Agent substrate can power agents on Kubernetes with kagent | Solo.io

Open original

Solo.io and Google teamed up to integrate Agent Substrate into kagent, enabling fast suspend/resume, scale-to-zero and secure sandboxing for AI agents on Kubernetes. It uses pre-provisioned worker pools, snapshots to storage, and lightweight isolation (gVisor or Firecracker) to cut boot times to milliseconds and minimize idle resource use.

  • kagent now integrates Google's Agent Substrate to suspend/resume/scale-to-zero AI agents on Kubernetes in 50-200ms instead of the seconds pods normally take to boot
  • It works by pre-provisioning worker pools and snapshotting idle agent states to object storage (S3/GCS), then restoring them on demand
  • A parallel control layer handles deploy/suspend/resume logic instead of routing everything through the Kubernetes API, avoiding overload
  • Each agent instance can be sandboxed with gVisor, Firecracker, or Bubblewrap/Landlock/seccomp, with Solo.io's agentgateway enforcing egress control and credential injection at the network edge

Solo.io’s kagent now plugs into Google’s open-source Agent Substrate to handle AI agents on Kubernetes. Traditional pods sit idle until they’re needed—wasting resources and spinning up in seconds, not milliseconds. Agent Substrate solves this by letting you pre-provision worker pods (or VMs) and then rapidly suspend, snapshot, resume, or scale to zero individual agents in 50–200 ms. It avoids overloading the Kubernetes API by offloading deploy/suspend/resume logic to a parallel control layer, and it snapshots entire agent states to object storage like S3 or GCS. Under the hood, you declare a WorkerPool of generic pods, and an ActorTemplate for each agent type. When a client request arrives, Agent Substrate’s control plane routes it to an already-loaded actor or boots a new one from the template. Idle actors get snapshotted and evicted to free up capacity. You can sandbox each agent instance using gVisor, Firecracker microVMs, Bubblewrap/Landlock/seccomp, and enforce security and governance through Solo.io’s agentgateway, which handles all egress and can inject credentials at the network edge. In kagent, you can choose between standard one-agent-per-pod deployments or Agent Substrate mode. Once you pick Substrate, agents like LangChain, CrewAI, or OpenClaw harnesses run as suspended actors inside the worker pool. The UI shows which actors live in which pods, and you manage them alongside your existing Kubernetes workloads. This brings sub-second startup, strict isolation, and storage-backed snapshots to AI agents without rebuilding your cluster.

Ask about this article