Skip to main content

Agent Runtimes

Agent Runtimes provide a unified way to deploy, run, and supervise AI agents across interfaces such as Platform, VS Code, JupyterLab, and CLI. They combine managed execution with explicit policy controls so teams can move fast without losing runtime ownership.

Use Agent Runtimes when you need to:

  • Run long-lived agent workflows reliably.
  • Control tool usage through approvals and constraints.
  • Keep configurations portable with versioned specs.
  • Observe execution behavior with logs, traces, and evals.
  • Connect runtime workflows to interactive chat and UI outputs.

What You Get​

  • Portable Specs: Define goals, model choices, tools, skills, memory, and constraints in reusable specs.
  • Durable Execution: Run in managed runtimes with restart and recovery semantics.
  • Governance: Apply identity- and permission-scoped guardrails for safer production behavior.
  • Tool Control: Configure automated or human-approved tool execution.
  • Observability: Inspect execution via status indicators, traces, and evaluation signals.
info

For interactive conversational capabilities, see AI Chat.

For runtime lifecycle operations (launch, assign, pause/resume, terminate), see Runtimes.

Agent Runtimes are designed for teams that need both fast delivery and long-term maintainability: policy-first automation, predictable operations, and low migration friction across providers and environments.

Feature Coverage Map​

The following feature areas are fully documented in this guide: