Skip to main content
The Settings page controls how your workspace behaves — from AI configuration to spam handling to data retention. This page covers every setting available.

​
Widget

The master switch for your chat widget, plus workspace-wide chat behavior.
  • Widget enabled — When disabled, the widget does not load on your site at all. Visitors will not see the chat bubble. Use this to temporarily take your widget offline without removing the embed code from your site. When you re-enable it, the widget starts working immediately.
  • Track visitor pages — Show which pages visitors browse during conversations.
  • Human handoff — Allow visitors to request a human agent from the widget.
  • Auto-close inactive conversations — Automatically closes AI-handled web conversations when the visitor stops responding. A warning message is sent 5 minutes before closing. When enabled, set Close after (minutes) — 10 to 1440 minutes (24 hours).
  • Allow AI to resolve conversations (Pro+) — When enabled, the AI closes conversations after the visitor confirms their question is answered, with a 30-second grace period to keep talking.
  • Workspace timezone — Used for business hours routing rules and SLA business-hours-only counting.

​
AI Provider

Configure which AI model powers your chatbot.
  • Provider — Choose between OpenAI and Anthropic.
  • Model — Select a specific model from the chosen provider. Options vary by provider. See AI Provider Setup for model comparisons.
  • API Key — Enter your API key. Once saved, the key is masked for security and shows a Configured badge (or Key Error if it can no longer be decrypted). See Security & Privacy for how keys are stored.
  • Max Conversation Tokens — Set a cap on how many tokens a single conversation can use, from 1,000 to 500,000. This prevents runaway costs from unusually long conversations.
  • Test Connection — Click to verify your API key works and the model is reachable. A status indicator shows whether the connection succeeded.
If no provider or key is configured yet, an inline setup guide walks you through creating an API key on OpenAI’s or Anthropic’s site.

​
Allowed Origins

A list of domains where your widget is allowed to run. The widget will only load on pages served from these origins. Enter one origin per line, including the protocol:
This prevents your widget from being embedded on unauthorized sites. For local development, add http://localhost:5173 or whichever port you use.

​
Knowledge Base Thresholds

Fine-tune how your knowledge base search works.
  • Minimum Similarity — The lowest relevance score a knowledge base chunk must have to be considered for a response. Higher values mean stricter matching (fewer but more relevant results). Range: 0 to 1.
  • High Confidence Threshold — When a chunk scores above this threshold, the bot treats it as a highly confident match. Range: 0 to 1.
Leave these at their defaults unless you are seeing too many irrelevant responses (raise the minimum) or too many “I don’t know” answers (lower the minimum).

​
Help Center

Publish your knowledge base as a public, SEO-friendly help center.
  • Public help center — Toggle to enable. When enabled, published articles are available at your public URL.
  • Help center URL — A slug at <your-domain>/help/<slug> (lowercase letters, numbers, and dashes). Auto-generated when you first enable the help center; edit it here if you want a custom slug. The live URL is shown with a link and a copy button once set.
Individual articles are toggled live from the Knowledge Base page, not here. See Help Center for the full setup and article-management guide.

​
Response Cache

Toggle caching on or off. When enabled, ReplyBubble caches AI responses for similar questions. This means:
  • Faster responses — Cached answers are returned instantly without calling the AI.
  • Lower cost — Cached responses use zero tokens.
  • Consistent answers — Visitors asking the same question get the same response.
The cache is automatically invalidated when you update your knowledge base (add, edit, or delete an article) or switch personalities, so answers stay current. Entries also expire automatically after 24 hours.

​
Business Hours

Set your team’s weekly availability, interpreted in your workspace timezone.
  • Enable business hours — Toggle the schedule on or off.
  • Timezone — All times below are interpreted in this timezone (shared with the Workspace timezone field above).
  • Per-day schedule — Toggle each day (Monday–Sunday) on or off, with a start and end time when on. An end time earlier than the start time spans midnight (e.g. 22:00 to 02:00); identical start and end times mean open 24 hours. Days left off are treated as closed.
  • Take the widget offline outside hours — Outside your schedule, human handoff is hidden by default (visitors can still chat with the AI or leave a message). Enable this to switch the whole widget to your offline experience instead when you’re closed.
Outside business hours, routing rules with a business-hours condition also won’t match. If you enable “take the widget offline outside hours” without an offline contact prompt configured, a warning reminds you visitors may have no way to reach you outside your schedule.

