Fetch.ai SDK Guide: Search Agent

💡️

This is currently a work in progress and may be subject to further updates and revisions.

This guide explains the implementation of a Search Agent and Query Agent using the Fetch.ai SDK. These Agents work together to handle user queries, search the Fetch.ai network for relevant Agents, and return results.

Below, we outline the Search Function and the Complete Agent Implementation in details.

Installation

Once you successfully have Python installed, you can start setting up your environment.

Create a new folder; call it fetchai-search-agent:


mkdir fetchai-search-agent && cd fetchai-search-agent

Then, install the required libraries:


poetry install uagents fetchai openai

Search Function

The Search Function is the core of the Search Agent. It interacts with the Fetch.ai network to find Agents matching a specific query and sends them a message.

Here's how it works:

Functionalities

Search the Fetch.ai Network: The fetch.ai(query) method searches for Agents that match the user's query.
Set Sender Identity: Each message is sent using a unique sender identity generated by Identity.from_seed.
Dynamic Payload: Uses OpenAI's GPT to generate payload based on the query.
Send Messages: For every matching Agent, a payload is constructed and sent using the send_message_to_agent function.


from fetchai import fetch
from fetchai.crypto import Identity
from fetchai.communication import send_message_to_agent
from uuid import uuid4
 
def search(query):
    # Search for agents matching the query
    available_ais = fetch.ai(query)
    
    # Create sender identity for communication
    sender_identity = Identity.from_seed("search_sender_identity", 0)
    
    for ai in available_ais.get('ais'):  # Iterate through discovered agents
 
        prompt = f"""
        you will take the following information: query={query}.
        You must return a results according to the query"""
  
        completion = client.chat.completions.create(
            model="gpt-4o",
            messages=[
                {"role": "user", "content": prompt}
            ]
        )
        
        payload = {
            "Response": completion.choices[0].message.content
        }
 
        other_addr = ai.get("address", "")  # Get agent's address
        print(f"Sending a message to an AI agent at address: {other_addr}")
 
        # Send the payload to the discovered agent
        send_message_to_agent(
            sender=sender_identity,
            target=other_addr,
            payload=payload,
            session=uuid4(),
        )
 
    return {"status": "Agent searched"}

Complete Agent Implementation

The implementation involves two Agents: Query Agent and Search Agent, which interact with each other as follows:

Query Agent

It sends the user's query to the Search Agent on startup.

it then waits for a response from the Search Agent.


query_agent = Agent(name="query_agent", seed="query_agent recovery phrase")

Search Agent

-Receives the query, processes it using the search() function, and sends a response back to the Query Agent.


search_agent = Agent(name="search_agent", seed="search_agent recovery phrase")

Overall script

fetchai-search-agent.py
from fetchai import fetch
from fetchai.crypto import Identity
from fetchai.communication import (
    send_message_to_agent
)
 
from openai import OpenAI
 
client = OpenAI()
 
 
from uuid import uuid4
from uagents import Agent, Bureau, Context, Model
 
class Query(Model):
    message: str
 
class Response(Model):
    status: str
 
def search(query):
    available_ais = fetch.ai(query)
    sender_identity = Identity.from_seed("whatever i want this to be, but i am searching", 0)
    print(f"[results] available ais : {available_ais} ")
    for ai in available_ais.get('ais'):
        
        prompt = f"""
        you will take the following information: query={query}.
        You must return a results according to the query"""
  
        completion = client.chat.completions.create(
            model="gpt-4o",
            messages=[
                {"role": "user", "content": prompt}
            ]
        )
        
        payload = {
            "Response": completion.choices[0].message.content
        }
 
        other_addr = ai.get("address", "")
 
        print(f"[INFO] Sending message to AI at address: {other_addr}")
 
        send_message_to_agent(
            sender=sender_identity,
            target=ai.get("address", ""),
            payload=payload,
            session=uuid4()
        )
         
        return {"status": "Agent searched"}
 
query_agent = Agent(name="query_agent", seed="query_agent recovery phrase")
search_agent = Agent(name="search_agent", seed="search_agent recovery phrase")
 
user_query = input("Enter your query: ")
 
@query_agent.on_event("startup")
async def send_message(ctx: Context):
    ctx.logger.info("[STARTUP] Query agent starting up and sending user query to search agent.")
    await ctx.send(search_agent.address, Query(message=user_query))
 
