Agent Configuration

After creating an agent, you can adjust its behavior, connect external resources, and configure advanced capabilities.
This guide describes all available options on the Agent Configuration screen, accessible when editing an existing agent.

---

​

Accessing Agent Configuration

On the Agents screen, you can view all agents created in the workspace. Main elements:

• Agents List: cards containing agent name, type (LLM), model, and status
• Actions menu (⋯ Actions): appears when hovering over the agent card
  • Edit: opens the configuration screen for the selected agent
  • Delete: permanently removes the agent

Step by step:

1. Locate the desired agent
2. Click ⋯ Actions
3. Click Edit to configure

---

​

Profile – Agent Basic Information

The Profile section is responsible for defining who the agent is and how it behaves. Available fields:

• Agent name: Internal name of the agent (e.g., sales_assistant)
• Description: Brief description of the agent’s purpose
• Role: Agent’s role in the conversation (e.g., Customer support assistant)
• Goal: Agent’s main objective (e.g., Resolve customer questions)
• Behavior: Defines the agent’s tone and personality (e.g., You are a sales specialist assistant)

Buttons:

• Review Prompt: allows reviewing the final generated prompt
• Generate with AI: automatically generates the behavior
• Test your agent: opens a test chat
• Save: saves all changes

---

​

Sub Agents – Agent Composition

The Sub Agents section allows an agent to execute other agents as part of the flow. Components:

• No sub-agent selected: indicates that no subagent has been added
• Available Agents: list of available agents
• Search Agents: search field
• Add: adds the selected agent as a subagent

Step by step:

1. Click Sub Agents in the sidebar menu
2. Search for the desired agent
3. Click Add
4. Save changes

---

​

Knowledge – Knowledge Base and Memory

The Knowledge section allows the agent to access information stored in knowledge bases and use memory from previous conversations. This functionality is essential for the agent to provide contextualized and personalized responses.

​

Load Knowledge – Loading Knowledge Base

The Load Knowledge functionality allows the agent to access documents, files, and information previously loaded into a vector knowledge base. Available settings:

• Toggle Load Knowledge: Activates or deactivates the use of the knowledge base by the agent
• Knowledge Base: Selects which knowledge base will be used (e.g., evoai via Pinecone)
• Preload Knowledge: Optimization option that loads knowledge before the query
• Maximum Results: Defines the maximum number of results returned from the knowledge base
• Knowledge Tags: Allows filtering knowledge by specific tags (e.g., Sales, Support, Product)

How it works: When activated, the agent automatically searches for relevant information in the knowledge base before responding to the user. This allows it to access technical documents, manuals, company policies, and other relevant content without needing to have this information directly in its instructions.

​

Load Memory – Managing Conversation Memory

The Load Memory functionality allows the agent to maintain context from previous conversations, making interactions more natural and contextualized. Available settings:

• Toggle Load Memory: Activates or deactivates memory use by the agent
• Memory Base: Selects where memory will be stored (e.g., evoai via Pinecone)
• Maximum Messages in Short-Term Memory: Defines how many recent messages the agent keeps in short-term memory
• Compression Interval for Medium-Term Memory: Defines after how many messages compression to medium-term memory occurs

How it works: The memory system works in levels:

• Short-Term Memory: Keeps the most recent messages complete for immediate context
• Medium-Term Memory: Compresses old messages into summaries, keeping important information without taking up much space
• Long-Term Memory: Stores relevant information that can be retrieved in future conversations

Memory cleanup:

• Clear All Memory: Removes all stored memory from the agent. Use with caution, as this action is irreversible.

---

​

Tools – Agent Tools

The Tools section defines which tools the agent can use during conversations.

​

Agent Tools

Allows selecting other agents as tools that can be called during execution. How to add:

1. Click Add in the Agent Tools section
2. Select the desired agents from the list

Options:

• List of available agents with checkbox for selection
• Confirm: confirms the agent selection
• Cancel: closes without saving

​

Custom Tools

Allows creating custom HTTP tools to integrate with external APIs. Options:

• Add Tool: Creates custom HTTP tools
• Create: Starts the process of creating a new tool

​

Native Tools (starting in v1.0.0-rc3)

Native tools embedded in the processor that become available without configuring custom HTTP. They appear automatically when the prerequisite is satisfied: Pipeline manipulation and label management usage hints are automatically injected into the system prompt when these tools are active, guiding the LLM on when to call them.

​

”Allow Managing Labels” Toggle

Starting in v1.0.0-rc3, the agent configuration exposes the Allow managing labels toggle (allow_manage_labels). When active:

• The agent receives the manage_conversation_labels tool.
• The agent can add/remove labels during the conversation.

When inactive, the tool is not exposed — the agent has no way to tag conversations.

---

​

Products – Linked Catalog (starting in v1.0.0-rc3)

In the agent’s Products tab, select which catalog items the agent should know about. Linked products are injected into the agent’s context at runtime — changes in the catalog reflect in the next conversation without reconfiguring the prompt. Only products with active status are available for linking. See Products Catalog.

