Agent runtimes

Agent runtimes are the deployed surface for long-lived agent-backed logic. Use them when the thing you are shipping is not just a web frontend or a bursty function, but an operational agent system that should stay live as a product surface.

Pricing

Agent runtime pricing in Compute Tokens

Agent runtimes are billed as always-on deployed surfaces. Use them when a continuously available agent-backed runtime is the product you need to ship.

Agent runtime

0.3 CT / minute

Billed while the runtime stays deployed and available.

What agent runtimes are for

Use agent runtimes when ACP should publish an always-available agent-backed service surface. This is the right fit for operational copilots, agent APIs, embedded assistants, and product logic that should stay live outside the lifetime of one thread.

Agent runtimes are the deployment surface for agent-backed products, not just internal execution.

Bind runtimes to auth, databases, or web apps when the agent should sit inside a larger product boundary.

Use the same lifecycle methods as other ACP resource managers.

Full agent runtime lifecycle in JavaScript

The runtimes manager mirrors the generic resource lifecycle while keeping your code explicit about the resource you are deploying.

Use client.runtimes or client.agentRuntimes depending on your preferred naming.

The manager supports creation, deployment, invocation, bindings, logs, runtime inspection, files, and deletion.

Operate an agent runtime end to end

Full agent runtime lifecycle in Python

Python exposes the same runtime surface for backend services and automation layers that need to control deployed agent systems.

Use client.runtimes or client.agent_runtimes depending on the imported alias you prefer.

The Python manager mirrors the same runtime lifecycle operations as JavaScript.

Operate an agent runtime end to end