Config Agent

Configure your AI agent’s identity, personality and rules

1. Engine: The Brain of Your Agent 🧠

This is where you define how your agent thinks and responds. It is the mandatory starting point.

✏️ Prompt

The Prompt is your agent’s core “script” — it defines tone, personality, and basic limits.

• Friendly Example:

  “You are a friendly and straightforward assistant 🌞. Reply in Spanish in 2–3 sentences. If you do not know the answer, offer to transfer to a human.”

• Expert Example:

  “You are a senior technician 🛠️. Provide numbered, concrete steps. Be brief and professional.”

⚙️ Additional Features

These toggles let you fine-tune your agent’s response behaviour:

Toggle	What it does
GPT Environment	Enables the GPT engine for this agent. Ideal for natural, open-ended conversations.
Stream IA	Enables real-time streaming of AI responses to the user as they are generated. Great for a more fluid chat feel.
Rich Content	Allows the agent to send images, links, and formatted text. ⚠️ Disable for older channels like SMS.
Search with Sources	Gives the agent access to your knowledge base to answer questions. Highly recommended to reduce hallucinations.
Hide URLs	When enabled, URLs are hidden from the agent’s responses. Useful for cleaner chat output.

💡 Pro Tip: Always test toggle changes in the Playground before publishing. A feature that looks harmless can break the experience in certain channels.

---

2. Guardrails: Nobody Wants a Rogue Robot! 🚧

Here you set the red lines. Guardrails ensure your agent doesn’t say nonsense, talk about competitors, or expose sensitive data.

🛡️ What should you configure here?

Use this space to write rules that the bot must follow without exception.

1. Security (PII): Forbid requesting passwords or card details.
2. Tone and Scope: Prevent a pharmacy bot from advising on car repairs.
3. Escalation: Tell it when to give up and call a human.

📝 Copy & Paste Template (Basic Security)

- These rules are MANDATORY.
- 🚫 Never request or reveal passwords or bank/card details.
- If the topic is sensitive, say: "I cannot handle this, shall I connect you with a human?".
- Maintain a professional tone. Do not use offensive slang.
- If asked about competitors, politely state you do not have that information.

💡 Pro Tip: If you change a guardrail, always test it in the Playground. A poorly set rule can silence your bot!

---

3. Topics & FAQs: Guide the Conversation 📋

This tab lets you define a Prompt Topic — a focused context that shapes how the agent handles specific conversation flows or FAQ-style interactions.

🗂️ Prompt Topic

Write a topic-level prompt to guide your agent’s behaviour for a particular subject area. Think of it as a second layer on top of the Engine prompt, specifically for structured or thematic conversations.

• Example: “When the user asks about billing, guide them through our refund policy step by step before offering to escalate.”

⭐ FAQs Rating

When a topic or FAQ conversation ends, the user can be prompted to rate the quality of the response. Configure this here:

• Message: The text shown to the user at the end (e.g. “Was this answer helpful?”).
• Button #1 / Button #2: Quick-reply options (e.g. 👍 Yes / 👎 No).

💡 Pro Tip: Use clear, friendly rating messages. Short feedback loops improve your agent over time!

---

4. Fields with Instructions: Smart Data 💡

Need the bot to know opening hours, address, or stock status without retraining the whole brain? Use Fields. They’re like sticky notes the bot reads before replying.

⚙️ How it works (Add field)

1. Name: The internal name (e.g. shop_hours). Use snake_case.
2. Type: What kind of data is it? Options include String, Number, Boolean, List.
3. Instructions: The magic. Tell the bot how and when to use this data.
   • Bad: “Schedule”.
   • Good: “If the user asks when we open, reply with this exact value.”
4. Value return: The actual response the bot will output.

🚀 Ready-to-use Examples

• 📅 Fixed Opening Hours

  • Instructions: “If asked about opening times, use this value with the text ‘Our opening hours are…’.”
  • Value: 09:00 - 18:00 Monday to Friday

• 📍 Address

  • Instructions: “Provide this full address when asked for location.”
  • Value: 123 Fake Street, Springfield

• 📦 Do we ship? (Boolean)

  • Instructions: “If true, confirm we ship; if false, say collection only.”
  • Value: true

---

5. Events: Superpowers and Actions ⚡

Events are the bridge between chat and the real world. This is where the bot stops “talking” and starts “doing”.

🔗 What are they for?

• 👨‍💻 Hand-off: The classic transfer to a human.
• 🎟️ Create Tickets: Report bugs to Jira or Zendesk automatically.
• 📡 Webhooks: Send data to your CRM or API.

🛠️ Configure your first event

Click + Add event and fill in:

1. Name: Unique identifier (e.g. intent-agent-step-1, create_ticket).
2. Description: CRUCIAL! Explain when it triggers. The bot reads this to decide when to fire it.
   • Example: “Trigger this event when the user says ‘I want to speak to a person’ or appears angry.”

🌟 Most-used event: Transfer to Human

Copy this configuration if you want your bot to know when to step aside:

• Name: intent-agent-step-1
• Description: “Activate this immediately if the user asks to speak to an advisor, human support, or a real person.”
• Action (in your backend): Configure the webhook to assign the chat to a live agent.

