In this article, you'll learn how to create and configure your AI agent in Callbell — from the initial setup to adding it to your bot flow to start handling real conversations.
| Note: Once your AI agent is ready, the next step is to add it as a bot action in your bot flow. This is what makes it active and accessible to your customers. |
How to create a new AI agent
- Go to the AI agents tab and click New.
- Fill in the configuration fields described below.
- Test your agent using the simulator before adding it to your bot flow.
Configuration fields
1. Name and avatar
Choose a name and avatar to identify the agent internally. These are only visible to your team inside Callbell — your customers will never see them.
2. Agent goal
This is one of the most important fields. The goal tells the agent exactly what it needs to accomplish before ending the conversation. Be as specific as possible and break it down into clear steps — the more precise the goal, the more predictable the agent's behavior.
Example:
- Collect the client's name and email
- Ask what's the client's need
- Recommend one product and share pricing
- Confirm the customer wants to buy before finishing the flow
3. Agent behavior
This is where most of your agent's instructions live. Use this field to define how the agent should handle the conversation from start to finish.
You can include things like:
- What the first message should say
- The sequence of questions or requests the agent should follow
- How to handle specific scenarios or objections
- What qualifies as a good lead
- When to share specific product information, links, or pricing
The more detailed and structured this section is, the better your agent will perform.
4. Agent output (internal)
This field defines what the agent returns once its goal is completed. The output is not sent to the customer — it's used internally and can be passed as a variable in your bot flow.
Examples of what to use it for:
- A summary of the conversation
- Collected contact information organized in topics
- A lead quality score based on specific criteria
5. Agent tone
Briefly describe the personality and communication style of the agent. Examples: friendly, formal, professional, enthusiastic, concise.
6. Forbidden rules
List things the agent must never say. This helps prevent unwanted behavior and keeps the agent within safe boundaries.
Examples:
- Make up facts or information it doesn't know
- Mention competitors
- Discuss pricing
- Talk about topics unrelated to its training
7. Q&A
Add specific question-and-answer pairs to build the agent's knowledge base. This is the most reliable way to ensure the agent responds accurately to common questions.
Examples of what to include:
- Business hours
- Product pricing and availability
- Frequently asked questions from customers
- Return or refund policies
Testing your agent
Before activating your agent in a bot flow, use the simulator to test how it behaves in real conversations. The simulator shows the cost of each message generated by the agent, so you can also use it to estimate your AI credit consumption.
| Tip: Run through different scenarios — including edge cases — to make sure your agent handles them the way you expect. |
You will see the cost generated for each message and the total cost of the chat. Learn more about AI agent cost here.
Next step: adding the agent to your bot flow
Once you're happy with how your agent performs in the simulator, add it with the action "AI Agent flow" in your bot flow. This is what connects the agent to your actual conversations and makes it live for your customers.
FAQ
Can I attach files to the agent's training?
Not yet. For now, we recommend adding all resources directly in the Q&A section or in the Agent behavior field.
Can the agent understand attachments sent by customers?
Yes! The agent can receive and understand images and files sent by customers during a conversation.
Can the agent listen to audio messages?
Not yet. Support for audio messages is planned for a future update.
Can the agent send media (images, files, videos)?
No. The agent can only send text messages.
Can the agent assign conversations to agents or teams, or add tags?
No. The agent cannot perform actions inside Callbell beyond sending messages. All actions — such as assignment, tagging, or closing chats — should be configured in the bot flow after the agent finishes its goal.
Can the agent open or access links?
No. The AI agent does not have access to the internet, so it cannot open links — either during training or in live conversations. Make sure all relevant information is included directly in the Q&A or Agent behavior fields.