Agents management
Learn how to create, configure and manage your virtual assistants
An agent is like a virtual employee that works for you. It can chat with your customers, answer questions, give recommendations and perform specific tasks that you teach it.
💡 Think of it like this:
If you had to hire someone to handle your support chat 24/7, you would need to train them, give them manuals and teach them about your products. An agent works the same way, but it's digital, never gets tired and can assist many people at the same time.
Step by step:
Go to the Agents page
In the left sidebar, click "Agents".
Menu → Agents → "Create Agent" buttonGive it a descriptive name
Choose a name that clearly indicates what the agent does.
✅ Good examples:
- • "Sales Assistant"
- • "Technical Support"
- • "Appointment Scheduler"
- • "Product Consultant"
Write a clear description
Explain which tasks this agent will perform.
Example:
"This agent helps customers find the perfect product for their needs, answers questions about features, prices and availability, and can guide them through the purchase process."
Define its personality
In the "Core Model" section, write how you want it to behave.
Personality example:
"You are a friendly and professional assistant. You speak clearly and simply, always with a positive tone. When you don't know something, admit it honestly and offer to contact a human. Use emojis occasionally to feel more approachable 😊"
Main settings:
🧠 AI model
The "brain" your agent will use. Different models have different capabilities:
Basic
Optimized for simple tasks and low cost. Ideal for tests and simple flows.
Standard
A balance between performance and cost. Recommended for most use cases.
Advanced
Maximum capacity and answer quality. Use it when you need the best possible reasoning.
🎨 Temperature (Creativity)
Controls how "creative" or "predictable" the agent is:
0.0 - 0.3
Very predictable. It always gives similar answers. Ideal for precise technical information.
0.4 - 0.7
Balanced. Answers vary a bit but remain coherent. Recommended.
0.8 - 1.0
Very creative. Answers are varied and original, but can be unpredictable.
📏 Maximum answer length
Controls how long answers can be. A low value (100–300) generates short, direct responses. A high value (500–1000) allows more detailed, explanatory answers.
💰 Note: Longer answers consume more tokens and cost more.
Before publishing your agent, it's important to test it to make sure it responds correctly:
Click the "Chat" button
You'll find it on your agent card, next to "Configure".
Ask typical customer questions
Try with the questions your customers ask most often.
Evaluate the quality of the answers
Are they accurate? Is the tone appropriate? Is any information missing?
Adjust as needed
Go back to the configuration and improve the personality or add more information.
💡 Pro tip:
Ask someone on your team to test the agent as well. Other people may spot issues you don’t see because you know your business too well.
Activate / Pause
When an agent is "Active" (green badge), it can receive and answer messages. If you set it to "Paused" (grey badge), it will temporarily stop working without losing its configuration.
Configure
Access all agent options: change its personality, add knowledge, configure connections with other agents, view usage metrics and generate API keys.
Open chat
Open a chat window to talk directly with the agent and test its responses in real time.
Delete
Permanently delete the agent. ⚠️ This action cannot be undone. You will lose all configuration, conversation history and metrics.
- ✓Be specific about the personality: The clearer you are about how it should speak and behave, the better its answers will be.
- ✓Test frequently: Whenever you make changes, test the agent again with different questions.
- ✓Give it clear boundaries: Tell it which topics it can handle and which ones it should escalate to a human.
- ✓Start simple: It's better to have an agent that does one thing very well than one that tries to do many things poorly.