Agent Builder
The Agent Builder transforms bot configuration from a simple toggle page into a full training experience for your AI employee. Configure everything from personality and business knowledge to sales tactics and compliance rules — all in one place.
Navigate to Conversations > Agent Builder (or the old Bot Config route, which now redirects here).
Overview
The Agent Builder is organized into 8 tabs, each controlling a different aspect of your AI agent:
| Tab | What It Controls |
|---|---|
| Identity | Name, avatar, personality sliders, system prompt, languages, reply settings |
| Knowledge | Company story, custom instructions, business policies, store locations, FAQ, pricing rules |
| Sales | Sales mode, discounts, promo codes, upsell/bundle rules, lead qualification, closing techniques |
| Support | Objection handling, complaint flows, refund authority, shipping issues, handoff, follow-up scheduling |
| Media | Voice messages, images, video, location, documents, unsupported media fallback |
| Platforms | Per-platform behavior (Instagram, Facebook, WhatsApp, TikTok) — tone, features, reply length |
| Performance | Live stats dashboard — total threads, bot vs. human replies, resolution rates |
| Safety | Blocked topics, never-say phrases, escalation keywords, approval actions, data privacy, compliance |
All tabs modify a single BotConfiguration object. Click Save once to persist all changes.
Identity Tab
Agent Persona
- Agent Name — The name your AI uses when introducing itself (e.g., "Sarah", "ShopBot")
- Avatar URL — Optional profile image for the bot
- System Prompt — The master instruction that shapes all AI behavior (up to 5,000 characters). Think of it as the agent's job description.
Personality Sliders
Fine-tune your agent's tone with four 0–100 sliders:
| Slider | Low (0) | High (100) |
|---|---|---|
| Humor | Strictly professional | Playful, uses jokes |
| Formality | Casual, friendly | Very formal, corporate |
| Emoji Usage | No emojis | Frequent emojis |
| Verbosity | Brief, concise | Detailed, thorough |
Languages
Select which languages the agent can reply in (English, Arabic, French, Spanish, German, Italian, Portuguese, Turkish, Hindi, Chinese). Enable Auto-detect language to let the AI match the customer's language automatically.
Reply Settings
- Auto-reply to comments / DMs — Toggle each channel independently
- Max auto-replies per thread — Prevent infinite loops (default: 3)
- Reply delay — Add a human-like delay in seconds before responding
- Confidence threshold — Minimum AI confidence to send automatically; below this, replies are held for review
- Working hours — Restrict auto-replies to business hours with timezone support
Knowledge Tab
Company Story
A free-text description of your business that the AI references when answering questions (up to 5,000 characters).
Custom Instructions
Additional directives beyond the system prompt — useful for seasonal campaigns, temporary policies, or specific response styles.
Business Policies
Structured fields for your core policies:
- Return policy
- Refund policy
- Shipping policy
- Warranty policy
The AI references these directly when customers ask policy-related questions.
Store Locations
Add physical store locations with name, address, phone, and hours. The AI can direct customers to the nearest location.
FAQ
Manually add question-answer pairs, or use the Auto-Generate button to create FAQ entries from your conversation history.
Pricing Rules
- Can quote prices — Allow the AI to mention specific prices
- Currency — Default currency for price quotes
- Show tax — Include tax information in price responses
Sales Tab
Enable Sales Mode to let your AI agent assist with commerce.
Discount Authority
Control what discounts the agent can offer autonomously:
- Maximum discount percentage
- Maximum fixed discount amount
- Currency for fixed amounts
Promo Codes
Add allowed promo codes as tags. The agent will only suggest codes from this list.
Upsell Rules
Create trigger-based upsell messages. Each rule has:
- Trigger keywords — Words that activate the upsell
- Upsell message — What the agent says
Bundle Offers
Define product bundles with a name, discount percentage, and description.
Lead Qualification
Enable AI-powered lead scoring with a configurable threshold (0–100). Leads scoring above the threshold are flagged for human follow-up.
Closing Techniques
Toggle sales techniques the agent can use:
- Urgency — "Limited stock available!"
