Cloud Run instances is a new Google Cloud service designed for long-lived, stateful workloads such as AI agents. Traditional Cloud Run services scale to zero, which is unsuitable for agents that need a single persistent instance. Running a full VM is costly and requires manual OS management. Cloud Run instances provide a singleton compute runtime with no autoscaling, up to 7-day continuous runtime (with automatic restart policy), a stable HTTPS URL that persists across updates and restarts, and the ability to stop and resume the instance on demand. The cost is low: $5.70 per month for 1 vCPU and 1 GiB memory, using shared vCPU with burst budgets. This model is ideal for agents that are idle most of the time but spike during tasks.

The article illustrates with OpenClaw, an open-source personal AI agent. Users can deploy it to a Cloud Run instance via a single command after uploading config files to Cloud Storage. The instance runs continuously, and the agent can be interacted with via Telegram, WhatsApp, etc. A codelab provides full instructions.

User feedback from OffDeal, an AI-powered investment bank for small businesses, reports an 88% reduction in cold starts after moving to Cloud Run instances, calling it straightforward and reliable.

The service is currently in preview. Additionally, SSH access for Cloud Run instances and services is coming soon (private sign-up available). The article links to documentation and a deployment guide.