AI Agents

Last updated: March 2026

Learn how to create, configure, and deploy AI-powered virtual users that autonomously perform tasks, interact with your data, and integrate with external services.

What Are AI Agents

AI Agents in Webooto are AI-powered virtual users that can be programmed to perform tasks autonomously within your workspace. Each agent operates as an independent entity with its own identity, instructions, and set of capabilities. Agents can query and manipulate your data, communicate with external services, send emails and notifications, generate documents, and trigger workflows -- all without manual intervention.

Every agent is defined by a profile that includes a first name, last name, email address, avatar, and avatar color, giving it a recognizable identity across the platform. Under the hood, each agent is driven by a large language model (LLM) that interprets instructions written in natural language and decides which tools to invoke to accomplish a given task.

Agents support multiple AI models from the Gemini family, allowing you to balance capability, speed, and cost depending on the use case. A temperature setting (ranging from 0 to 1) controls how creative or deterministic the agent's responses are: lower values produce more predictable outputs, while higher values encourage varied and creative responses.

Whether you need an agent that automatically processes incoming support tickets, enriches CRM records with data from external APIs, sends weekly summary reports, or joins meetings to take notes, the AI Agents system provides a flexible and extensible framework for building intelligent automation.

Creating an Agent

Setting up a new AI Agent involves defining its identity, selecting a model, writing its system prompt, and configuring the tools it is allowed to use. Follow the steps below to create and deploy your first agent.

Define the agent identity

Provide a first name, last name, and email address for the agent. Choose an avatar and avatar color so the agent is easily recognizable in conversations, activity logs, and notifications. The email address is used as the sender when the agent sends emails.

Select an AI model

Choose from the available models: Gemini 2.5 Flash (fast and cost-effective), Gemini 2.5 Pro (advanced reasoning), Gemini 1.5 Flash (lightweight tasks), or Gemini 1.5 Pro (balanced performance). The model determines the quality of the agent's reasoning and its token limits.

Set the temperature

Adjust the temperature slider between 0 and 1. Use a low value (0.0 - 0.3) for tasks requiring precision and consistency, such as data extraction or record updates. Use a higher value (0.5 - 0.8) for creative tasks like drafting emails or generating summaries. Avoid values very close to 1 unless you want highly varied outputs.

Write the system prompt

The system prompt is the core instruction set for the agent. Write clear, detailed instructions in natural language describing what the agent should do, how it should behave, what tone to use, and any constraints it must follow. For example: "You are a customer support agent. When a new ticket is created in the Support Requests entity, classify it by urgency, draft a response, and assign it to the appropriate team member."

Enable tools

Review the list of available tools and enable only the ones the agent needs. For example, an agent that reads and updates records needs entity_query, entity_get, and entity_update. An agent that sends reports needs generate_pdf and send_email. Restricting tools follows the principle of least privilege and prevents unintended actions.

Configure HTTP allowlist

If the agent uses http_get, http_post, or read_url tools, define the HTTP allowlist to control which external URLs the agent is permitted to access. This prevents the agent from making requests to unauthorized or untrusted endpoints. Specify allowed domains or URL patterns.

Set max iterations

Define the maximum number of iterations the agent can perform in a single run. This acts as a safety limit to prevent runaway executions. A typical value is between 5 and 20, depending on the complexity of the task. If the agent exceeds this limit, the run is stopped with a timeout status.

Activate the agent

Toggle the isActive flag to enable the agent. Once active, the agent can be triggered via chat, scheduled tasks, webhooks, or workflow steps. You can deactivate the agent at any time to pause its operations without deleting its configuration.

Tip

Start with a narrow scope and a small set of tools. You can always expand the agent's capabilities later once you have validated its behavior in a controlled setting.

Available Tools Reference

Each AI Agent can be granted access to a specific set of tools. Tools are the actions an agent can perform during a run. You enable or disable each tool individually in the agent's configuration. Below is a comprehensive reference of all available tools, their purpose, and typical use cases.

