> Source: https://builder.ema.ai/v2/autopilot/permissions-scoping > Title: Permissions and Scoping # Permissions and Scoping Autopilot works inside your account's boundaries by design. Every session runs as you, in your tenant, so Autopilot sees exactly the AI Employees (AIEs), Knowledge bases, Tools, and conversations you can already see — nothing more. When it needs to do something significant, it pauses and asks you first. This page covers what Autopilot can reach, when it asks for approval, how to respond, and how your role shapes the interface. ## What Autopilot can access Autopilot operates within your tenant — the organization-level account you belong to on the Ema platform — and acts with your identity and your permissions. It can read and act on the same resources you can reach through the regular interface: Resource What Autopilot can do AI Employees View, create, update, clone, delete, and test AI Employees in your tenant Workflows Load, edit, validate, and deploy the workflow behind an AI Employee Knowledge bases List, upload, search, copy, and remove data sources Tools and integrations Browse and search available integrations and actions, and wire Tools into a workflow Conversations Start, send to, rename, and delete test conversations Dashboards Upload data, view rows, and trigger a workflow run for a row Autopilot also knows which page you are on. If you are viewing a specific AI Employee, it already has that context — you can say "add a step that summarizes the ticket" without naming the AIE. > [INFO] > **Tenant isolation.** Autopilot cannot reach AI Employees, data, or conversations that belong to a different tenant. If you are a member of more than one organization, each Autopilot session is scoped to the tenant you are currently signed into. Switching tenants starts from that tenant's resources only. **Example prompts** > Show me all the AI Employees in my tenant. > What data sources are connected to this AI Employee? ## Approval prompts Some actions have lasting consequences. Before Autopilot does anything destructive or hard to reverse, it pauses the conversation and asks for your explicit approval. This keeps you in control — Autopilot does not make irreversible changes on its own. ### When Autopilot asks before acting Autopilot shows an approval prompt before it does things like: - **Deleting an AI Employee** — Autopilot names the AI Employee before removing it. - **Deleting a data source** — this is permanent, so Autopilot always confirms. - **Deleting a test conversation** — the conversation history is lost. - **Deleting an AI Employee group** — the grouping is removed (the AIEs inside are not deleted). - **Deploying a workflow** — this pushes changes to your live environment. For routine, reversible actions — listing resources, loading a workflow to edit, searching a Knowledge base — Autopilot proceeds without interrupting you. ### Responding to an approval prompt When Autopilot needs permission, the conversation shows an **Approval needed** card with the action it wants to take and two buttons: 1. **Approve** — Autopilot carries out the action and resumes the task immediately. 2. **Deny** — Autopilot stops, acknowledges your decision, and does not proceed. Once you respond, the card collapses into a compact resolved row showing what you decided (for example, "you replied Approve"), so the conversation history stays readable. > [TIP] > **Example.** If you say "Delete the Customer Support AI Employee," Autopilot shows an approval card naming that AI Employee before anything is removed. Nothing is deleted until you select Approve. ### Answering follow-up questions Sometimes Autopilot needs more information rather than permission. In that case it asks a **Question** instead of showing approval buttons. A question appears as a card with one of two input shapes: - A set of **option buttons** when Autopilot has a few specific choices in mind — select one to answer without typing. - A **text box** (placeholder "Type your reply…") with a **Send** button when the answer is free-form. You can also **Cancel** to skip the question. After you respond, the card collapses into the same resolved row ("you replied …") as an approval prompt. > [TIP] > **Example.** If you say "Create a new AI Employee for handling refunds," Autopilot may ask a follow-up such as "What tone should this AI Employee use with customers?" before it builds. ## When something is out of scope If you ask Autopilot to do something it cannot do, it tells you directly rather than failing silently. Common cases: - **The resource doesn't exist.** If you reference an AI Employee or data source it can't find, Autopilot says so and may suggest checking the name. - **A prerequisite step was skipped.** Some actions depend on an earlier one — editing a workflow's structure requires loading it first, for example. Autopilot either handles the prerequisite for you or tells you what it needs to do first. - **A turn is already running.** While Autopilot is working on a request in the same session, the composer shows a **Stop** control instead of a send button. Wait for the current turn to finish, or stop it, before sending the next message. - **You're at the session limit.** Each tenant has a cap on concurrent Autopilot sessions. When you reach it, **New chat** is disabled and the welcome screen notes you're at the limit. Close a running session from **Recent chats** to free a slot. See [Limits, availability, and troubleshooting](/builder/v2/autopilot/limits-availability-and-troubleshooting). You can track every session from the **Recent chats** list. Each row carries a status indicator: Indicator Meaning Spinner The session is running — Autopilot is working on a turn Person icon and a "Needs your reply" pill Autopilot is waiting on an approval prompt or a question Warning icon The session ended in an error Green dot There is unread activity in that session you haven't opened yet ## Role-based visibility Your role in the Ema platform determines what appears in the interface. For Autopilot, the practical difference is whether you can create AI Employees: - If your role includes permission to create AI Employees, the **Build your own AI Employee** option is available on the AI Employees page, and you can also ask Autopilot to create one for you. - If your role does not include that permission, the create option is hidden — not shown as a grayed-out control — and Autopilot cannot create, clone, or update AI Employees on your behalf. Every other Autopilot capability is available to any authenticated user in the tenant. Editing within your permissions, testing, listing, searching, deploying, generating documents, and asking questions all work the same regardless of role. Because Autopilot acts as you, it can never do something your own account is not allowed to do. ## What's next - [Core actions](/builder/v2/autopilot/core-actions) — the full set of things Autopilot can build and change. - [Modes, feedback, and tips](/builder/v2/autopilot/modes-feedback-and-tips) — Fast vs Thorough, the two views, and getting better results. - [Limits, availability, and troubleshooting](/builder/v2/autopilot/limits-availability-and-troubleshooting) — limits, the availability flag, and the fleet-status banner.