GCP Backend Setup¶

The GCP setup uses Cloud SQL (PostgreSQL + pgvector) for storage and Vertex AI for embeddings.

Prerequisites¶

A GCP project with billing enabled
gcloud CLI authenticated
Cloud SQL Admin API enabled
Vertex AI API enabled

Step 1: Create a Cloud SQL instance¶

gcloud sql instances create distill-db \
  --database-version=POSTGRES_15 \
  --tier=db-f1-micro \
  --region=us-central1

Step 2: Create the database¶

gcloud sql databases create distill --instance=distill-db

Step 3: Set up Cloud SQL Proxy¶

# Download the proxy
curl -o cloud-sql-proxy https://storage.googleapis.com/cloud-sql-connectors/cloud-sql-proxy/v2.15.2/cloud-sql-proxy.linux.amd64
chmod +x cloud-sql-proxy

# Start the proxy (runs in background)
./cloud-sql-proxy your-project:us-central1:distill-db &

Step 4: Configure environment¶

export BACKEND=postgres
export DATABASE_URL="postgresql://postgres:password@127.0.0.1:5432/distill"
export EMBEDDING_PROVIDER=vertex
export GCP_PROJECT=your-project
export GCP_LOCATION=us-central1

Step 5: Start Distill¶

python -m distill_mcp

The PostgreSQL store will automatically:

Create the memories table with pgvector extension
Set up full-text search indexes
Initialize RLS policies (if AUTH_ENABLED=true)

Cost¶

Approximately $11/month for a db-f1-micro instance. Vertex AI embedding costs are negligible at typical usage volumes.

Authentication (optional)¶

Enable git-based identity with Row-Level Security:

export AUTH_ENABLED=true

This uses your git identity (user.email) to tag memories and enforce ownership via PostgreSQL RLS policies. Anonymous users get read-only access.