---

​

Integrations – External Integrations

The Integrations section allows connecting the agent to external services. Available integrations:

• ElevenLabs: Allows the agent to generate audio responses using voice synthesis. Button: ACTIVATE
• Google Calendar: Allows creating, viewing, and managing calendar events. Button: ACTIVATE
• Google Sheets: Allows creating, reading, and updating Google spreadsheets. Button: ACTIVATE

Integration Configuration: When activating an integration, you will need to provide:

• API Key: Authentication key for the external service
• Apply Settings: Saves the integration configuration

---

​

MCP Servers – MCP Servers

The MCP Servers section allows connecting MCP (Model Context Protocol) servers, which expand the agent’s capabilities. Available MCP integrations: GitHub, Notion, Stripe, HubSpot, Linear, Monday.com, PayPal, Canva.
Current status: All displayed as Coming soon. Custom MCPs:

• Add Custom MCP: Adds a previously created MCP server
• Add MCP Server: Creates a new custom MCP server

Select Custom MCPs:

• Search field to filter MCPs
• No MCP server configured message when no MCPs are configured
• Add MCP Server button to add a new server

---

​

Configuration – Advanced Agent Settings

The Configuration section is where you define the agent’s technical and behavioral adjustments. This section is divided into four main tabs: General, System, Message Handling, and Inactivity Actions.

---

​

General Tab – General Settings

The General tab contains the fundamental settings that define how the agent processes and responds to requests.

​

Model and API – AI Model and API Key

This section defines which artificial intelligence model will be used by the agent and which authentication key will be used to access this model. Available fields:

• API Key: Select the API key that the agent will use to connect to the AI provider. The selected key determines which models will be available for selection
• Language Model: Choose the specific language model (examples: GPT-4.1 Nano, GPT-4, Claude 3.5 Sonnet, etc.). Each model has different characteristics of speed, cost, and capability
• Manage: Button that opens the API key manager, allowing you to add, edit, or remove keys

Why this is important: The model choice directly affects response quality, processing time, and agent operating costs. More advanced models tend to provide more accurate responses but may have higher costs.

---

​

Agent Capabilities – Advanced Capabilities

This section activates special functionalities that expand the agent’s capabilities. Available options:

• Planner (toggle): When activated, the agent executes a planning process before responding. This means it first analyzes the request, creates an action plan, and then executes that plan step by step. This capability significantly improves response quality in complex tasks that require multiple steps or structured reasoning.

When to use Planner:

• Tasks that require multiple steps
• Complex analyses that need structuring
• Situations where it’s important for the agent to “think” before acting

---

​

Output Format – Structured Output Format

Output Format allows defining a structured schema for agent responses, ensuring it always returns data in a consistent and predictable format. Basic settings:

• Output Key: Unique identifier that will be used to reference this agent’s output. Useful when you have multiple agents and need to identify which one is returning the data
• Output Schema: Defines the data structure that the agent must return

Structured Schema – Creating the Output Schema: The structured schema allows defining exactly which fields the agent should return and in what format. How to configure:

1. Click First Field or Add Field to add a new field to the schema
2. For each field, define:
   • Field Name: Field name (e.g., “customer_name”, “total_value”, “status”)
   • Type: Field data type:
     • String: Text (e.g., “John Silva”, “Approved”)
     • Number: Number (e.g., 150.50, 42)
     • Boolean: True or false (e.g., true, false)
     • Array: List of values (e.g., [“item1”, “item2”, “item3”])
     • Object: Complex structure with multiple fields
   • Description: Describe the field’s purpose so the agent understands what to include there
3. Use the Edit, Remove, or Save buttons to manage created fields

Usage example: If you’re creating a sales agent, you can define a schema like:

• customer_name (String): Customer’s full name
• email (String): Contact email
• quote_value (Number): Total quote value
• products_interest (Array): List of products the customer showed interest in
• negotiation_status (String): Current negotiation status

Why use Output Format:

• System integration: Facilitates agent integration with CRMs, ERPs, and other systems
• Consistency: Ensures data always comes in the same format
• Automation: Allows automating processes based on structured agent responses
• Data analysis: Facilitates analysis and reports on interactions

---

​

System Tab – Behavior and Permissions

The System tab controls the agent’s permissions and behavior within the system during conversations.

​

Conversation Behavior – Conversation Behavior

Defines what the agent can or cannot do during interactions with users. Available permissions:

• Transfer to human: When activated, allows the agent to identify situations where it’s necessary to transfer the conversation to a human agent. The agent can do this automatically when it detects customer frustration, complex requests it cannot resolve, or when the user explicitly asks to speak with a person.
• Allow registering reminders: Allows the agent to create reminders and appointments. For example, if a customer asks to be reminded about a promotion next month, the agent can register this reminder automatically.
• Allow editing contacts: Grants the agent permission to modify contact information, such as updating phone numbers, emails, addresses, or other registration information when the user provides new data.
• Allow pipeline manipulation: Allows the agent to move contacts between different stages of the sales or service pipeline. For example, moving a lead from “First Contact” to “Negotiation” or from “Qualification” to “Proposal Sent”.

