AI Agents is a generative AI platform that allows any Zenvia customer to create and manage their own AI agents, natively integrated into the Zenvia Customer Cloud ecosystem. An agent does more than just answer questions: it performs tasks and takes actions, retrieves information from the CRM, updates data in the ERP, creates support tickets, qualifies leads, and resolves customer requests. All of this is done according to the data, permissions, and business rules configured by the customer.
We are moving beyond chatbots and into the era of AI agents, with less friction, less manual work, zero coding, and native context from Zenvia Customer Cloud.
Prerequisites
Before getting started, make sure you have:
Access to Zenvia Customer Cloud with the AI Agents platform enabled.
A configured customer service channel, such as a WhatsApp account.
The documents that will be used as knowledge sources, in PDF, DOCX, or text format.
The API documentation for the systems you want to integrate, if you intend to use external tools.
1. Understanding the main AI Agents platform concepts
Before configuring your first agent, it is important to understand some of the concepts used on the platform. These concepts help you understand how agents are configured, how conversations are processed, and how changes are applied.
1.1. How does the customer service flow work?
When a customer sends a message through an integrated channel, it goes through several steps before reaching the responsible agent.
In simplified terms, the flow works as follows:
The customer sends a message through the customer service channel.
The message is received by the channel gateway.
The platform checks the configured routing rules.
The corresponding agent is identified.
The agent processes the message based on its instructions, tools, and knowledge sources.
The response follows the same path back to the customer.
⚠️ Attention: When a conversation already has an active session, the platform keeps the agent and version associated with the beginning of that session. Therefore, a newly published version will only be used for new conversations and will not automatically change interactions that are already in progress.
2. How to create and configure an agent
Creating an agent involves defining its identity, behavior, the artificial intelligence model it will use, and how long it should retain the context of a conversation.
Go to Zenvia Customer Cloud > AI Agents.
Click Create agent +.
Enter a Name and the Session expiration time due to inactivity.
Example name: Hotel Concierge.
In minutes: 2000.
Click Save.
After clicking Save, the platform will create the agent's first version. If no version has been created for this agent yet, click Create first version.
How to create a new agent version
Creating a new version or the first version allows you to configure or update an AI agent's characteristics, defining its identity, behavior, knowledge sources, tools, artificial intelligence model, and protection mechanisms.
Select the agent you want to configure.
In the Versions section, click New version +.
The platform will display the configuration screen, divided into sections to make agent customization easier.
1. Configure the agent's identity
In the Identity section, you will find the fields used to identify the agent.
Name: Enter the agent's name. This field identifies the version and the agent it belongs to.
Description: Enter a description explaining the agent's purpose, main responsibilities, or the type of customer service it provides.
💡Tip: Use a clear and concise description to make it easier to identify the agent's purpose.
2. Configure the agent's behavior
In the Behavior section, you will find the System prompt field. The System prompt contains the instructions that guide how the agent should act and respond during conversations. This is where you define its role, personality, tone of voice, customer service rules, and limitations.
System prompt
Use the text editor to write the agent's instructions.
The editor provides formatting features to help organize the content, such as:
Bold and italic formatting.
Headings and subheadings.
Links.
Code formatting.
Quotations.
Numbered and bulleted lists.
Other text organization features.
What should you include in the System prompt?
To guide the agent's behavior, we recommend including:
Agent role: Explain who the agent is and what its function is.
Objective: Describe what it should accomplish during customer interactions.
Tone of voice: Define how it should communicate with customers.
Customer service rules: Establish the steps it should follow during conversations.
Restrictions: Specify which actions or responses are not allowed.
Handling questions: Explain what the agent should do when it cannot find information or resolve a request.
Example instruction: "You are a customer service agent for the company. Your role is to help customers with questions about orders and products. Use polite, clear, and concise language. Always check the knowledge base before providing specific information. If you cannot find the answer or resolve the request, transfer the conversation to a human agent."
💡Tip: The clearer and more organized the instructions are, the more accurately the agent will follow the expected behavior.
3. Add a knowledge base
In the Knowledge section, you can associate knowledge bases with the agent. These knowledge bases provide company documents and information that the agent can consult during conversations.
To add a knowledge base:
Click Add KB+.
Select the knowledge base you want to associate with the agent.
Confirm your selection.
You can use knowledge bases containing information such as:
Product and service catalogs.
Frequently asked questions.
Customer service policies.
Internal manuals and procedures.
Company information.
⚠️ Attention: Before associating a knowledge base, make sure its documents have been processed and are available for consultation.
For more information, see the article How to create and configure a knowledge base.
4. Add tools to the agent
In the Tools section, you can associate tools that allow the agent to retrieve information or perform actions in external systems.
To add a tool:
Click Add tool.
Select the tool you want to make available to the agent.
Confirm your selection.
For example, you can add a tool to:
Check an order status.
Retrieve customer information.
Query data from an external system.
Perform operations through an API.
Tools expand the agent's capabilities, allowing it to retrieve information and perform actions beyond simply responding to messages.
⚠️ Attention: The tool must already be registered on the platform before it can be associated with the agent.
For more information, see the article How to register and configure tools.
5. Select the artificial intelligence model
In the Model section, choose the provider and artificial intelligence model that the agent will use.
The platform offers options from different providers:
Amazon Bedrock
Anthropic
Google
OpenAI
After selecting a provider, choose the corresponding available model.
Example: Google – Gemini 1.5 Flash.
Your model selection should take into account the agent's requirements and the complexity of the tasks it will perform.
In general, simpler models may be suitable for straightforward tasks, while more advanced models may be better suited to activities that require greater interpretation and reasoning capabilities.
💡Tip: Evaluate the agent's expected behavior and the characteristics of the available models before making your selection.
6. Configure Guardrails
Guardrails are protection mechanisms that help identify and control content or behaviors that may pose risks during agent interactions. On the version creation screen, you will find four protection options:
Prompt injection detector
Detects attempts to manipulate the agent through prompt injection or jailbreak attacks.
This mechanism helps protect the agent's original instructions against messages designed to change its behavior.
System prompt leakage detector
Prevents the agent from revealing its System prompt or internal instructions.
This mechanism helps protect the guidelines used to configure the agent.
Harmful content detector
Flags and blocks harmful, unsafe, or policy-violating content.
Its purpose is to reduce exposure to inappropriate content during interactions.
Personally identifiable information (PII) detector
Detects and masks personally identifiable information (PII) in inputs and outputs.
This mechanism helps protect personal information that may appear in incoming messages or agent responses.
⚠️ Attention: Evaluate which protection mechanisms are necessary for your agent's intended use case.
7. Configure context enrichment
In the Context Enrichment section, you can select additional information that will automatically be made available to the agent during conversations. This feature allows you to provide contextual data without requiring the agent to ask the customer for it directly.
The platform offers two options:
Visitor
Provides visitor-related information to give the interaction additional context.
Current date and time
Provides the current date and time so the agent can take this information into account during customer interactions.
After completing all sections, review the configured information and click Create version. Wait for the process to finish. The platform will create a new version with the configurations you have specified.
⚠️ Attention: Creating a version does not necessarily mean that it is already in production. To make it available for customer interactions, you must associate it with the tag used in the routing rules, according to your organization's publishing strategy.