Native Tools

Native tools are predefined tools integrated into the platform that allow the AI model to perform specific actions during a conversation. These tools extend the capabilities of the AI agent beyond text generation, allowing it to interact with the system, query information, manipulate data, and execute concrete actions. Native tools are found and configured in the AI Agent node. To access them:

Select the AI Agent node in your flow
Open the TOOLS section in the node configuration
Click the ”+ Add tools” button
In the “Select a Tool” field, you will see the list of all available tools, including native ones (marked with the text “(Native)”)

Native tools configuration in the AI Agent node

Below are descriptions of all native tools available in the AI Agent node:

Product search

Searches products in the available catalog to recommend to the user. Uses a search engine to find matches based on the product name or related terms. Parameters:

Name	Type	Required	Description
`product`	`string`	Yes	Search term or product name.
`top_k`	`number`	No	Maximum number of results to return.
`score_threshold`	`number`	No	Minimum similarity score (0.0 to 1.0) to consider a result relevant.
`client_reference_id`	`string`	No	Customer reference ID for personalized searches.

Prompt Example:

“If the user asks about the availability of a specific item or model, you MUST use the search_products tool to query the catalog before responding.”

Arguments Example:

{
  "product": "iPhone 15 Pro",
  "top_k": 5,
  "score_threshold": 0.85
}

Response: Returns a list of products found or a message indicating that no results were found. In case of error, returns a descriptive system message. Notes:

Requires the channel to have a product catalog configured and accessible.

Transfer to agent

Transfers the user’s current conversation to a human agent or an attention queue. Allows specifying routing strategies and priorities. Parameters:

Name	Type	Required	Description
`assignment_type`	`enum`	No (Default: `direct`)	Assignment method: `direct` (immediate) or `queue` (ticket queue).
`assignment_by`	`enum`	No (Default: `shuffle`)	Routing strategy: `shuffle` (auto), `team`, `operators`, `general`.
`teamId`	`number`	No*	Team ID. *Required only if `assignment_by` is ‘team’.
`operatorId`	`number`	No*	Operator ID. *Required only if `assignment_by` is ‘operators’.
`priority`	`number`	No (Default: 0)	Queue priority (0-10), only relevant for `assignment_type='queue'`.

Prompt Example:

“When the user expresses frustration or explicitly requests to speak with a real person, execute the transfer_to_agent function immediately.”

Arguments Example: Case: Automatic direct transfer (most common)

{
  "assignment_type": "direct",
  "assignment_by": "shuffle"
}

Case: Direct transfer to a team

{
  "assignment_type": "direct",
  "assignment_by": "team",
  "teamId": 4501
}

Case: Create ticket in general queue

{
  "assignment_type": "queue",
  "assignment_by": "general",
  "priority": 5
}

Response: Returns a JSON object with the result of the operation indicating whether the transfer was initiated successfully. Notes:

Can end the AI flow immediately.
When using assignment_type: "queue" with assignment_by: "general", a ticket is created in the general attention queue.
Important: Team or operator IDs must not be made up; only those explicitly configured should be used.

Interactive messages

Interactive messages brings together the button messages the agent can send: Buttons and list, Webview, and Call to Action. Each message has its own toggle, so you turn on only the ones the agent needs. The agent decides when to send each one based on the criteria you give it. The Interactive messages row appears in the Tools tab of the AI Agent node and shows how many messages are active (for example, 2/3). Click it to open the settings panel.

General settings: Buttons and list

Buttons and list turns on the native tool Buttons and list. With it, the agent chooses the format (buttons or list) and the content of the message in each conversation. You configure it once per agent. Once it is on, you can edit its Description: the text the model reads to decide when to use it.

Custom messages: Webview and Call to Action

Custom messages let you define in advance what the agent sends and in which situation.

Message	What it sends	What happens next
Webview	A button that opens a form inside the chat	The agent waits for the user to complete the form and continues the conversation with the response
Call to Action	A button with a link	The agent continues the conversation without waiting

Turn on the message

Turn on the Webview or Call to Action toggle and click the arrow on the row to open its settings.

Give it a name

Enter a Name that identifies the message, for example Book an appointment or Sign-up form. It can’t be the same as, or very similar to, the name of the other custom message.

Define when the agent should use it

In When the agent should use it, describe the customer intent that triggers the message. You have up to 1000 characters.