Tool	Description	Use Case
entity_query	Search and filter entity records using query parameters. Supports filtering by fields, sorting, pagination, and full-text search.	Finding all overdue invoices, listing active customers by region, searching for records matching specific criteria.
entity_get	Retrieve a specific entity record by its unique ID. Returns the full record with all fields.	Fetching a customer profile to read contact details, loading a ticket to check its status before updating.
entity_create	Create a new record in any entity. Accepts field values as input and returns the created record.	Creating a follow-up task after processing a support ticket, adding a new lead from enriched data.
entity_update	Update one or more fields on an existing entity record identified by ID.	Marking a task as completed, updating a contact's email address, changing an order status.
entity_delete	Permanently delete an entity record by ID. Use with caution as this action is irreversible.	Removing duplicate records, cleaning up test data, deleting cancelled orders.
http_get	Make an HTTP GET request to an external API. The target URL must be in the agent's HTTP allowlist.	Fetching weather data, retrieving stock prices, pulling information from a third-party REST API.
http_post	Make an HTTP POST request to an external API with a JSON body. The target URL must be in the HTTP allowlist.	Submitting data to a webhook, posting a message to an external chat service, triggering a third-party action.
read_url	Read and extract content from a web page URL. Useful for scraping or summarizing web content.	Reading a knowledge base article, extracting product information from a supplier page, summarizing a blog post.
send_email	Send an email using the agent's configured email address as the sender. Supports subject, body (HTML), recipients, CC, and BCC.	Sending a weekly summary report, notifying a customer of a status change, delivering a generated PDF invoice.
send_notification	Send an in-app notification to one or more users within the Webooto workspace.	Alerting a team lead about a high-priority ticket, notifying an assignee about a new task, broadcasting system updates.
run_workflow	Trigger the execution of an existing workflow by its ID, optionally passing input parameters.	Kicking off an approval process after data validation, chaining multiple automation steps, launching complex multi-step operations.
generate_pdf	Generate a PDF document from a predefined template, populating it with dynamic data.	Creating invoices, generating contracts, producing monthly reports with charts and tables.
log	Write a log message for debugging and auditing purposes. Logs are visible in the agent run history.	Tracking intermediate values during data processing, recording decision points, debugging tool call failures.

Warning

The entity_delete tool permanently removes records. Only enable it for agents that specifically require deletion capabilities, and always test with non-production data first.

Note

When using http_get, http_post, or read_url, all target URLs must be explicitly added to the agent's HTTP allowlist. Requests to URLs not on the allowlist will be blocked.

Scheduled Tasks

Scheduled tasks allow an agent to run automatically on a recurring basis using cron expressions. This is ideal for periodic operations such as daily data syncs, weekly report generation, hourly monitoring checks, or monthly cleanup routines.

Each scheduled task is defined by a name, an instruction (the prompt sent to the agent when the task triggers), and a cron schedule expression. The cron expression follows standard cron syntax with five fields: minute, hour, day of month, month, and day of week. For example, "0 9 * * 1" runs the task every Monday at 9:00 AM.

Each scheduled task has an active/inactive toggle, allowing you to pause individual schedules without deleting them. The platform tracks the last run timestamp for each task, making it easy to verify that schedules are executing as expected.

When a scheduled task fires, the agent receives the task's instruction as its input and executes using the same tools, model, and system prompt configured on the agent. The resulting run is recorded in the agent's run history with a "scheduled" trigger mode.

Create a new scheduled task

Navigate to the agent's configuration and open the Scheduled Tasks section. Click "Add Task" and provide a descriptive name for the task (e.g., "Daily Invoice Reminder" or "Hourly Lead Enrichment").

Write the task instruction

Enter the instruction that will be sent to the agent each time the task runs. Be specific and self-contained, as the agent does not retain memory between scheduled runs. Example: "Query all invoices where status is overdue and dueDate is before today. For each one, send a reminder email to the customer contact."

