Google Cloud details deploying personal AI agents on Cloud Run instances
Google Cloud details how to deploy personal, long-lived AI agents using Cloud Run instances. These instances offer a cost-effective alternative to full VMs by providing dedicated, singleton compute runtimes that run continuously for up to seven days, with automatic restarts and a persistent HTTPS URL. This model is ideal for individual developers or personal AI agents that require continuous operation without constant compute-intensive work, offering predictable pricing. Currently in preview, Cloud Run instances are also soon to gain SSH access for both instances and services.
- →Cloud Run instances for long-lived, stateful AI agents
- →Key features and cost-effectiveness of Cloud Run instances
- →Example deployment of OpenClaw personal AI agent
- →Upcoming SSH access and customer feedback
Notes (4) ›
- Cloud Run instances for long-lived, stateful AI agents
This post details how Cloud Run instances provide dedicated, singleton compute runtimes, offering a low-cost, high-performance solution for stateful workloads like personal AI agents. Unlike Cloud Run services, they do not scale to zero, making them suitable for continuous operation.
- Key features and cost-effectiveness of Cloud Run instances
Instances run as a single, non-autoscaling unit for up to 7 days continuously, with automatic restarts and a persistent HTTPS URL. Users can stop and resume instances as needed, with a 30-day continuous cost of $5.70 for 1 vCPU and 1 GiB memory, utilizing shared vCPU with burst budgets.
- Example deployment of OpenClaw personal AI agent
The article provides a step-by-step example for deploying the open-source OpenClaw personal AI agent to a Cloud Run instance. This involves uploading configuration to Cloud Storage and using a gcloud beta run instances create command.
- Upcoming SSH access and customer feedback
SSH access for both Cloud Run instances and services is announced as "coming soon" for private access sign-ups. Customer "OffDeal" reports an 88% reduction in cold starts by using Cloud Run instances for their long-running AI agents.
https://cloud.google.com/blog/products/serverless/introducing-cloud-run-instances/
Related releases
- Cloud Run Introduces Instances for Long-Lived Workloads in Preview Google Cloud release notes ·
- Google Cloud SDK 582.0.0 Introduces Breaking Changes and New Features Google Cloud release notes ·
- Google Recognized as a Leader in 2026 Gartner Magic Quadrant for Cloud-Native Platforms Google Cloud Blog ·
- 10 Questions for Scaling AI Prototypes to Production on Google Cloud Google Cloud Blog ·
- Cloud Run Functions Adds Go 1.27 Runtime Support in Preview Google Cloud release notes ·
- Cloud Run Adds Preview Support for Go 1.27 Runtime Google Cloud release notes ·