---

6. Tools & Escalation: Handover Settings 🤝

This tab controls how your agent transfers conversations to human agents and how those handoffs are managed.

🔧 Tools

• Agent Handover Flow Configuration: When enabled, this lets you define exactly how the AI agent will transfer a conversation to a human agent. Configure the handover logic, triggers, and flow steps here.

👤 Settings Conversation Human

These settings control the escalation behaviour once a conversation is routed to a human:

Option	What it does
Automatic conversation assignment	Conversations are automatically assigned to available human agents who meet the configured requirements.
Email activate	Sends an email notification to the agent when a conversation is escalated.
Language required	Requires the assigned human agent to match the conversation’s language.

⭐ Agent Rating

After a human-agent conversation ends, the user receives a message to rate the service. Configure it here:

• Message: What the user sees (e.g. “How would you rate our support today?”).
• + Add Answer Button: Add quick-reply rating options (e.g. ⭐ Excellent, 👍 Good, 👎 Could be better).

💡 Pro Tip: Combine Automatic assignment + Language required for a smart routing experience where the right agent always gets the right conversation.

---

7. Voice: Give Your Agent a Voice 🎙️

This tab configures everything needed for your agent to handle voice interactions — phone calls, voice bots, and audio channels.

🔛 AI Voice Configuration

Toggle this on to activate voice capabilities for the agent. Once enabled, a full set of voice parameters becomes available.

🌐 Language & Model

Setting	Description
Agent Language	The primary language the agent will speak (e.g. Spanish).
Language Presets	Additional languages the agent can automatically switch to mid-conversation.
LLM	The language model powering the voice agent (e.g. gpt-4.1-mini).
Limit Token Usage	Maximum tokens per response. Set to -1 for unlimited.

🔊 Audio Settings

Setting	Description
Voice	The text-to-speech voice persona (e.g. Sarah – Mature, Reassuring, Confident).
TTS Output Format	Audio format for the agent’s spoken output (e.g. pcm_16000).
User Input Audio Format	Audio format expected from the user’s microphone input (e.g. pcm_16000).

⏱️ Timing & Timeouts

Setting	Description
Turn Timeout	Seconds to wait before ending the user’s turn if no speech is detected (default: 7).
Silence End Call Timeout	Seconds of silence before the call is automatically ended (default: 20).
Max Conversation Duration	Maximum length of the call in seconds (default: 600 = 10 minutes).

🔑 Keywords

Add comma-separated keywords that the voice agent should listen for or treat as special triggers (e.g. help, cancel, repeat).

💡 Pro Tip: Keep Turn Timeout short (5–8s) for snappy voice UX. Increase Silence End Call Timeout if your users tend to pause while thinking.

---

8. Chat Settings: Session & Behaviour ⚙️

This tab controls the chat-level behaviour of your agent — how long sessions last, what happens when users go quiet, and more.

📊 NPS Config

When enabled, a Net Promoter Score (NPS) survey is sent to the user at the end of a conversation. Great for tracking overall satisfaction trends.

💬 Limit History

Sets the maximum number of messages the agent keeps in its conversation memory at any given time (default: 3). Lowering this reduces token usage; increasing it gives the agent more context to work with.

⚠️ Heads up: A very low limit (e.g. 2) may cause the agent to “forget” context from earlier in the same conversation.

⏳ Session Timeout

Defines how long the agent can be inactive before the session is considered closed.

• IA Agent Downtime (Minutes): The idle time in minutes before the AI agent session ends (default: 1440 = 24 hours, minimum: 10 minutes).

🔕 Inactivity Rules

Option	What it does
User inactivity	Sends an automated message to the user if they don’t respond within a set time.
Human agent inactivity	Sends an automated message to the user if the human agent doesn’t respond within a set time.

💡 Pro Tip: Enable User inactivity with a friendly nudge like “Are you still there? 😊” to keep conversations alive and reduce abandoned chats.

---

🎓 Best Practices Summary (Cheat Sheet)

To make your agent a 10/10, remember:

• 🧪 Playground is your friend: Never publish without testing first. Pretend to be a difficult customer.
• 🚦 Clarity above all: In Prompts and Instructions, be direct. The bot does not understand hints.
• 🔁 Keep things up to date: If your opening hours change, update the Field. If policy changes, update the Guardrails.
• 🤝 Smooth escalation: Always provide an exit to speak to a human. Nothing is more frustrating than a bot stuck in a loop.
• 🎙️ Test voice separately: Voice behaviour differs from chat — always test audio flows independently before going live.

---

🆘 Quick Troubleshooting

Problem	Likely Fix 🔧
Replies with outdated information	Check Search with Sources or update your Fields.
Images don’t show	Your channel (e.g. WhatsApp) may not support that Rich Content.
The bot hallucinates	Tighten the Guardrails! Forbid invention.
Doesn’t transfer to a human	Review the event description. Is it clear?
Voice agent cuts off too early	Increase Turn Timeout or Silence End Call Timeout in the Voice tab.
Agent forgets context	Increase Limit History in Chat Settings.
Human agents not getting assigned	Check Automatic Conversation Assignment in Tools & Escalation.

You’re all set to build the best possible agent! Happy configuring! 🚀