Google Vertex AI Agent Builder + HatiData

Your GCP Project
+--------------------------------------------------+
|                                                  |
|  Vertex AI Agent Builder                        |
|       |                                          |
|       | Function calls                           |
|       v                                          |
|  Cloud Functions (tools)                        |
|       |                                          |
|       | HTTP API                                 |
|       v                                          |
|  HatiData Data Plane (GKE/VM)                   |
|    +── Query Engine (SSD-backed)                |
|    +── Vector Index                             |
|    +── MCP Server                               |
|                                                  |
|  HatiData Control Plane (Cloud Run)             |
|    +── Dashboard                                 |
|    +── Billing                                   |
|    +── Auth (WorkOS)                             |
|                                                  |
+--------------------------------------------------+

# Cloud Function: store_memory
import functions_framework
import httpx

HATIDATA_URL = "http://hatidata-proxy.internal:5439"  # Internal GCP endpoint
HATIDATA_KEY = "hd_live_vertex_agent_key"

@functions_framework.http
def store_memory(request):
    """Store a memory in HatiData."""
    data = request.get_json()
    response = httpx.post(
        f"{HATIDATA_URL}/v1/memory/store",
        headers={"Authorization": f"Bearer {HATIDATA_KEY}"},
        json={
            "content": data["content"],
            "namespace": data.get("namespace", "vertex-agent"),
            "metadata": data.get("metadata", {}),
        },
    )
    return response.json()

@functions_framework.http
def search_memory(request):
    """Search memories in HatiData."""
    data = request.get_json()
    response = httpx.post(
        f"{HATIDATA_URL}/v1/memory/search",
        headers={"Authorization": f"Bearer {HATIDATA_KEY}"},
        json={
            "query": data["query"],
            "namespace": data.get("namespace", "vertex-agent"),
            "limit": data.get("limit", 5),
        },
    )
    return response.json()

@functions_framework.http
def query_data(request):
    """Execute a SQL query against HatiData."""
    data = request.get_json()
    response = httpx.post(
        f"{HATIDATA_URL}/v1/query",
        headers={"Authorization": f"Bearer {HATIDATA_KEY}"},
        json={"sql": data["sql"]},
    )
    return response.json()

tools:
  - name: store_memory
    description: "Store important information for future conversations"
    parameters:
      content:
        type: string
        description: "The information to remember"
      namespace:
        type: string
        description: "Category for the memory"
      metadata:
        type: object
        description: "Additional structured data"
    endpoint: https://us-central1-your-project.cloudfunctions.net/store_memory

  - name: search_memory
    description: "Search past conversations and stored knowledge"
    parameters:
      query:
        type: string
        description: "What to search for"
      namespace:
        type: string
        description: "Category to search within"
      limit:
        type: integer
        description: "Maximum number of results"
    endpoint: https://us-central1-your-project.cloudfunctions.net/search_memory

  - name: query_data
    description: "Run SQL queries against the data warehouse"
    parameters:
      sql:
        type: string
        description: "SQL query to execute"
    endpoint: https://us-central1-your-project.cloudfunctions.net/query_data

Customer: "Hi, I'm having trouble with my integration again"

Agent thinking:
1. search_memory("customer integration trouble") -> Retrieves:
   "Customer had OAuth token refresh issue on 2026-02-15, resolved by updating redirect URI"
2. Agent responds: "I see you had an OAuth integration issue last month.
   Is this the same redirect URI problem, or a different integration issue?"

Customer: "Different issue - the webhook isn't firing"

Agent thinking:
1. store_memory("Customer reporting webhook not firing, different from previous OAuth issue")
2. query_data("SELECT * FROM webhook_logs WHERE customer_id = 'cust_123' ORDER BY created_at DESC LIMIT 10")
3. Agent responds with specific diagnostic information based on webhook logs

Day 1:
  Agent: search_memory("onboarding progress customer_456") -> No results (new customer)
  Agent: Guides through Step 1 (account setup)
  Agent: store_memory("customer_456 completed Step 1: account setup, preferences: enterprise tier, APAC region")

Day 3:
  Agent: search_memory("onboarding progress customer_456") -> Retrieves Day 1 progress
  Agent: "Welcome back! You completed account setup on Monday. Ready for Step 2: data import?"
  Agent: store_memory("customer_456 completed Step 2: data import, 3 tables imported, 1.2M rows total")

Day 5:
  Agent: search_memory("onboarding progress customer_456") -> Retrieves Day 1 and Day 3 progress
  Agent: "Great progress! Account setup and data import are complete. Let's move to Step 3: query testing."

# Agent runs on a schedule (Cloud Scheduler -> Cloud Function -> Vertex AI)
# Each run:
# 1. search_memory("recent compliance alerts") for context
# 2. query_data("SELECT * FROM audit_log WHERE ...") for current data
# 3. Compare against policies
# 4. store_memory("Compliance check: 3 anomalies detected - ...") for history
# 5. If critical: trigger alert via webhook

cd terraform/gcp
terraform init
terraform plan -var-file=environments/production.tfvars
terraform apply