@search_agent.on_message(model=Query)
async def sigmar_message_handler(ctx: Context, sender: str, msg: Query):
    ctx.logger.info(f"[RECEIVED] Query received from {sender}. Message: '{msg.message}'")
    results = search(msg.message)
    ctx.logger.info("[PROCESSING] Searching completed. Sending response back to the query agent.")
    await ctx.send(query_agent.address, Response(status=results["status"]))
 
@query_agent.on_message(model=Response)
async def slaanesh_message_handler(ctx: Context, sender: str, msg: Response):
    ctx.logger.info(f"[RECEIVED] Response received from search agent {sender}. Status: '{msg.status}'")
 
bureau = Bureau()
bureau.add(query_agent)
bureau.add(search_agent)
 
if __name__ == "__main__":
    print("[INFO] Starting Bureau with Query Agent and Search Agent...")
    bureau.run()

Running the Agent

We run fetchai-search-agent.py with the following commands:


python fetchai-search-agent.py

Expected output

The expected output from the fetchai-search-agent should be similar to the following:


Enter your query: Buy me a pair of shoes
[INFO] Starting Bureau with Query Agent and Search Agent...
INFO:     [query_agent]: [STARTUP] Query agent starting up and sending user query to search agent.
INFO:     [search_agent]: [RECEIVED] Query received from agent1qdpstehd8x39n3jr0mas3adcy9d7rh4ss8wtw6euch0mq04tqu66kpfcu3q. Message: 'Buy me a pair of shoes'
INFO:httpx:HTTP Request: POST https://agentverse.ai/v1/search/agents "HTTP/1.1 200 OK"
[results] available ais : {
  "ais": [
    {
      "address": "agent1qw7802t7qf98kg775k7f5v3f9h864c72eja2r94pumxnvyx3492xyzu8fmg",
      "name": "My AI's Name",
      "readme": "\n<description>Used for getting the nike shoes</description>\n<use_cases>\n    <use_case>Used to get the nike shoes</use_case>\n</use_cases>\n<payload_requirements>\n<description>nike ai shoess</description>\n<payload>\n    <requirement>\n        <parameter>question</parameter>\n        <description>to gwt the nike shoese</description>\n    </requirement>\n</payload>\n</payload_requirements>\n",
      "protocols": [
        {
          "name": "",
          "version": "",
          "digest": "proto:a03398ea81d7aaaf67e72940937676eae0d019f8e1d8b5efbadfef9fd2e98bb2"
        }
      ],...
    },...
  ]
}
INFO:httpx:HTTP Request: POST https://api.openai.com/v1/chat/completions "HTTP/1.1 200 OK"
[INFO] Sending message to AI at address: agent1qv5jr3ylluy45hzwftgctt0hnaa75h2m0ehflxdw8rz3720pjmaegy2447r
{"version":1,"sender":"agent1qdtxzn2e0dg8y2v5y53p7frplt4w6wq36rfapv38g8x9ukgpc28fgfqnjug","target":"agent1qv5jr3ylluy45hzwftgctt0hnaa75h2m0ehflxdw8rz3720pjmaegy2447r","session":"924e26cb-bae2-4442-95a6-02292125c4f4","schema_digest":"model:708d789bb90924328daa69a47f7a8f3483980f16a1142c24b12972a2e4174bc6","protocol_digest":"proto:a03398ea81d7aaaf67e72940937676eae0d019f8e1d8b5efbadfef9fd2e98bb2","payload":"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","expires":null,"nonce":null,"signature":"sig1lsc7kttap4utjfuhs0w34vy396vse2tuekyw6aak77hr9sdwj967u3u3u96l9c20yzn25qgsw70pjtd4tn2hgtlmp8x7u39ad9jhurcgaru0n"}
INFO:fetchai:Got response looking up agent endpoint
https://staging-api.flockx.io/v1/chats/webhook_agent/
INFO:fetchai:Sent message to agent
INFO:     [search_agent]: [PROCESSING] Searching completed. Sending response back to the query agent.
INFO:     [bureau]: Starting server on http://0.0.0.0:8000 (Press CTRL+C to quit)
INFO:     [query_agent]: [RECEIVED] Response received from search agent agent1qgj8y2mswcc4jm275tsnq948fa7aqe8d9v0jd78h0nx9ak6v3fnxj6m6pkj. Status: 'Agent searched'

Was this page helpful?

You can also leave detailed feedbackon Github

Multi-file agent pipeline for AI Engine: Hugging face API to create a multi agent pipeline Introducing dialogues