GCP Tech Influencer Agent

A serverless, autonomous agent that posts trending tech content to X (Twitter). Built with Google Cloud Run Jobs, Vertex AI (Gemini + Veo), and Firestore.

Architecture

• Orchestrator: Cloud Run Jobs (Python container)
• Brain: Vertex AI Gemini 1.5 Flash (Strategy & Scripting)
• Video: Vertex AI Veo (Video Generation)
• Database: Firestore (History & Deduplication)
• Secrets: Secret Manager (API Keys)
• Scheduler: Cloud Scheduler (7 daily triggers, AWST timezone)
• IaC: Terraform

Prerequisites

• Google Cloud Project with Billing enabled
• X (Twitter) Developer Account with v2 API access
• gcloud CLI installed and authenticated
• Terraform installed (v1.0+)
• Firestore database already created in your project

Quick Start (Terraform)

# 1. Clone and navigate to terraform directory
cd terraform

# 2. Run the setup script with your project ID
./setup.sh YOUR_PROJECT_ID

The setup script will:

1. Create APIs, Artifact Registry, and Service Account
2. Build and push the Docker container
3. Create Cloud Run Job and Cloud Scheduler triggers

Manual Deployment

If you prefer manual steps:

# Set your project
export PROJECT_ID="your-project-id"
export REGION="us-central1"

# Initialize and apply Terraform (prerequisites first)
cd terraform
terraform init
terraform apply -target=google_project_service.required_apis \
    -target=google_artifact_registry_repository.phantom_repo \
    -target=google_service_account.phantom_sa

# Build container image
cd ..
gcloud builds submit --tag ${REGION}-docker.pkg.dev/${PROJECT_ID}/phantom-influencer/phantom-influencer:latest .

# Apply remaining Terraform (Cloud Run Job, Scheduler)
cd terraform
terraform apply

Configure Twitter Secrets

After deployment, add your Twitter API credentials:

echo -n 'your-consumer-key' | gcloud secrets versions add TWITTER_CONSUMER_KEY --data-file=-
echo -n 'your-consumer-secret' | gcloud secrets versions add TWITTER_CONSUMER_SECRET --data-file=-
echo -n 'your-access-token' | gcloud secrets versions add TWITTER_ACCESS_TOKEN --data-file=-
echo -n 'your-access-token-secret' | gcloud secrets versions add TWITTER_ACCESS_TOKEN_SECRET --data-file=-
echo -n 'your-bearer-token' | gcloud secrets versions add TWITTER_BEARER_TOKEN --data-file=-  # Optional

Rebuild After Code Changes

After pushing new code to GitHub and pulling in GCP Cloud Shell:

cd ~/phantom
git pull

# Build new image and update Cloud Run job
gcloud builds submit --tag ${REGION}-docker.pkg.dev/${PROJECT_ID}/phantom-influencer/phantom-influencer:latest .
gcloud run jobs update phantom-influencer-job --image ${REGION}-docker.pkg.dev/${PROJECT_ID}/phantom-influencer/phantom-influencer:latest --region ${REGION}

Useful Commands

# Test the job (force post)
gcloud run jobs execute phantom-influencer-job --region us-central1 --update-env-vars FORCE_POST=true

# View logs
gcloud logging read 'resource.type=cloud_run_job AND resource.labels.job_name=phantom-influencer-job' --limit=50

# View scheduler status
gcloud scheduler jobs list --location=us-central1

# Check Terraform outputs
cd terraform && terraform output

Environment Variables

Variable	Description	Default
PROJECT_ID	Your GCP Project ID	Required
REGION	GCP region for resources	us-central1
BUDGET_MODE	True to disable video generation	True
TIMEZONE	Scheduler timezone	Australia/Perth
CURRENCY	Currency for budget display	AUD

Development

# Install dependencies
pip install -r requirements.txt

# Run locally (requires GCP credentials)
python main.py

License

MIT