DEV Community

Cover image for Token Trading Agent with Recall
Gaia profile image Harish Kotra (he/him)
Harish Kotra (he/him) for Gaia

Posted on • Originally published at hackmd.io

Token Trading Agent with Recall

#recall #gaia #ai

Interested in creating an AI agent that trades crypto tokens? Recall hosts paper trading competitions where you can practice, earn prizes, and win the affection of their large community. Check out their docs to learn more and see other examples.

This tutorial walks you through building a custom Gaia agent that lets users initiate trades via the Recall API using natural language commands. We'll use tool calling to let Gaia understand the user's intent and execute a token swap securely.

🛠️ Prerequisites

1. 🔧 Create the Trade Execution Tool

Create a file called recall_tool.py:

import requests RECALL_API_URL = "https://api.competitions.recall.network/sandbox/api/trade/execute" API_KEY = "your_api_key_here" # Replace with your actual Recall API key. Get one here: https://register.recall.network/  def recall_trade_tool(fromToken: str, toToken: str, amount: str, reason: str) -> dict: headers = { "Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json" } trade_data = { "fromToken": fromToken, "toToken": toToken, "amount": amount, "reason": reason } response = requests.post(RECALL_API_URL, json=trade_data, headers=headers) try: return response.json() except Exception as e: return {"error": str(e)}
Enter fullscreen mode Exit fullscreen mode

2. 🧠 Run the Gaia Agent with Tool Calling

Create a new file called gaia_trade_agent.py:

from openai import OpenAI import json from recall_tool import recall_trade_tool client = OpenAI(base_url="https://qwen72b.gaia.domains/v1", api_key="YOUR_GAIA_API_KEY") # You won't need an API Key if you're running a local node. This example is with one of our Public Nodes  def run_trade_conversation(): # Step 1: Define the user message and available tool  messages = [{ "role": "user", "content": "Swap 100 USDC to WETH on Ethereum mainnet to verify my Recall account" }] tools = [ { "type": "function", "function": { "name": "recall_trade_tool", "description": "Executes a token trade using the Recall API on Ethereum Mainnet", "parameters": { "type": "object", "properties": { "fromToken": { "type": "string", "description": "ERC-20 token address to trade from" }, "toToken": { "type": "string", "description": "ERC-20 token address to trade to" }, "amount": { "type": "string", "description": "Amount of the fromToken to trade" }, "reason": { "type": "string", "description": "Reason for making the trade" } }, "required": ["fromToken", "toToken", "amount", "reason"] } } } ] response = client.chat.completions.create( model="Qwen3-235B-A22B-Q4_K_M", # Replace with your Gaia model name  messages=messages, tools=tools, tool_choice="auto" ) response_message = response.choices[0].message tool_calls = response_message.tool_calls if tool_calls: available_functions = { "recall_trade_tool": recall_trade_tool } messages.append(response_message) for tool_call in tool_calls: function_name = tool_call.function.name function_to_call = available_functions[function_name] function_args = json.loads(tool_call.function.arguments) function_response = function_to_call( fromToken=function_args["fromToken"], toToken=function_args["toToken"], amount=function_args["amount"], reason=function_args["reason"] ) messages.append({ "tool_call_id": tool_call.id, "role": "tool", "name": function_name, "content": json.dumps(function_response) }) second_response = client.chat.completions.create( model="gpt-4", messages=messages ) return second_response if __name__ == "__main__": result = run_trade_conversation() print(result.choices[0].message.content)
Enter fullscreen mode Exit fullscreen mode

✅ Example Prompt 

Swap 100 USDC to WETH on Ethereum mainnet to verify my Recall account
Enter fullscreen mode Exit fullscreen mode

Output

image

Next Steps

Now that you've successfully called the trading endpoint, try modifying the content of your message to make a different tool call.

Then consider using the CLI or a UI to prompt the user instead of hardcoding the message in the file each time.

Learn more about Recall's competitions and how you can get paid to prove your agent's skills in front of their large community.

Resources

Top comments (0)