Skip to content

Bot Configuration

View as Markdown

The Settings tab (/ai-bots/{botId}/config) controls how a bot behaves on live customer conversations. Every change saves automatically as soon as you select it.

Choose how the bot participates in chats and its confidence threshold for auto-sending:

  • Off: Bot is paused and will not draft or auto-reply.
  • Draft: Bot prepares suggested replies in the chat composer for human agent review before sending (DRAFT).
  • Cautious: Auto-sends (AUTO_SEND) only when knowledge base documents match very closely (semantic distance threshold 0.42).
  • Balanced (Recommended): Auto-sends when the answer meets the standard confidence threshold (0.50).
  • Bold: Auto-sends even when knowledge matches moderately (0.60).
  • Always: Bypasses the confidence gate and immediately auto-sends every drafted reply (anti-spam safeguards still apply).
  • Language Model: Choose from 7 supported models:
    • DeepSeek V3 (Recommended) — deepseek/deepseek-chat
    • Gemini 2.5 Flash Lite — google/gemini-2.5-flash-lite
    • GPT-4o Mini — openai/gpt-4o-mini
    • Gemini 2.5 Flash (Ultra-fast) — google/gemini-2.5-flash
    • DeepSeek R1 (Deep Reasoning) — deepseek/deepseek-r1-0528
    • Gemini 2.5 Pro — google/gemini-2.5-pro
    • GPT-4o — openai/gpt-4o
  • Creativity Level (Temperature):
    • Precise & Strict (0.3): Strictly adheres to knowledge base facts.
    • Balanced & Natural (0.7, default): Friendly tone while staying faithful to your data.
    • Adaptive & Creative (1.0): More flexible phrasing for consultative sales.
  • Response Length: Choose Concise (1–2 short chat sentences) or Detailed & Comprehensive.

3. Human Response Delay & Typing Simulation

Section titled “3. Human Response Delay & Typing Simulation”
  • Simulate Human Typing Rhythm: Choose Off (Send immediately), 2 – 4 seconds (Fast), 3 – 7 seconds (Natural - Recommended), 5 – 10 seconds (Deliberate), or Custom minimum (0–60 seconds) and maximum (up to 120 seconds).
  • Show “Typing…” Indicator: Toggle sending typing status to Zalo while the bot prepares a reply (enabled by default).
  • 24/7 Continuous: Bot operates around the clock.
  • Custom Shifts / Working Hours: Set Working Hours (for example, 08:30 – 18:00) and Applicable Days of Week (Mon–Sun).
  • Action Outside Shift:
    • Draft suggestion: Only drafts replies outside working hours for agents to review later.
    • Pause Bot: Disables AI outside working hours. When selected, you can optionally enable Send off-hours auto-responder with a custom message (zero AI token cost and anti-spam protected).
    • Auto-send: Lets the bot auto-reply outside the configured daytime shift.
  • When Information is Not Found in Knowledge Base:
    • Save as Draft (DRAFT): Avoids ungrounded guesses and leaves the reply for human agents.
    • Send Waiting / Escalation Message: Sends a custom holding message, with an optional toggle to Pause Bot after sending waiting message.
  • AI Pause Duration in Conversation: Choose 1 hour, 4 hours (default), 12 hours, 24 hours, or Until manually re-enabled.
  • Pause on Negative Sentiment / Complaints: Automatically halts bot replies when upset customers, profanity, or complaints are detected (enabled by default).
  • Max Consecutive Bot Replies: Choose Unlimited, 1, 2, 3 (Recommended), 5, 10, or Custom (1–100 messages) before pausing for staff takeover.
  • Human Support Request Keywords: When a customer message matches any configured keyword, the bot steps aside for a human agent. Default keywords include gặp nhân viên, gặp người thật, tư vấn viên, alo shop, and hotline. You can add or remove keywords at any time.
  • Customer Message Buffer (Debounce): Choose Off (Reply each text), 5 seconds, 10 seconds (Recommended), or 15 seconds. The bot waits after the customer’s latest message to bundle short fragmented messages into a single reply.