Dark mode
Support AI assistant configuration

Support AI assistant configuration

The AI Assistant settings pair a support agent persona with guidance — instructions that shape how the AI speaks and behaves in your conversations.

Selecting a support agent

The Support agent select on the AI Assistant page chooses which agent persona handles support conversations. If none is selected ("No support agent selected"), AI support cannot start — enabling the assistant with no agent selected is blocked with a warning.

Agents available here are the workspace's configured support agents (see the Automation module's agent management for creating and editing agents). The selected agent's prompt defines its base behavior; your guidance adds workspace-specific instructions on top.

Guidance

Add guidance creates instruction items the AI follows alongside its base prompt. Each guidance item has:

  • Title — a short label for the rule.

  • Question patterns — phrases or intents this guidance applies to (multiple patterns, one per line).

  • Answer/instruction — the guidance text: what the AI should do when these patterns appear.

  • Language (optional) — scope the guidance to a language.

  • Status — active or disabled; disabled guidance is retained but ignored.

Writing effective guidance

Guidance works best as short, imperative rules tied to observable situations:

  • Good: "When asked about refunds, first ask whether the request is within 30 days of purchase."

  • Good: "Tone: concise, no emoji, never promise timelines."

  • Less useful: "Be helpful and friendly." (The base prompt already covers this.)

Patterns should reflect what customers actually type — include common phrasings ("cancel", "unsubscribe", "delete my account") rather than your internal terms.

How guidance interacts with answers

Guidance shapes how the AI responds, not what it can know:

  • Guidance is applied on top of retrieved content; it cannot make the AI answer from content that isn't in your knowledge sources.

  • Guidance does not override safety behavior — hard escalations (abuse, PII, safety) always win.

  • Conflicting guidance items are resolved by the agent; keep instructions consistent to avoid unpredictable behavior.

Disabling the assistant safely

Turning Enable off stops AI handling for new messages. Existing AI-handled conversations complete their current turn; no new AI replies are generated. Conversations that were escalated remain in their queues. Re-enabling resumes the pipeline for new messages.

Was this article helpful?