Google has introduced Cloud Run instances, a new product now in preview.
The offering gives developers a single dedicated Cloud Run runtime for workloads that need to stay active for extended periods, rather than scale up and down in response to web traffic.
Cloud Run instances are aimed at long-lived, stateful applications such as AI agents that often run continuously for one user at a time. That puts them outside the normal Cloud Run model, which is designed for stateless applications and can scale to zero when requests stop.
Under this model, each instance runs as a singleton with no autoscaling. Runtimes can stay active for up to seven days at a time and restart automatically under a default policy.
Each instance also gets an HTTPS address that remains the same across updates and restarts. Users can stop an instance when it is not needed and resume it later.
Google positioned pricing well below the cost of a dedicated virtual machine for some use cases. A Cloud Run instance with 1 vCPU and 1 GiB of memory running continuously for 30 days would cost $5.70, according to the company.
The service uses shared vCPU with burst budgets, a model Google described as suitable for workloads that remain online for long periods but use significant compute resources only intermittently. That profile is common among AI agents that sit idle and then perform a task when prompted.
Target workloads
The launch reflects a broader shift in developer demand as software agents and other persistent applications move from experimental tools on personal machines to hosted services that need stable uptime without the overhead of managing a full server.
Google contrasted the new product with a dedicated virtual machine, which typically requires continuous payment for compute resources along with management of operating system updates, firewall configuration and HTTPS endpoints.
One example Google highlighted was OpenClaw, an open-source personal AI agent. Many users start by running such tools on laptops before looking for a hosted environment that does not stop when the local device goes to sleep, the company said.
Google also pointed to Hermes and other personal-use agents as examples of software that may run continuously while serving a single user. In those cases, the economics and operating model differ from conventional cloud-native web services built around high request volumes and elastic scaling.
User feedback
Google cited OffDeal as an early user of the service for long-running agents.
"We're currently using Cloud Run instances as our primary infrastructure for our long-running agent. It reduced cold starts by 88%. Everything was very straightforward to implement, and it has been very reliable," said Luis Ruiz Morel, Member of Technical Staff at OffDeal.
That customer account highlights one of the key technical issues Google is trying to address. Cold starts, which occur when an application has to initialise after being spun up on demand, can be a problem for software expected to remain continuously available.
By offering a fixed single instance rather than an autoscaled service, Google is creating a middle ground between fully managed serverless infrastructure and conventional virtual machines. The approach keeps Cloud Run's managed deployment model while changing runtime behaviour to suit persistent software.
The launch also suggests Google sees sustained demand for hosted AI agents from individual developers and small teams that do not want to operate infrastructure directly. Those users often need a predictable endpoint, steady uptime and low monthly cost more than scale-out request handling.
Cloud Run has traditionally been associated with containerised applications that respond to inbound requests and scale according to traffic. Cloud Run instances broaden that scope by adding a separate option for applications whose main requirement is continuity rather than elasticity.
For Google, the move extends an established serverless product into an area where developers have often had to choose between convenience and persistence. For users building always-on agents, Google is betting that a singleton managed runtime with a low entry price will be enough to shift some workloads away from virtual machines.