Fill in the message fields

For each field, choose how its value is obtained and fill in the required fields. The fields of each type are described below.

Each message field has three modes:

Mode	How the value is obtained
AI	The agent writes the value in each conversation. You can add an optional description to guide it; if you leave it blank, the AI decides what goes in the field.
Static	You use fixed text, the same in every conversation.
Variable	You use the value of a flow variable, for example `{{$memory.payment_link}}`.

If you turn on a message but it is missing information, the row shows Active but incomplete and the agent can’t send it yet. Hover over the badge to see what is missing: the Name, When the agent should use it, or an empty required field in Static or Variable mode. Fields in AI mode are never incomplete, because the agent fills them in when it sends the message.

Webview fields

Field	Required	Limit	Description
URL	Yes	2048 characters	Address of the form that opens inside the chat.
Title	No	120 characters	Message header.
Message	No	900 characters	Main text of the message.
Button text	No	80 characters	Text of the button that opens the form.
Message footer	No	60 characters	Supporting text below the message.
Input variable	No	1024 characters	Data sent to the form as URL parameters.

Call to Action fields

Field	Required	Limit	Description
Header	No	60 characters (text)	None, Text, Image, Video, or Document. For an image, video, or document, enter the variable that holds the public URL of the file.
Content	Yes	1024 characters	Main text of the message.
Footer	No	60 characters	Supporting text below the message.
Button name	Yes	18 characters	Button text.
URL	Yes	4096 characters	Link the button opens.

Supported channels

Custom messages are available on these channels:

Channel	Webview	Call to Action
WhatsApp	Yes	Yes
Slack	Yes	Yes
Teams	Yes	Yes
Facebook Messenger	Yes	No
Instagram	Yes	No

If the conversation’s channel doesn’t support a custom message, the agent doesn’t get it as an option.

On Facebook Messenger and Instagram, the Webview arrives as a Meta button message: the text and the button are shown, without a title or footer. On Instagram, that button only shows in the mobile app; the web version of Instagram doesn’t display it.

Write the When the agent should use it criteria as the customer’s intent, not as a technical instruction. For example: “The customer wants to book an appointment or asks about available times”.

While you have custom messages active, the native tool Send Call to Action is no longer offered to the agent. Configure the Webview and the Call to Action from Interactive messages.

Buttons and list

Without the Interactive messages catalog, this tool appears in the panel as Send interactive messages. Sends a structured or interactive message to the user. The exact format depends on the channel (WhatsApp, Web, etc.), but generally includes buttons or option lists. Parameters:

Name	Type	Required	Description
`text`	`string`	Yes	Main content of the message. Maximum 1024 characters.
`title`	`string`	No	Header/title of the message. Maximum 60 characters.
`caption`	`string`	No	Subtitle or complementary description. Maximum 60 characters. In WhatsApp with 4+ options, used as the button text that expands the list.
`options`	`array`	Yes	List of options/buttons. See structure below.

Structure of options (Array of objects):

title (string, required): Text of the option/button. Maximum 20 characters. Duplicate titles are not allowed.
description (string, optional): Additional description of the option. Maximum 72 characters.

Prompt Example:

“If the user asks what services we offer, you MUST use send_interactive_message to display a menu with the available options.”

Arguments Example (Buttons - 3 or fewer options):

{
  "text": "How would you like to continue with your query?",
  "title": "Service Options",
  "caption": "Select an option",
  "options": [
    {
      "title": "Talk to Agent",
      "description": "Transfer to human support"
    },
    {
      "title": "View Catalog",
      "description": "Browse available products"
    }
  ]
}

Arguments Example (WhatsApp List - 4+ options):

{
  "text": "We have several departments available to assist you.",
  "title": "Departments",
  "caption": "View options",
  "options": [
    { "title": "Sales", "description": "Product and pricing queries" },
    { "title": "Technical Support", "description": "Help with technical issues" },
    { "title": "Billing", "description": "Payment queries" },
    { "title": "Returns", "description": "Manage returns" }
  ]
}

In WhatsApp, when you send 4 or more options, the message is displayed as a dropdown list. The caption value is used as the button text that opens the list (in the example above, “View options”).

Response: Returns "Message sent successfully." if the sending was successful or an error message otherwise. Notes:

Sends a real message to the user.
The message type adapts automatically according to the channel and number of options.

