Managani Docs

Appearance

AI Automation ​

AI Automation contains two site-scoped agents: Chat AI for customer conversations and lifecycle agents for event-driven outreach.

Chat AI ​

Chat AI can review eligible widget conversations, answer from approved knowledge sources, and escalate when a human should take over. Configure its instructions, enable state, escalation recipients, and run behavior before turning on scheduled processing.

Use the run history and outgoing-email list to verify what the agent did. Keep customer-only knowledge scoped to the matching customer.

Lifecycle Agents ​

Lifecycle agents watch signed customer activity and send timely onboarding, support, or sales email. Each agent has:

  • A personality and message instructions
  • A verified site email identity using system SMTP or site-specific SMTP
  • Daily outreach and reply limits
  • Retry limits, quiet hours, and waiting timeouts
  • Human escalation recipients and instructions

Rules can trigger when a customer first appears, performs an action, does not perform an action, becomes inactive after an action, has no activity, or reaches a trial-day window. Rules can filter plans, watch named tracking actions, wait before acting, and run in priority order.

Operate Safely ​

Start agents disabled. Test the email identity, confirm Tracking emits the intended actions, review matched accounts and conversation history, then run manually before enabling scheduled work.

Use escalation when customer intent is ambiguous or a request needs account-specific human judgment.

Review a Rule Before Enabling It ​

Write down the customer situation the rule should recognize and the response that would be helpful. Check that the tracked action and plan filters match real customer activity. A missing event can mean that an integration is incomplete, rather than that a customer needs another reminder.

Run a small manual review first. Inspect the selected customer, available conversation history, proposed message, and sending identity. Verify quiet hours and outreach limits against the experience you want customers to receive.

Keep escalation instructions specific enough that the agent can stop and involve a person. Review run history after changes to knowledge, instructions, or tracking, since those changes can alter which customers qualify and how the agent responds.