Security tip: Only activate permissions that the agent really needs. If the agent is only informative, it’s not necessary to allow contact editing or pipeline manipulation.

​

Agent Timezone – Agent Timezone

Defines the timezone that the agent will use for all date and time calculations. Configuration:

• Select the appropriate timezone for your operation
• Example: America/Sao_Paulo (GMT-3) for Brasília time

Why this is important: The correct timezone ensures that:

• Reminders are sent at the right time
• Reports show times correctly
• Appointments are made considering the business’s local time

---

​

Message Handling Tab – Message Processing

The Message Handling tab defines how the agent processes, formats, and sends its messages.

​

Processing and Sending Settings

Message Wait Time (seconds) – Wait Time: Defines how many seconds the agent should wait before processing and responding to a message.

• When to use: Useful when users usually send multiple consecutive messages. Instead of responding to each message individually, the agent waits for the configured time and responds considering all messages sent in the period.
• Example: If configured for 3 seconds and the user sends “I want”, “to buy”, “a notebook”, the agent waits 3 seconds and processes the three messages together, understanding the complete request.

Message Signature – Message Signature: Text that will be automatically added to the end of each message sent by the agent.

• Common use: Add contact information, service hours, or disclaimers
• Example: “This is an automatic message. To speak with an agent, type ‘speak with human’.”

Enable text segmentation – Text Segmentation: When activated, the agent divides long messages into multiple smaller messages.

• Advantage: Improves readability, especially on mobile devices
• Behavior: A long response is divided into separate paragraphs or topics

Use Emojis in Responses – Using Emojis: Allows the agent to include emojis in its responses to make them more friendly and expressive.

• When to activate: Informal channels like WhatsApp, Instagram
• When to deactivate: Formal corporate communications, professional emails

Send message as reply in conversation – Reply to Specific Messages: When activated, the agent responds directly to specific messages using the “reply” or “quote” feature available on most messaging platforms.

• Advantage: Keeps context clear, especially in conversations with multiple questions
• Example: If the user asks three different questions, the agent can answer each one by quoting the specific question

---

​

Inactivity Actions Tab – Inactivity Actions

The Inactivity Actions tab allows configuring automatic actions that the agent will execute when a user becomes inactive (doesn’t respond) for a certain period.

​

How It Works

This functionality is essential for maintaining user engagement and preventing important conversations from being forgotten. Screen structure:

• View of all configured inactivity rules
• + Add previous action button to create new rules

​

Creating an Inactivity Action

Step by step:

1. Click + Add previous action
2. Configure the following fields:

If not respond in – If not responding in: Defines the inactivity time that will trigger the action.

• Options: Minutes, hours, or days
• Example: 2 minutes, 1 hour, 1 day, 3 days

the agent should – the agent should: Defines which action will be executed after the inactivity period.

• Interact with client: Send a re-engagement message
• Send reminder: Send a reminder about something pending
• Close conversation: Automatically close the conversation
• Transfer to human: Transfer to human agent
• Move in pipeline: Move to another funnel stage

What should the agent say? – What should the agent say: Free text field where you write the message that will be sent. Message examples:

• For re-engagement: “Hello! I saw you were interested in our products. Can I help with anything else?”
• For reminders: “Hi! Remember that our promotion ends tomorrow. Would you like to continue with your purchase?”
• For qualification: “I noticed you didn’t respond. Are you still interested in learning about our solution?”

3. Click Save to save the rule

​

Creating Multiple Cascading Rules

You can create multiple inactivity rules with different times to create a re-engagement sequence. Strategy example:

1. 2 minutes of inactivity: “Are you still there? Can I clarify any questions?”
2. 1 hour of inactivity: “Hi! I saw we talked earlier. If you need help, I’m here!”
3. 1 day of inactivity: “Hello! I noticed you showed interest in [product]. Would you like to resume our conversation?”
4. 3 days of inactivity: “This is my last contact attempt. If you wish, we can talk in the future. See you soon!”

​

Best Practices

• Don’t be invasive: Avoid sending many messages in short periods
• Be relevant: Reference the context of the previous conversation
• Offer value: Each message should offer something useful to the user
• Respect limits: If the user doesn’t respond after several attempts, stop sending messages

---

​

Final Considerations

The Agent Configuration screen is the complete control center for customizing behavior, connecting external resources, and defining how the agent operates in different scenarios. Important recommendations:

• Always click Save in the upper right corner after making changes
• Use Test your agent to validate behavior before putting the agent into production
• Configure the knowledge base and memory for more contextualized conversations
• Adjust system permissions according to the desired level of autonomy for the agent
• Define inactivity actions to maintain user engagement

After configuring all sections, the agent will be ready for use in complex flows, integrations, and advanced automations.