Skip to content

Getting started

This page is the map. It tells you what you need before you begin and the five things you actually do with the platform, each linking to the detail.

What you need

  • A Microsoft 365 tenant you manage through GDAP. The platform acts on a client tenant using granular delegated admin (GDAP) plus consented app access — never a stored global-admin password. See Onboarding a client.
  • A source of tickets. Either a connected PSA (HaloPSA today; others on the roadmap) or the intake API. Tickets can also be entered directly for testing.
  • Operator access to the Portal. The Portal is where you review plans, approve or deny them, watch execution, and manage settings. Access is role-based once RBAC is enabled (see Admin controls → Users & permissions).

The five things you do

  1. Onboard a client. Establish GDAP, consent the platform's app access, and let the platform discover what it can execute in that tenant. This is the one step with real Microsoft-side friction; everything after it is in-app.
  2. Connect a PSA so real tickets flow in automatically — or pull/enter them on demand while you're getting set up.
  3. Let a ticket become a plan. When a ticket resolves to a managed tenant, you generate a plan; the AI reads the live tenant and constructs the change. See Running the platform.
  4. Review and approve. Read the AI's reasoning and every proposed operation — including the exact cmdlet and parameters for anything that runs in the customer tenant — then approve, deny, or ask for clarification.
  5. Watch it execute and report back to the ticket. Execution runs as a background job with per-step evidence; the outcome posts back to the originating PSA ticket.

Two things worth knowing up front

Plans and executions are background jobs

A complex ticket can take the AI many minutes to plan, and execution runs step by step against M365. Both are asynchronous — you submit, then the Portal polls and shows progress. You never sit on a frozen request.

Approval is mandatory and real

The platform is built so that generate → approve → execute are separate stages with a genuine pause in the middle. A plan cannot execute until a human approves it. That pause is the safety model, not a formality.