Guardrails define what the agent will and won't do. The three core guardrails are brand voice (how it sounds), claim restrictions (what it can't say), and suppression lists (who it can't contact). Configure all three before going live.
Brand voice
Settings → Brand Voice. Upload 3–5 sample emails or marketing assets. The agent reads these and calibrates its draft style — sentence length, formality level, signoff style, vocabulary. You can have multiple voice configs (e.g. "founder voice" vs "AE voice") and assign different ones to different agents.
Claim restrictions
Settings → Compliance → Claim Restrictions. Add specific claims the agent cannot make: "never quote a specific dollar price," "never claim we're cheaper than Salesforce," "never make ROI guarantees." The agent's drafter validates against this list before any send.
Suppression lists
Settings → Suppression. Three suppression scopes:
- Workspace-wide — no agent ever contacts these.
- Per-agent — this specific agent skips them.
- Per-campaign — only this campaign skips them.
Upload via CSV or add manually. Email addresses, domains, or company names all work.
Conditional guardrails
Advanced: conditional guardrails fire only under specific circumstances. Example: "if recipient is in the EU, require explicit opt-in language," or "if recipient has > 5000 employees, route to human review before send." Configure in Settings → Compliance → Conditional Rules.
