Skip to main content

​
Agent Overview

​
Audience

This page is for developers, automation integrators, and maintainers accessing AiOS through agent flows, APIs, or a future CLI.

​
Goal and Scope

This page explains:
  1. how Agent / future CLI relate to Web
  2. why they share one server-side execution chain
  3. why production Agent/CLI flows still require step-up / approval

​
Core Concepts

  • Web, API, future CLI, and Agent are entry channels, not separate business systems
  • core capabilities are owned by the server-side control plane
  • the unified guardrail chain is AuthN -> AuthZ -> Approval -> Execute -> Audit

​
Standard Workflow

  1. acquire identity context
  2. call the unified server-side API
  3. let the control plane enforce authorization and required approvals
  4. receive task state, error semantics, and audit-linked results

​
Planned CLI Examples (Not Yet Implemented)

The repository does not currently ship these CLI subcommands. They remain illustrative of how a future command surface must reuse the same server-side execution chain.

​
Web UI Path

CLI and Agent do not depend on Web directly, but they share the same resource, approval, and error model as Web.

​
Pitfalls / Risk Notes

  • an Agent is not a backdoor around approval
  • Future CLI field names and resource models must not diverge from Web/API semantics