Using bq_agent_mick Agent Engine

Your BigQuery agent is successfully deployed to Vertex AI Agent Engine! 🎉

Note: If you’ve updated the agent code (especially instructions), you’ll need to redeploy:

make deploy-agent-engine

Important: After redeployment, update the REASONING_ENGINE_ID in test_agent.py with the new ID from the deployment output, or set it via environment variable:

export REASONING_ENGINE_ID="your-new-id"

Current Status

✅ Deployed: Agent is live and running
✅ API Endpoints Found: :query and :streamQuery?alt=sse
✅ Working Format: Successfully tested with streamQuery endpoint!

Agent Engine Information

  • Agent Name: bq_agent_mick
  • Project: qwiklabs-asl-04-8e9f23e85ced
  • Location: us-central1
  • Reasoning Engine ID: 6060143054440890368
  • Console: https://console.cloud.google.com/vertex-ai/agents/agent-engines?project=qwiklabs-asl-04-8e9f23e85ced

How to Query Your Agent

Option 1: Use the Test Script (Easiest!) ✅

The test_agent.py script is now working:

# Single query
python bq_agent_mick/test_agent.py "What are the top 10 services by cost?"

# Interactive mode
python bq_agent_mick/test_agent.py

Option 2: Use Python Code Directly

from google.auth import default
from google.auth.transport.requests import Request
import requests
import json

# Authenticate
credentials, _ = default()
if not credentials.valid:
    credentials.refresh(Request())

# Configuration
PROJECT_ID = "qwiklabs-asl-04-8e9f23e85ced"
LOCATION = "us-central1"
REASONING_ENGINE_ID = "6060143054440890368"

# Query endpoint (streaming - recommended)
stream_url = f"https://{LOCATION}-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/{LOCATION}/reasoningEngines/{REASONING_ENGINE_ID}:streamQuery?alt=sse"

headers = {
    "Authorization": f"Bearer {credentials.token}",
    "Content-Type": "application/json",
}

payload = {
    "input": {
        "message": "What are the top 10 services by cost?",
        "user_id": "test_user",
        "session_id": "test_session"
    }
}

# Send request and read SSE stream
response = requests.post(stream_url, headers=headers, json=payload, stream=True, timeout=120)

if response.status_code == 200:
    for line in response.iter_lines(decode_unicode=True):
        if line and line.startswith("data: "):
            data = json.loads(line[6:])
            print(data)  # Process each streaming chunk
else:
    print(f"Error: {response.status_code} - {response.text}")

Option 3: Use cURL

# Get access token
ACCESS_TOKEN=$(gcloud auth print-access-token)

# Query the agent
curl -X POST \
  "https://us-central1-aiplatform.googleapis.com/v1/projects/qwiklabs-asl-04-8e9f23e85ced/locations/us-central1/reasoningEngines/6060143054440890368:streamQuery?alt=sse" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "message": "What are the top 10 services by cost?",
      "user_id": "test_user",
      "session_id": "test_session"
    }
  }'

API Details

Endpoints

  1. Stream Query (Recommended - Real-time streaming): 
    POST https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT}/locations/{LOCATION}/reasoningEngines/{ID}:streamQuery?alt=sse
    • Returns: Server-Sent Events (SSE) stream
    • Best for: Interactive queries with real-time responses
  2. Query (Alternative - Single response): 
    POST https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT}/locations/{LOCATION}/reasoningEngines/{ID}:query
    • Returns: Single JSON response
    • Note: Requires agent to have a default query method (currently uses stream_query)

Request Format

{
  "input": {
    "message": "Your question here",
    "user_id": "user_id_string",
    "session_id": "session_id_string"
  }
}

Response Format (StreamQuery)

Server-Sent Events (SSE) format:

data: {"content": "response text chunk 1"}
data: {"content": "response text chunk 2"}
...

Troubleshooting

Common Issues:

  1. Empty Stream Response:
    • The agent may take 30-60 seconds to process complex queries
    • Be patient and wait for the stream to start
    • Check agent logs in the Console if no response after 2 minutes
  2. Authentication Errors: 
    gcloud auth application-default login
  3. Timeout Errors:
    • Increase timeout to 120 seconds for complex queries
    • The streamQuery endpoint can take time to start returning data
  4. 400 Errors:
    • Ensure payload uses {"input": {...}} structure
    • Include message, user_id, and session_id in the input object

Alternative: Use Locally

While we figure out the Agent Engine API, you can always test the agent locally:

python main.py

This uses the same agent code, just running locally instead of in Agent Engine.

Help Us Improve

If you discover the correct API format, please share it so we can update the scripts!