Invocable agents

Invocable agents execute on demand, triggered by a workflow node or an API call. Unlike autonomous agents (which listen for events continuously), invocable agents process a specific input and return a structured result.

How they work


Agent node in workflow (or API call)
  → Invokes the agent with input data
  → Agent processes the information
  → Agent returns structured result (JSON)
  → The workflow continues with the returned data

Differences from autonomous agents

Aspect	Autonomous	Invocable
Activation	Automatic via events	Manual or from workflow
Execution	Continuous in the background	On demand
Input	System event	Data passed by the workflow
Output	Executed actions	Structured data returned
Typical use	Triage, automatic response	Data extraction, preparation

Available invocable agents

Data Extractor

Purpose: Extract structured data from documents, emails and forms.

Input: Document (PDF, image, text) or email content.

Output: JSON with extracted data, structured by type:


{
  "tipo": "factura",
  "emisor": "Empresa XYZ SL",
  "nif": "B12345678",
  "fecha": "2026-03-15",
  "importe_total": 1250.00,
  "iva": 262.50,
  "base_imponible": 987.50,
  "conceptos": [
    { "descripcion": "Servicio de consultoria", "cantidad": 1, "precio": 987.50 }
  ],
  "confianza": 0.95
}

Use cases in workflows:

Process supplier invoices received by email
Extract data from identity documents
Digitize hand-filled forms
Capture information from contracts

Meeting Prep

Purpose: Prepare pre-meeting briefings with full context on the contact.

Input: Contact ID and/or case associated with the meeting.

Output: Structured briefing with:

Contact and company summary
Recent interaction history
Open cases and their status
Latest relevant communications
Pending points from previous meetings
Suggested topics to address

Typical use case: A workflow scheduled 30 minutes before each meeting invokes Meeting Prep and sends the briefing to the assigned operator via email or in-app notification.

Use in workflows

Invocable agents are integrated into workflows via the Agent node:

Drag an Agent node onto the workflow canvas
Select the invocable agent to use
Configure the input data (workflow variables)
Define how the output will be used (destination variables)
Connect the node to the rest of the flow

Example workflow with invocable agent


Start (document received by email)
  → Agent: Data Extractor (extracts invoice data)
  → Decision (amount > 5000?)
    → Yes: Approval (supervisor approval)
    → No: User Task (direct accounting entry)
  → Notification (confirm registration to sender)
  → End

Invocable agents can also be called directly from the BiVelio REST API, enabling custom integrations with external systems.

Creating custom invocable agents

In addition to marketplace templates, you can create custom invocable agents:

Define the system prompt with the agent’s instructions
Specify the expected input format
Define the output format (JSON structure)
Configure the AI models and parameters (temperature, max tokens)
Test with sample data before integrating into workflows

Credits

Each invocable agent execution consumes AI credits. Consumption depends on:

Length of the input processed
Complexity of the prompt
AI model used
Length of the generated output

Typical consumption: 2-8 credits per execution.