Source: kagent-dev/kagent README (Apache-2.0, CNCF sandbox project originally from Solo.io); 1.0 GA announced November 2026. Surfaced by André Lindenberg.
What changed
Before 1.0, kagent ran each agent on Kubernetes like a web service: a long-lived Deployment that served every conversation. In 1.0, agents run on a Substrate layer that manages their lifecycle (suspension, resumption, snapshots), and, per the post, each conversation gets its own sandbox pinned to the prompt and model it started with. A prompt change therefore reaches only conversations opened after it.
Building blocks (from the README)
Agents and AgentTemplates as Kubernetes custom resources: system prompt, tools, subagents, model config.
Harnesses choose how an agent runs: kagent's own Go and Python ADKs, Codex, Claude, or a custom runtime. Third-party coding agents run like native agents.
Sessions persist conversations and task history in PostgreSQL, independent of the running agent; you can suspend, resume, checkpoint and fork.
MCP tools for Kubernetes, Istio, Helm, Argo, Prometheus, Grafana, Cilium and others, shared through RemoteMCPServer resources.
Why it matters
Instruction stability. A running agent should not change instructions mid-task. Pinning per conversation is the agent equivalent of not hot-swapping code under an in-flight request.
Isolation. Per-conversation sandboxes limit blast radius when an agent runs tools or code.
Rollouts look like deployments. New prompts and models roll out to new conversations, so you can compare old and new side by side.
Tradeoffs
More sandboxes means more scheduling and cold-start cost; long-lived conversations keep old prompts, including old bugs, until they end.
You need a way to force-migrate or end conversations when a prompt has a security fix.
Session state in a shared database becomes the critical dependency.
Review checklist
Can a prompt or model change reach a conversation mid-task in our system?
How do we force an urgent fix onto running conversations?
Is each agent run isolated from other users' runs?
Can we fork a conversation from a checkpoint to debug it?
Related entries
R084 (always-on agent primitives), R035 (durable execution), R074 (policy on every mutation), R021.