Send Call to Action (CTA) (Legacy)

Replaced by Interactive messages. Sends a call-to-action button that opens an external URL. It can be configured as a WebView for forms or payments that require waiting for a user response. Parameters:

Name	Type	Required	Description
`text`	`string`	Yes	Main content of the message.
`url`	`string`	Yes	URL that opens when the button is clicked.
`display_text`	`string`	Yes	CTA button text. Maximum 20 characters.
`title`	`string`	No	Header/title of the message.
`caption`	`string`	No	Subtitle to complement the message.
`is_webview`	`boolean`	No	If `true`, sends as a WebView that pauses the flow and waits for a callback. Default: `false`.
`expiration_time`	`number`	No	Seconds to wait for the callback before expiring. Default: 300. Only applies when `is_webview=true`.
`pending_message`	`string`	No	Message to show if the user sends text while waiting for the WebView callback.
`response_variable`	`string`	No	Name of the variable where the callback response will be saved.
`input`	`string` \| `object`	No	Data to pass to the WebView as URL parameters.

Prompt Example:

“When the user confirms they want to make the payment, use send_call_to_action with is_webview=true to send the payment link and wait for confirmation.”

Arguments Example (Simple CTA):

{
  "text": "Visit our online store to see all available products.",
  "url": "https://store.example.com/catalog",
  "display_text": "View Catalog",
  "title": "Online Store"
}

Arguments Example (WebView waiting for response):

{
  "text": "Click the button to complete your payment securely.",
  "url": "https://payments.example.com/checkout",
  "display_text": "Pay Now",
  "title": "Complete Payment",
  "is_webview": true,
  "expiration_time": 600,
  "pending_message": "Please complete the payment in the opened window before continuing.",
  "response_variable": "payment_result",
  "input": {
    "order_id": "12345",
    "amount": 99.99
  }
}

Response:

Normal CTA: "Call-to-action message sent successfully."
WebView: "WebView message sent. Waiting for user action."

Notes:

Available only for WhatsApp providers that support this type of message.
When is_webview=true, the flow pauses until the WebView returns a callback or the time expires.
The input value is converted to URL query string parameters.
The callback response is saved in the variable specified in response_variable.

Reply with voice notes

This tool converts the agent’s replies into voice notes instead of text. It is enabled like any other native tool from the Tools tab.

It is only available on WhatsApp, Instagram, and Facebook channels.

Open the AI Agent node

Select the node in a WhatsApp, Instagram, or Facebook channel.

Enable the tool

In the Tools tab, enable Reply with voice notes.

Adjust the description (optional)

Open the tool settings to edit the description that guides the agent on when to use it, and save.

Save the node

From then on, the agent can reply with voice notes.

Behavior

The agent’s turn ends after sending the voice note. The tool always pauses; this is not configurable.
The platform chooses the voice. There is no field to select it.
You can only edit the tool’s description; it has no parameters or behavior section.

Current date and time

Gets the current date and time. Allows specifying a time zone to get the correct local time. Parameters:

Name	Type	Required	Description
`timezone`	`string`	No	Desired time zone in IANA format (e.g., “America/Guayaquil”, “UTC”).

Prompt Example:

“Before processing any time-dependent request (such as ‘good morning’ or appointments for ‘today’), call get_current_date_time to know the exact time.”

Arguments Example:

{
  "timezone": "America/Mexico_City"
}

Response: Returns a string with the formatted date and time, for example: "Monday, December 08, 2025 10:30 AM". Notes:

Has no visible side effects for the user.
It is purely informational for the model.

Day of the week

Calculates and returns the day of the week corresponding to a specific date provided. Parameters:

Name	Type	Required	Description
`date`	`string`	Yes	Date to query (ISO format or standard readable format, e.g., “2024-12-25”).

Prompt Example:

“If the user wants to schedule for a specific date, first verify what day of the week it is using get_weekday_of_a_date to ensure it is a business day.”

Arguments Example:

{
  "date": "2025-12-10"
}

Response: Returns the name of the day of the week, for example: "Wednesday". Notes:

Useful for validating appointments or schedules.

Memory management

The following tools are enabled from the Memory management section of the AI Agent node’s Context tab. They give the agent working memory, access to the inbox CRM, flow context variables, and platform information. See the Memory management section of the AI Agent node for how to enable them and configure their permissions.

