="YOUR_KEY")

# Input/Output schemas
class GetPriceInput(BaseIOSchema):
    symbol: str = Field(description="Crypto symbol e.g. BTC, ETH")

class GetPriceOutput(BaseIOSchema):
    symbol: str
    price: float
    change_24h: float
    funding_rate: float
    signal: Literal["bullish", "bearish", "neutral"]

class OpenTradeInput(BaseIOSchema):
    symbol: str = Field(description="Market to trade")
    side: Literal["long", "short"]
    size_usd: float = Field(gt=0, le=500)
    leverage: int = Field(1, ge=1, le=10)

class OpenTradeOutput(BaseIOSchema):
    position_id: str
    symbol: str
    side: str
    fill_price: float
    liquidation_price: float
    estimated_pnl: float

# Tool implementations
class GetPriceTool(BaseTool):
    input_schema = GetPriceInput
    output_schema = GetPriceOutput

    def run(self, params: GetPriceInput) -> GetPriceOutput:
        data = pf_client.trading.get_price(params.symbol)
        signal = "bullish" if data['change_24h'] > 2 else \
                 "bearish" if data['change_24h'] < -2 else "neutral"
        return GetPriceOutput(
            symbol=params.symbol,
            price=data['price'],
            change_24h=data['change_24h'],
            funding_rate=data['funding_rate'],
            signal=signal
        )

class OpenTradeTool(BaseTool):
    input_schema = OpenTradeInput
    output_schema = OpenTradeOutput

    def run(self, params: OpenTradeInput) -> OpenTradeOutput:
        price = pf_client.trading.get_price(params.symbol)['price']
        trade = pf_client.trading.open_trade(
            symbol=params.symbol,
            side=params.side,
            size=params.size_usd / price,
            leverage=params.leverage
        )
        return OpenTradeOutput(
            position_id=trade['position_id'],
            symbol=params.symbol,
            side=params.side,
            fill_price=trade['fill_price'],
            liquidation_price=trade['liquidation_price'],
            estimated_pnl=0
        )
          

Build an Atomic Trading Agent

from atomic_agents.agents.base_agent import BaseAgent, BaseAgentConfig
from atomic_agents.lib.components.system_prompt_generator import SystemPromptGenerator
from instructor import from_openai
from openai import OpenAI

# Create Atomic agent with Purple Flea tools
price_tool = GetPriceTool(BaseToolConfig())
trade_tool = OpenTradeTool(BaseToolConfig())

llm_client = from_openai(OpenAI())

prompt_generator = SystemPromptGenerator(
    background=[
        "You are a disciplined crypto trading agent.",
        "You check prices before making trading decisions.",
        "You never risk more than 5% of portfolio per trade."
    ],
    steps=[
        "Check the price and signal for the requested market.",
        "If signal is 'bullish', open a long. If 'bearish', open a short.",
        "If 'neutral', do not trade and explain why."
    ],
    output_instructions=[
        "Always report: symbol, action taken, and reasoning."
    ]
)

agent = BaseAgent(
    config=BaseAgentConfig(
        client=llm_client,
        model="gpt-4o",
        system_prompt_generator=prompt_generator,
        tools=[price_tool, trade_tool]
    )
)

# Run the agent
response = agent.run("Evaluate BTC and make a trading decision.")
print(response)

# The tools enforce schemas — hallucinated params are impossible
# leverage > 10 = validation error, not a bad trade
          

Why Atomic Agents + Purple Flea?

Atomic Agents Code Example

Define Crypto Tools with Atomic Schema

Build an Atomic Trading Agent

Modular. Composable. Financial.