Set the cron schedule

Enter a cron expression to define the schedule. Common examples: "0 */2 * * *" (every 2 hours), "30 8 * * 1-5" (weekdays at 8:30 AM), "0 0 1 * *" (first day of each month at midnight). The platform validates the expression and shows a human-readable description of the schedule.

Activate and monitor

Toggle the task to active. Monitor the "Last Run" timestamp and check the agent's run history to verify that the task executes correctly. If a run fails, review the tool call details and logs to diagnose the issue.

Tip

Use descriptive task names that include the frequency and purpose (e.g., "Weekly Sales Summary Report") so they are easy to identify in the scheduled tasks list.

Chat & Conversation

The Chat interface provides a direct, conversational way to interact with your AI Agent. You can send messages, ask questions, and issue instructions in natural language. The agent processes your input using its system prompt and configured tools, then responds with text, data, or the results of actions it has taken.

Conversations support multi-turn memory, meaning the agent retains context from previous messages within the same session. This allows for iterative workflows: you can ask the agent to find records, review the results, and then instruct it to update specific entries -- all within one continuous conversation.

Each message exchange is recorded as part of the agent's run history. You can review past sessions to see exactly what was said, which tools were called, what inputs and outputs each tool produced, and how long the entire interaction took. This transparency makes it easy to audit agent behavior and refine prompts.

The chat interface is particularly useful for ad-hoc tasks, exploratory data analysis, testing new system prompts, and interactive debugging. For repetitive tasks, consider converting a successful chat interaction into a scheduled task or webhook-triggered flow.

Note

Multi-turn memory is maintained within a single chat session. If you start a new session, the agent begins with a fresh context. For tasks that require historical awareness, include relevant context in the system prompt or instruct the agent to query past records.

Webhook Integration

Webhook integration enables external systems to trigger an AI Agent by sending an HTTP request to a dedicated webhook URL. This is the primary mechanism for event-driven automation: when something happens in an external service (a form is submitted, a payment is received, a build completes), the external service sends a webhook to Webooto, and the agent processes it.

Each agent can be configured with a unique webhook endpoint. When the endpoint receives a request, the payload is passed to the agent as input. The agent then processes the data according to its system prompt and tools, performing any necessary actions such as creating records, sending notifications, or calling external APIs.

Webhook-triggered runs are recorded in the agent's run history with a "webhook" trigger mode. The run details include the incoming payload, all tool calls made during processing, the final output, iteration count, duration, and status (success, failed, or timeout).

The agent run history provides comprehensive tracking for every execution, regardless of trigger mode. Each run record includes: the session ID, the trigger source (chat, scheduled, or webhook), input and output messages, detailed tool call logs (tool name, input parameters, output results, and duration for each call), the total number of iterations, the final status, and the overall run duration. This data is invaluable for debugging, performance optimization, and compliance auditing.

Enable the webhook

Open the agent's configuration and navigate to the Webhook section. Enable the webhook toggle to generate a unique webhook URL for the agent. Copy this URL to use in your external service's webhook configuration.

Configure the external service

In your external service (e.g., Stripe, GitHub, Typeform, or a custom application), set up a webhook that sends HTTP POST requests to the agent's webhook URL. Ensure the payload is formatted as JSON.

Handle the payload in the system prompt

Update the agent's system prompt to describe how it should interpret incoming webhook payloads. For example: "When you receive a webhook from Stripe, extract the event type and customer ID. If the event is payment_succeeded, update the customer record's payment status to paid and send a confirmation email."

Test and monitor

Send a test webhook from the external service and verify the agent processes it correctly. Check the run history for the webhook-triggered run, review tool call details, and confirm the expected actions were performed.

Tip

Combine webhook-triggered agents with the run_workflow tool to orchestrate complex, multi-step automations that span multiple entities and external services.