AI Configuration

You can tune AI behavior without changing code.

Where to configure

Go to AI Agent in the main sidebar. The hub starts with a standalone Overview page (high-level activity, drafts pending review, recent auto-sends), then splits into two groups:

Monitoring

• Knowledge — what the KB retrieval is matching against and where coverage gaps are
• Lessons — review queue for eval-flagged conversations, pending coaching notes, and the active lessons the AI applies at draft time. See How the AI learns from your team.
• Spend — usage and cost breakdowns, plus the workspace spend cap

Configuration

• Features — on/off toggles for auto-triage, draft replies, auto-reply (plus its threshold), similar tickets, translation, knowledge retrieval
• Persona — tone, company name, brand voice, custom instructions
• Rollout — staged rollout controls for gradually enabling AI per segment
• API Keys — OpenRouter (required) and optional direct Anthropic / OpenAI / Google keys

Live widget conversations are monitored from the dedicated Live Chat section in the main sidebar, not under AI Agent.

Recommended configuration order

1. Set brand tone and voice description.
2. Add strict policy instructions (what never to say/do).
3. Verify your knowledge base has current policy/process content.
4. Confirm templates for common reply types.
5. Start with conservative confidence thresholds and tighten after review.

Writing strong instructions

Good instructions are explicit and testable:

• "Never confirm refunds without manager approval."
• "Escalate legal/privacy requests immediately."
• "Always include next steps and expected timeline."

Avoid vague instructions like "be smart" or "sound better".

Knowledge base and templates

AI quality depends heavily on KB quality.

• Keep articles specific, current, and de-duplicated.
• Prefer short, actionable policy entries.
• Use templates for frequent workflows so AI has stable structure.

Related:

• Knowledge Base retrieval
• Creating articles
• AI feature toggles