Skip to main content
Agent Application Platform turns tested Agent Projects into immutable releases and operates long-lived instances with durable workspaces, identity management, security controls, governance, observability, evaluation, and fleet management.
The platform turns a framework-native project into an operating product. It resolves and verifies dependencies, packages the complete graph as an immutable release, creates instances at the application’s privacy boundary, attaches a durable workspace and scoped authority, and runs the sessions that perform the work. In the paper’s seven-layer stack, L5 is the Agent Application Platform. A platform may also provide development tools for creating and testing the framework-native L4 Agent Application Project. The optional L6 Agent Application Store handles acquisition, entitlements, billing, publisher payments, and monetization. A platform learns what a project needs from the framework’s native configuration or from a portable Agent Application Manifest. It also supplies shared services for identity and authority, scheduling, event delivery, artifacts, evaluation and observability, governance, fleet upgrades, cost allocation, and usage enforcement. An agent registry is a governance catalog inside one organization: which agents, tools, MCP servers, and skills are approved, who owns them, and where they run. It belongs with the platform’s governance services. A store distributes applications across organizations and handles payment.

Agent Cloud and platform

Ordinary cloud infrastructure provides compute, storage, networking, and databases. Agent Cloud adds infrastructure for persistent agents, including durable workspaces, resumable execution, sandboxed tools, checkpoints, and per-instance metering. An Agent Application Platform is the larger product layer. It may use an Agent Cloud, provide those services itself, or assemble them from ordinary cloud services. See the paper’s Agent Application Platform and Agent Application Store sections.