Create a guarded AI Agent

  1. Configure and enable Settings → AI Provider.
  2. Open Auto Reply → Agents → Create AI Agent.
  3. Set Agent name, routing description/keywords, Agent prompt, Allowed topics, and Blocked topics.
  4. Choose a fallback language, previous-message context, Active/Inactive status, and reply approval mode.
  5. Save the agent. Add only approved knowledge files and permitted attachments.

New AI agents default to manual approval. Existing migrated agents may retain Automatic; check the saved mode rather than assuming. Enabling AI Provider does not assign or activate an agent.

AeroMailer interface: Create guarded AI auto-reply agents
Sample-data app screenshot. Select the image to view it at full size; your accounts and content will differ.

Import an agent folder

Use Import agent folder in the creation form, or write a prompt directly. The folder needs a top-level agent.md. Review supported imported knowledge and attachment resources. This creates managed resources, not a continuous source-code scanner. Never import credentials, private customer data, or material the agent must not disclose.

Test, assign, and approve

Use AI routing dry run / Run dry test to inspect routing, language, topic checks, and resource warnings without a provider call or sent reply. Assign the agent under Account assignments, then inspect Approval Queue or History. Real generation can share selected knowledge and conversation context with the configured provider and incur API charges.

Manage safely

The combined Agents list supports edit, status, archive/restore, and confirmed individual/bulk deletion. The agent can follow the sender's language; configure a fallback when unclear. Built-in guards exclude inappropriate automated/no-reply/security/marketing messages and unsubscribe requests. A prompt is not a confidentiality guarantee: keep resources narrow and review output. AI-off pauses AI-dependent work without deleting the saved agent.