Configure
Brand the assistant, control AI behaviour, and tune the widget — identity, appearance, behaviour, and greeting quick replies.
Last updated 2026-05-23
The configure tab carries every per-assistant setting that doesn't belong to sources or scenarios. Save at the bottom of the page; the new config is published to the CDN within seconds (no re-embedding required).
Identity
Name
Displayed in the chat header. Up to 80 characters. Default: the name you gave during assistant creation.
Description
Internal only — visitors never see it. Use it for your own notes.
Avatar style
Choose one of three tiles:
- Bottts — a deterministic DiceBear robot generated from the name. Different name, different robot.
- Silhouette — a neutral person icon.
- None — no avatar in the chat header at all.
Greeting
The first message a visitor sees when they open the chat. Plain text; one or two short sentences works best.
Show greeting bubble
Toggle. When on, a small bubble of the greeting pops next to the launcher five seconds after the page loads, then auto-dismisses. Once a visitor dismisses it, it stays dismissed in their browser for 24 hours.
Appearance
Brand color
Hex colour picker. Sets the launcher background, the send-button colour, and the primary accent inside the chat panel. Default #18553D.
Position
- Bottom right — default.
- Bottom left — flips the launcher to the other corner.
Behaviour
Max tokens
Upper bound on AI-generated reply length, measured in tokens (roughly four characters each). Range 0–2000, default 256. Lower values feel snappier; higher values let the model elaborate.
Scenario match threshold
How tolerant the lexical matcher is when comparing visitor questions to scenarios:
- Strict — only very confident matches. Visitors who reword a question may fall through to AI. Use when correctness matters more than coverage.
- Balanced (recommended) — catches typical rewordings.
- Loose — permissive. Use only if your scenarios are unambiguous and you want maximum scenario coverage.
Scenario fallback message
Up to 200 characters. Shown on iOS (and other scenarios-only devices) when no scenario matches. Optional — if blank, the widget shows a generic “I don't have an answer for that yet”.
Greeting quick replies
Multi-select dropdown listing your published scenarios. Pick up to eight; their questions appear as buttons under the greeting in the chat panel, so visitors can one-tap a common question.
Saving
Click Save. The form validates, updates the assistant in Firestore, and enqueues an update-config task — that task republishes config.json (no re-embedding). A toast confirms the save.
config.json is republished, every fresh page load picks it up. Already-open chat sessions keep their current config until reload, or until the page calls element.refresh() — see JavaScript API.