How agent-planned deployments work

Every Nexplane deployment begins with structured intent: model source and deadline. The planner agent turns that into a concrete execution plan with accelerators, serving stack, reliability policy, and hourly/monthly serving cost estimates.

Plans stay visible during review. Orchestration steps — provisioning, runtime setup, endpoint registration — only appear after you approve. That separation keeps review focused on outcomes, not infra noise.

If something fails mid-run, recovery agents retry with context from prior steps rather than restarting blindly. Logs, metrics, and cost roll up to the run detail view.