A pod per AI agent session does not scale. Solo.io's Agent Substrate puts idle agents to sleep on Kubernetes

Teams that run AI agents on Kubernetes can now buy a supported runtime that suspends idle agent sessions instead of keeping a pod alive for each one. Solo.io made enterprise releases of Agent Substrate and kagent generally available on November 10, 2026, and says OutSystems and Swissquote already run them in production.

Key points

• Solo.io announced general availability of its enterprise releases of Agent Substrate and kagent on November 10, 2026, with OutSystems and Swissquote named as launch customers in production.
• Agent Substrate runs each AI agent session as an "actor" with its own sandbox and gateway, and Solo.io says idle sessions are suspended and resumed on demand on existing Kubernetes clusters.
• kagent is the layer for building and managing agents, tools and skills; the enterprise release adds granular access control, agent tracing, monitoring, evaluations and air-gapped installs, according to Solo.io.
• Solo.io co-maintains the open source Agent Substrate project with Google, and says the community's 1.0 release counts over 100 contributors across more than 40 organizations and more than 6,000 stars on GitHub.
• Solo.io also released a sizing calculator on the kagent site to help teams size environments and set service-level objectives.

Why is running AI agents on Kubernetes a problem?

Agent sessions strain a cluster because each agent session holds its own state across many turns and spends much of its time waiting. Kubernetes is already where most of these teams would put them: in the Cloud Native Computing Foundation's annual survey, released in January 2026, 82% of container users ran Kubernetes in production, and 66% of organizations hosting generative AI models used it for some or all of their inference.

The usual pattern, one pod per workload, was built for stateless services that stay busy. Solo.io's release puts the cost plainly: at tens of millions of sessions, a dedicated pod for each ties up resources while agents wait on models, tools or a person. In a post announcing kagent 1.0, Christian Posta, Solo.io's VP and global field CTO, adds that agents run code a model wrote, build up a workspace a restart would throw away, and act for individual users that one shared service account cannot tell apart.

How does Agent Substrate run AI agents on Kubernetes?

Agent Substrate treats each agent session as an actor, the basic unit it schedules, rather than as a pod. Every actor carries two required controls, according to Solo.io: a sandbox that isolates what the agent executes, and a gateway that decides who can reach it and what it can reach. The actor's state, identity and access rules move with it as the compute underneath changes.

When an actor goes idle, the runtime suspends it and hands its worker back; when a request arrives, it resumes the actor on an available worker. The Agent Substrate project site says each agent runs in its own microVM or gVisor sandbox and that a suspended actor resumes in under 500 ms on any ready worker.

Agent Substrate is trying to do for agentic workloads what kubernetes did for traditional deployments - to make the orchestration of them a non-issue.

— Tim Hockin, Distinguished Software Engineer Kubernetes and GKE, Google

What does the enterprise release of kagent add?

The enterprise release packages the open source projects as a supported runtime. Solo.io says it uses agentgateway as each actor's gateway, applying identity and policy checks to traffic in and out, while kagent gives developers one place to build and manage agents, tools and skills. On top, Solo.io lists granular access control, agent tracing, infrastructure monitoring, evaluations built from real execution traces, enterprise support and air-gapped installation for regulated environments.

The two launch customers describe why they moved:

With the number of agents we expect to support, running a dedicated, always-on pod for every agent simply isn't possible. Agent Substrate gives us sandboxed runtimes that suspend when idle and resume on demand.

— Ebenezer Schubert, VP Engineering, OutSystems

Our Codename Diego financial-crime investigation agents and MCP servers are already running in production on kagent GA.

— Andrew Boswell, Head Information Technology, Swissquote

What does this mean for CTOs planning agent workloads?

Our analysis: the release turns the question of where agents run into a platform decision rather than a per-team one, much like deciding when to build an internal developer platform. Three points stand out for engineering leaders:

• Capacity follows active work. If idle sessions give their compute back, cluster sizing tracks the agents that are working, not every open session; Solo.io's sizing calculator is built for that planning.
• Isolation is the default. Every session gets a sandbox and a gateway, which fits the principle that an AI agent should not hold the same access you do.
• It stays on Kubernetes. Solo.io says enterprises can scale on the Kubernetes infrastructure they already run, the same direction other vendors are taking, such as SUSE adding AI agents to Rancher Prime.

Frequently asked questions

Is Agent Substrate open source?

Yes. Agent Substrate is an open source project that Solo.io co-maintains with Google, and the community released version 1.0. Solo.io sells supported enterprise distributions of Agent Substrate and kagent alongside it.

Who is running it in production?

Solo.io names OutSystems and Swissquote as launch customers running the infrastructure in production. Swissquote says its financial-crime investigation agents and MCP servers run on kagent GA.

Can it run in an air-gapped environment?

Yes. Solo.io says its enterprise support and deployment options include air-gapped installations for environments with strict security and operational requirements.

How do I size a cluster for Agent Substrate?

Solo.io published a sizing calculator on the kagent project site, based on its performance and scale testing, to help teams size environments and set service-level objectives.

Sources

• Solo.io Brings Agent Substrate to Enterprise Production: press release from Solo.io via GlobeNewswire, 10 November 2026.
• Agent Substrate in Production: Announcing kagent 1.0: Christian Posta's technical announcement on the Solo.io blog, 10 November 2026.
• Kubernetes Established as the De Facto 'Operating System' for AI as Production Use Hits 82% in 2025 CNCF Annual Cloud Native Survey: the Cloud Native Computing Foundation's summary of its annual survey, 20 January 2026.
• Agent Substrate: the open source project's site.
• kagent: the open source project's site.