GCP Agent

A multi-agent travel assistant built with Google ADK and Gemini 2.5 Flash. It searches real flights via Google Flights, resolves airport IATA codes, and generates trip itineraries — all through a conversational interface.

Architecture

root_agent
└── planning_agent
    ├── FlightPlannerandApprovalAgent (SequentialAgent)
    │   ├── flight_planner_agent     — searches flights + resolves IATA codes
    │   └── request_flight_approval_agent — presents options & captures user selection
    └── itenary_agent                — builds itinerary using Google Search

Sessions are persisted in PostgreSQL. Traces are collected by Arize Phoenix.

Stack

• Google ADK — agent framework
• fast-flights — Google Flights scraper
• FastAPI + Uvicorn — HTTP server
• PostgreSQL — session persistence
• Arize Phoenix — observability / tracing
• Docker Compose — local orchestration

Setup

1. Copy env file

cp .env.example .env

Set GOOGLE_API_KEY (or configure Vertex AI). Adjust other values as needed.

2. Run with Docker

docker compose up --build

Service	URL
Agent API	http://localhost:8000
Phoenix UI	http://localhost:6006

3. Run locally (without Docker)

# requires uv
uv sync
uv run main.py

You'll need a running PostgreSQL instance and to set SESSION_SERVICE_URI accordingly.

Deploying to Cloud Run

Prerequisites

The scripts scripts/build-registry.py and scripts/deploy.py interact with several GCP services. The person or service account running them needs the following permissions.

Required IAM roles (deploying user)

Role	Purpose
roles/artifactregistry.admin	Create the Artifact Registry repo and push Docker images
roles/run.admin	Deploy and update Cloud Run services
roles/iam.serviceAccountUser	Act as the Cloud Run service account during deployment
roles/cloudsql.client	(if verifying connectivity) Connect to Cloud SQL

Grant them via the GCP Console under IAM & Admin → IAM, or with gcloud:

PROJECT=your-project-id
MEMBER=user:you@example.com   # or serviceAccount:sa@project.iam.gserviceaccount.com

gcloud projects add-iam-policy-binding $PROJECT --member=$MEMBER --role=roles/artifactregistry.admin
gcloud projects add-iam-policy-binding $PROJECT --member=$MEMBER --role=roles/run.admin
gcloud projects add-iam-policy-binding $PROJECT --member=$MEMBER --role=roles/iam.serviceAccountUser

Required IAM roles (Cloud Run service account)

The identity that the Cloud Run service runs as (by default <project-number>-compute@developer.gserviceaccount.com) needs:

Role	Purpose
roles/cloudsql.client	Open the Cloud SQL Auth Proxy socket to agentdb

SA=<project-number>-compute@developer.gserviceaccount.com

gcloud projects add-iam-policy-binding $PROJECT --member="serviceAccount:$SA" --role=roles/cloudsql.client

Required GCP APIs

The following APIs must be enabled in your project:

gcloud services enable \
  artifactregistry.googleapis.com \
  run.googleapis.com \
  sqladmin.googleapis.com \
  vpcaccess.googleapis.com

Cloud SQL — private IP note

agentdb uses a private IP only (no public IP). The deploy script configures Cloud Run with Direct VPC Egress (--network=default --subnet=default --vpc-egress=private-ranges-only) so the service can reach the private IP. If you recreate the Cloud SQL instance, make sure it is attached to the same VPC (default) and that private IP is enabled.

Run the deploy

# Full build + deploy
python scripts/deploy.py

# Deploy only (skip Docker build)
python scripts/deploy.py --skip-build

Environment Variables

Variable	Description
GOOGLE_API_KEY	Gemini API key
GOOGLE_GENAI_USE_VERTEXAI	1 to use Vertex AI instead of API key
SESSION_SERVICE_URI	PostgreSQL URI for session storage
PHOENIX_COLLECTOR_ENDPOINT	OTLP endpoint for Phoenix tracing
AIRPORTS_SQLITE_PATH	Path to the bundled airports.sqlite
PORT	Server port (default 8000)
WEB_ALLOWED	Enable ADK web UI