Internal memory

Lets the agent save, recall, and delete information from the conversation to reuse it later. It is per-user working memory with limited retention: it recalls the same user’s data across conversations, but it is not a permanent store. Operations:

Operation	Action	Description
Create	`save`	Saves a value to the user’s memory.
Recall	`recall`	Retrieves previously saved data.
Delete	`delete`	Deletes a saved value.
List	`list`	Lists the available saved data.

Notes:

Includes an optional Memory space field to isolate or group saved entries, preventing different contexts from mixing.
Each operation supports configuring its parameters in AI, variable, or static mode, along with a post-execution behavior (Continue, End, or Pause).

Inbox CRM

Reads, sets, and updates the contact’s CRM data (name, phone, tags, notes, and more) directly from the inbox. Operations:

Operation	Action	Description
Read	`readStoredParams`	Reads the contact’s stored data.
Set	`setStoredParams`	Sets contact data.
Update	`updateStoredParams`	Updates existing contact data.
List tags	`listTags`	Lists the contact’s tags.
Add tag	`addTag`	Adds a tag to the contact.
Remove tag	`removeTag`	Removes a tag from the contact.

Notes:

Permissions are managed per group (contact data / tags) with separate read and write access. You can restrict which tags the agent may access.
Requires at least one connected channel for contacts to exist to operate on.

$context

Reads and writes flow context variables to share data between nodes. Operations:

Operation	Action	Description
Get	`get`	Reads the value of a context variable.
Set	`set`	Writes or updates a context variable.

Notes:

It appears in the tools list as $context to distinguish it from the Context tab.
Variables written become available to later nodes of the flow, for example {{$context.my_variable}}.

Platform information

Read-only context blocks that inject platform data into the agent’s prompt. They do not consume a tool call: the information is added automatically as context when the source is active.

Source	Fields	Description
User information	`names`, `referenceId`, `botId`, `roomId`	Current user’s data.
Channel information	`name`, `type`	Connected channel’s data.
Company information	`name`, `plan`, `organizationId`	Company’s data.

Notes:

All three sources are read-only; the agent cannot modify them.
They are injected as context blocks, so they do not count as a tool invocation.

Databases

Native connection to Databases (Datum v2) from the Tools tab of the AI Agent node. It lets the agent search, create, update, and delete records in your collections without chaining auxiliary nodes. Operations: search, create, update, and delete records on the collections you enable.

Parameter	Type	Required	Description
`operation`	`enum`	Yes	`search`, `create`, `update`, or `delete`.
`collectionId`	`string`	Yes	Databases collection to operate on.
`query`	`object`	No*	Search filters. *Required for `search`.
`document`	`object`	No*	Document to insert or update. *Required for `create` and `update`.
`recordId`	`string`	No*	Record ID. *Required for `update` and `delete`.

Notes:

The agent can only operate on collections from the same company that you explicitly enable in the Configure popover.
Write operations persist immediately.
Post-execution behavior (Continue, End, or Pause) is configured per operation.

AI Routing

Sends the user’s request to the workflow that can handle it, when they ask for something this agent cannot do. It uses the project’s AI Routing to pick the destination, so the project must have AI Routing enabled. Parameters:

Name	Type	Required	Description
`mode`	`enum`	No (Default: `evaluate`)	`evaluate` (picks the destination automatically), `direct` (destination named by the user) or `search` (lists the available destinations).
`message`	`string`	No*	The user’s message verbatim. *Required for `evaluate`.
`target`	`string`	No*	Destination name or ID. *Required for `direct`.

Prompt example: Unlike other tools, ai_router does not need you to tell it when to fire: enabling it automatically appends the handoff rules to the agent’s instructions. All you have to define is the agent’s scope, because whatever falls outside it is what gets handed off.

“You only handle billing questions: account statements, payment dates and charge breakdowns.”

Argument examples: Case: the user asks something outside the agent’s scope (most common)

{
  "mode": "evaluate",
  "message": "I want to report a problem with my order"
}

Case: the user explicitly names the destination

{
  "mode": "direct",
  "target": "Support"
}

Case: check the available destinations before deciding

{
  "mode": "search"
}

Response: Returns a JSON object with the chosen destination, whether the handoff ran, and guidance for the model on what to do next. In search, and in direct with no match, it returns the available destinations and hands off nothing. Notes:

A successful handoff ends the agent’s turn: the user’s reply comes from the destination workflow. The agent must not answer as well.
If nothing matches or the handoff fails, the agent still owns the conversation and must keep helping the user.
Requires the project to have AI Routing enabled, with workflows that have a name and description configured. The post-execution behaviour is fixed: a handoff always ends the agent’s turn.
The agent can only hand off to workflows in the same project.