- Social proof — "500+ customers bought this last week"
- Scarcity — "Only 3 left at this price"
Support Tab
Objection Handling
Build a playbook of common objections and ideal responses. Each entry has:
- Objection — What the customer says (e.g., "too expensive")
- Response — How the agent should respond
- Category — Classification (price, quality, trust, shipping, other)
Complaint Resolution Flows
Define step-by-step flows for handling complaints:
- Trigger phrase — What activates the flow
- Steps — Toggle individual steps: Apologize, Investigate, Offer Solution, Escalate, Follow Up
Refund Authority
- Enabled — Allow the agent to process refunds
- Max refund amount — Upper limit for autonomous refunds
- Approval-above threshold — Amounts above this need human approval
Shipping Issue Handling
- Auto-lookup — Automatically check shipment status
- Offer compensation — Allow the agent to offer compensation
- Max compensation — Upper limit for compensation offers
Handoff Protocol
Configure what happens when the AI hands off to a human:
- Handoff message — What the customer sees during handoff
- Assign-to role — Which team role receives the handoff
- Preserve transcript — Include full conversation history
Conversation Goals
Select the goals your agent should pursue: convert to sale, resolve complaint, gather feedback, qualify lead, retain customer, or upsell.
Media Tab
In MENA and Arabic markets, voice messages are the primary communication method. Customers frequently send voice notes, product photos, and location pins. Without media handling, your AI agent is deaf and blind to the majority of customer interactions.
Voice Messages
- Enable voice — Process incoming voice messages
- Transcription languages — Select languages for voice-to-text (Arabic, English, French, Spanish, Hindi, Turkish, Urdu)
- Reply format — Choose how the agent replies to voice: text only, voice only, or both
Images
- Enable images — Process incoming images
- Use cases — Select what the AI can do with images: product lookup, damage detection, receipt parsing, visual search
Video
- Enable video — Process incoming videos
- Max duration — Maximum video length in seconds
Location
- Enable location — Process location pins (works with Store Locations in Knowledge tab)
Documents
- Enable documents — Process document attachments
- Document types — What types to parse: receipts, invoices, warranty cards, IDs
Unsupported Media Fallback
A custom message sent when a customer sends media the agent can't process.
Platforms Tab
Configure per-platform behavior for each connected social platform. Each platform card includes:
- Enable/disable for that platform
- Tone override — Platform-specific tone adjustment (e.g., more casual on Instagram)
- Max reply length — Character limit for replies on that platform
Platform-Specific Features
| Platform | Available Features |
|---|---|
| Story replies, Comment-to-DM redirect | |
| Quick replies, Comment-to-DM redirect | |
| Catalog messages, Quick replies, List messages | |
| TikTok | (Base features) |
Performance Tab
A live dashboard showing your AI agent's performance:
- Total conversations handled
- Bot replies vs. Human replies with ratio percentage
- Resolved conversations
- Escalated conversations
- Visual bar chart comparing bot and human activity
Click View Full Analytics to access detailed conversation analytics.
Safety Tab
Blocked Topics
Add topics the AI should never discuss (e.g., "politics", "religion", "legal advice"). The AI will gracefully deflect or escalate instead.
Never-Say Phrases
Words or phrases the AI must never include in any response (e.g., competitor names, inappropriate terms). The compliance filter strips these automatically.
Escalation Keywords
Words that trigger immediate escalation to a human agent (e.g., "lawyer", "sue", "manager").
Actions Requiring Approval
Define actions that need human approval before the AI can execute:
- Action name — What the action is
- Threshold — Dollar amount that triggers approval
- Approver role — Who can approve (Admin, Owner, Editor)
Data Privacy
Control what customer data the AI can reference:
- Reference purchase history
- Reference email address
- Reference phone number
- GDPR mode — Strict European privacy compliance
- Data retention — How many days to keep conversation data (1–3650)
Industry Compliance
Select your industry for specialized compliance rules: Retail, Healthcare, Finance, Legal, Food, or General.
Audit Trail
Enable full audit logging of all AI decisions and actions.
Tips
Don't try to configure all 8 tabs at once. Start with Identity (give your agent a name and personality) and Knowledge (add your company story and FAQ). Then expand to other tabs as needed.
If you serve Arabic-speaking customers, enable Voice Messages in the Media tab immediately. Voice notes are the default communication method in these markets.
Set the confidence threshold high (e.g., 0.9) initially so most replies go to review. Lower it gradually as you build trust in the AI's responses.
Related Pages
- AI Auto-Reply — The LLM pipeline behind auto-replies
- Unified Inbox — Where all conversations live
- Escalation Rules — Automatic escalation configuration