GCP Agent Harness
View on GitHub ↗What it is
A cloud-native harness for AI agents demonstrating infrastructure-as-code, automated CI/CD, and serverless observability on Google Cloud Platform.
Because AI workloads can have unpredictable latency, multi-step execution flows, and variable token costs, this harness provides a production-grade foundation for hosting agent endpoints backed by managed relational persistence and centralized telemetry.
How it works
The system is built around declarative cloud infrastructure and serverless container execution:
- Cloud Run API: Containerized Node.js and TypeScript service hosting agent harness endpoints, scaling dynamically to zero or under load.
- Terraform IaC: Fully defines and provisions all required cloud infrastructure including Cloud SQL (PostgreSQL), Secret Manager, and IAM permissions.
- Cloud Build CI/CD: Automated pipeline triggered on GitHub pushes to run Prisma database migrations and deploy new container revisions seamlessly.
- Operations & Telemetry: Collects structured logging and execution metrics in Google Cloud Operations for deep real-time observability across agent invocations.
Tech Stack
- Compute: Google Cloud Run (Serverless Containers)
- Database: Google Cloud SQL (PostgreSQL) with Prisma ORM
- Infrastructure as Code: Terraform
- CI/CD: Google Cloud Build & GitHub
- Runtime: Node.js, TypeScript, Express
- Observability: Google Cloud Operations (Structured Logging & Dashboards)