The handoff rules are appended automatically when you enable the tool: the agent already knows it must answer whatever its instructions cover, that it must not hand off because another tool failed or because it is unsure, and that it must not answer on top of handing off. You do not need to write any of this in the prompt.

Knowledge files: search by file

When you upload files in the Context tab of the AI Agent node, the agent automatically gets the search function to query them. With file search turned on, that same function searches better. No new functions or parameters appear:

It detects when a question refers to a specific document and searches only that document.
It searches every document of the agent, not only those in the first project.
It answers grounded in the documents: if nothing relevant is found, it says so.

File search is turned on per agent, in the AI Agent node configuration (knowledgeSettings.fileSearch). It is off by default: without that setting, the agent behaves exactly as before. To turn it on, open the Context tab of the AI Agent node and enable Search by document name. The switch appears once the node has knowledge files.

How the document is detected

The agent compares the question with the names of the documents uploaded to the node:

Case, accents, dashes and the extension are ignored, and small typos are tolerated: “tarifario 2026”, “TARIFARIO” and “Tarifario_2026.pdf” find the same file.
A document is detected when the question includes its full name (“how much is the basic plan in the tarifario 2026?”) or mentions it after a word such as “in”, “according to”, “search”, “check” or “document”. The cue words are matched in Spanish, the language of most agents: “en”, “según”, “busca”, “revisa”, “consulta”, “dice”, “documento”, “archivo”.
Generic words such as “manual”, “guide”, “policy” or “document” are not enough on their own to pick a file.

Examples of user phrases:

Question	Where it searches
”Según el tarifario, ¿cuánto cuesta el plan básico?”	Only in Tarifario 2026.pdf
”¿Qué dice el manual de reclamos sobre los plazos?”	Only in Manual de Reclamos.docx
”What are your opening hours?”	In every document

If no document matches, every document is searched, as always. If the mentioned document has nothing relevant, the search is repeated over every document before answering. Only the documents uploaded to that AI Agent node are considered. Files from other agents or other companies are never queried. If a file was renamed or deleted after the agent was saved, its current state is used.

Answers grounded in your documents

With file search on, the search result tells the agent to answer only from what the documents returned. When nothing relevant is found, the agent doesn’t get an empty list: it gets an explicit notice that there are no relevant results and that it must not answer from general knowledge. In our tests with questions the documents cannot answer, this change reduced invented answers from 11 in 15 to 1 in 15, without changing the agent’s instructions.

Give your files clear, distinct names, for example “Price list 2026.pdf” instead of “final document.pdf”. That way the agent recognizes when the user is talking about a specific one.

Getting Started

Developer

Jelou CLI

Outbound Messaging

Inbox

Marketplace

Channels

Configuration

Billing

Product search

Transfer to agent

Interactive messages

General settings: Buttons and list

Custom messages: Webview and Call to Action

Webview fields

Call to Action fields

Supported channels

Buttons and list

Send Call to Action (CTA) (Legacy)

Reply with voice notes

Current date and time

Day of the week

Memory management

Internal memory

Inbox CRM

$context

Platform information

Databases

AI Routing

Knowledge files: search by file

How the document is detected

Answers grounded in your documents

​Product search

​Transfer to agent

​Interactive messages

​General settings: Buttons and list

​Custom messages: Webview and Call to Action

​Webview fields

​Call to Action fields

​Supported channels

​Buttons and list

​Send Call to Action (CTA) (Legacy)

​Reply with voice notes

​Current date and time

​Day of the week

​Memory management

​Internal memory

​Inbox CRM

​$context

​Platform information

​Databases

​AI Routing

​Knowledge files: search by file

​How the document is detected

​Answers grounded in your documents

Product search

Transfer to agent

Interactive messages

General settings: Buttons and list

Custom messages: Webview and Call to Action

Webview fields

Call to Action fields

Supported channels

Buttons and list

Send Call to Action (CTA) (Legacy)

Reply with voice notes

Current date and time

Day of the week

Memory management

Internal memory

Inbox CRM

$context

Platform information

Databases

AI Routing

Knowledge files: search by file

How the document is detected

Answers grounded in your documents