​
Offline Mode

When offline mode is enabled and the AI provider is unavailable (no API key configured or a service outage), the widget falls back to a decision tree of canned responses instead of showing an error.

​
Configuration

  • Enable/Disable — Toggle offline mode on or off.
  • Welcome Message — The first message visitors see when offline mode is active.
  • Contact capture in offline mode — Toggle whether the offline flow shows a contact capture form at the end of the decision tree.
  • Contact capture prompt — Customize the message that asks for contact details.

​
Decision Tree Editor

Build a tree of questions and answers that guide visitors through common scenarios without AI.
  • Add response nodes with a question and up to several answer options
  • Each option can lead to another node (creating branches) or end with a final response
  • Maximum depth: 3 levels
  • Maximum nodes: 50 per workspace
The decision tree is useful for directing visitors to the right resource, collecting basic information, or providing quick answers to FAQs when AI is temporarily down.

​
Spam Detection

Control how ReplyBubble handles suspicious conversations. Shadow mode is useful when you want to monitor spam patterns before actively enforcing. Enforce mode is for workspaces that receive significant spam traffic. You can override individual spam classifications from the Contacts detail page.

​
Contact Capture

Control when and how the AI asks visitors for their contact information during conversations.
  • Auto-detect shared contact info — When a visitor shares their name, email, or phone in conversation, a brief confirmation card appears so they can verify and approve saving their details.
  • Capture mode (Pro+) — Determines how aggressively the AI asks for contact details:
    • Smart (recommended) — AI detects buying intent and asks naturally.
    • Conservative — Only asks when intent is very high. Best for support widgets.
    • Eager — Asks earlier in the conversation. Maximizes capture rate.
    • Manual only — Never auto-triggers. Form only appears via keyword triggers or API.
  • Minimum messages before asking (Pro+) — The AI waits at least this many visitor messages (2–10) before prompting for contact info.
  • High-intent keywords — When a visitor uses any of these words, the AI prioritizes asking for contact info. One per line or comma-separated.

​
Quotas

​
Data Retention

Set how many days of contact data to keep. Contacts older than the retention period are permanently deleted by the nightly cleanup; their linked conversations and form submissions are not deleted — they simply lose the contact link (conversation transcripts and analytics history are unaffected by this setting). Set to 0 to disable retention cleanup (never delete). The retention cleanup runs daily at 03:00 UTC.

​
Daily Token Quota

Set a daily spending cap for AI token usage. Once the quota is reached, new chat messages are rejected for the rest of the day (visitors see a generic “service unavailable” message, not a quota notice) while existing conversation history stays readable. Set to 0 for unlimited. The quota resets at midnight UTC.

​
Chat Behavior

Control real-time indicators shown to visitors in the chat widget.
  • Show typing indicators — Visitors see an animated indicator when an agent is composing a reply.
  • Show message delivery status — Visitors see checkmarks showing if their messages were sent and received.
  • Show visitor presence (Pro+) — See when visitors are online and their last activity in the inbox.

​
SLA Targets

Set first-response and resolution targets (Pro+). Conversations that miss a target are flagged in the inbox and can trigger notifications, and feed the SLA hit-rate column on the Agents leaderboard. Options: enable/disable, first-response target (1 minute – 7 days), resolution target (1 minute – 7 days), and whether to count business hours only (pausing the SLA clock outside your Business Hours schedule). See SLA for the full guide, including how targets interact with routing and notifications.

​
Backup & Transfer

​
Export Settings

Download a JSON backup covering the sections you choose:
  • Appearance — widget colors, theme, layout
  • AI Configuration — provider, model, system prompt
  • Personalities — bot personalities, active personality, and form overrides
  • Forms — contact capture forms and triggers
Use this to back up your configuration before making changes, or to transfer settings between workspaces. Other workspace-level settings on this page (AI thresholds, spam mode, quotas, business hours, etc.) are not included in the export.

​
Import Settings

Upload a previously exported JSON file. You’ll see a preview of what it contains before importing. If Personalities or Forms in the file have names that conflict with existing ones, choose how to resolve it:
  • Overwrite — Replace existing items with the imported version.
  • Rename — Create imported items with an “(imported)” suffix.
  • Skip — Keep existing items and skip duplicates.
Caution: Imported sections replace matching current configuration (per your conflict choice above). Export your current settings first if you want a backup.