Secure operators, not loose chatbots

Agent Operations

Agents should operate like controlled users of the system: dedicated machine identity, authenticated API calls, bounded permissions, visible actions, and reviewable logs.

Secure agent operations concept artwork
The agent is not the product by itself. The operating environment is the product.
A dedicated machine identity is useful when it is paired with API scopes, permissions, secrets management, and audit trails.
Generated apps make agent work visible because humans can inspect the same workflows, tasks, approvals, and results.
Agent operations foundry infographic

Machine identity

A dedicated machine gives the agent a concrete operating identity: known device, network context, local runtime, stored credentials, and controllable environment.

Dedicated workstationDevice identityNetwork presenceRuntime control

Authenticated API calls

The agent should act through approved API routes with role-bound permissions instead of unrestricted access to every system.

Scoped tokensService routesRole policyAction limits

Visible workflows

Agent work should appear inside dashboards, queues, approvals, drafts, holds, and event logs that people can inspect.

Task queuesApproval statesDraft reviewOperational logs

Human control

People need clear ways to pause, review, redirect, approve, revoke, or escalate agent activity.

PauseReviewEscalateRevoke

Agents need an operating system around them.

The dashboards page shows examples of the kinds of app surfaces people and agents can share.

View Dashboard Examples