Purple Flea for Portkey

Why Agents Need an AI Gateway

An autonomous trading or gambling agent making unobserved decisions is a liability. Portkey solves this.

👁️

Full Observability

See every LLM call your agent makes before executing a trade, placing a bet, or sending crypto. Log inputs, outputs, latencies, and costs in a central dashboard.

🛡️

Financial Guardrails

Define rules that prevent dangerous actions: "never trade with more than 10% of balance", "no leveraged positions after 3 losses", "flag any single bet over 0.1 ETH".

🔄

LLM Fallbacks

If your primary LLM (GPT-4o) is down or slow, Portkey automatically falls back to Claude or Gemini. Your trading agent stays operational 24/7.

⚡

Semantic Caching

Cache similar market analysis queries. "What's the BTC trend?" doesn't need a fresh LLM call every 30 seconds — Portkey reuses recent responses when appropriate.

📊

Cost Tracking

Track LLM costs alongside Purple Flea trading fees. Understand the true cost of each agent decision: LLM tokens + transaction fees combined.

🔐

Virtual Keys

Issue separate virtual API keys for each agent instance. Revoke a compromised agent's keys without affecting other agents or your master Purple Flea credentials.

Setup: Portkey + Purple Flea

Add an AI gateway to your crypto agent in 10 minutes.

Install & Configure

pip install portkey-ai openai purpleflea

from portkey_ai import Portkey

# Portkey as a drop-in for OpenAI
client = Portkey(
    api_key="YOUR_PORTKEY_KEY",
    virtual_key="OPENAI_VIRTUAL_KEY"
)

# Purple Flea tools (function calling format)
import purpleflea as pf

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

Add Guardrails for Trading

from portkey_ai import Portkey, Config, Guardrails

# Define financial safety guardrails
guardrails = Guardrails(
    on_request=[
        {
            "id": "no-large-bets",
            "type": "regex",
            "config": {
                "pattern": r'"size":\s*([5-9]\d+|\d{3,})',
                "action": "block",
                "message": "Position size too large, capped at 4"
            }
        }
    ]
)

config = Config(
    provider="openai",
    api_key="...",
    guardrails=guardrails,
    cache={"mode": "semantic"}
)

client = Portkey(config=config)
          

Observable Trading Agent

import json, purpleflea as pf

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

trading_tools = [
    {"type": "function", "function": {
        "name": "open_trade",
        "description": "Open perpetual futures position",
        "parameters": {
            "type": "object",
            "properties": {
                "symbol": {"type": "string"},
                "side": {"type": "string", "enum": ["long", "short"]},
                "size": {"type": "number"},
                "leverage": {"type": "integer"}
            }, "required": ["symbol", "side", "size"]
        }
    }}
]

# Every call logged by Portkey, guardrails enforced
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Trade BTC"}],
    tools=trading_tools,
    metadata={"agent_id": "trader-001", "session": "morning-run"}
)

# Handle tool calls with Purple Flea
for tc in response.choices[0].message.tool_calls or []:
    args = json.loads(tc.function.arguments)
    result = pf_client.trading.open_trade(**args)
    print(f"Trade opened: {result['position_id']}")
          

Multi-Provider Fallback