Overview

AI Agents allow you to integrate AI-powered automation steps directly into your process models in TIM Flow. An agent executes at a defined point in a process — similar to an Action Handler — and can read instance documents, process data, and write results back into process variables.

Setup and configuration

Before you can use AI Agents, the connection to an AI model must be configured. This includes setting up the required provider credentials and selecting the appropriate connector. Additionally, TIM Flow ships with a predefined set of agent templates that can be customized or extended to fit your needs.

For full details on both topics, refer to this page: https://tim-doc.atlassian.net/wiki/spaces/eng/pages/edit-v2/2010382371?draftShareId=28ed52ec-b08a-4c1b-8bf9-5092776c1f77

Using AI Agents in Workflow Design

TIM Flow ships with a comprehensive list of Agent Templates

Action Agents

• AI Action Agent (custom prompt, flexible)

• AI Text Extraction Agent

• AI Sentiment Analysis Agent

• AI Summary Agent

• AI Keyword Extraction Agent

• AI Task Detection Agent

• AI Language Detection Agent

• AI Date Extraction Agent

• AI Risk Detection Agent

• AI Role Identification Agent

• AI Location Extraction Agent

• AI Email Parsing Agent

Decision Agents

• AI Decision Agent (custom prompt, flexible)

• AI Approval Decision Agent

• AI Escalation Decision Agent

• AI Routing Decision Agent

• AI Priority Decision Agent

• AI Compliance Check Agent

• AI Risk Evaluation Agent

• AI Contract Review Agent

• AI Budget Allocation Agent

• AI Hiring Decision Agent

• AI Incident Response Agent

Adding an Agent to a Process

Agents are configured directly in Workflow Design. In the Workflow Items accordion, a new Agents section is available. Agents can be dragged and dropped onto the Node Enter or Node Leave sections of an Activity or XOR Gateway — the same way existing events are placed.

Open

Open

Agent Configuration Options

After placing an agent on a node, the following options are available:

Attachment Options

Specify whether the agent can read documents from the running process instance, and which documents are accessible. Two modes are available:

• Regex — Provide a regular expression to match documents by name.

• Configuration — Define access using the following sub-options:

  • Attachment Scope — Limit access to the First document, Last document, or All documents.

  • Filter Attachment Type — Restrict by document type: PDF, TXT, DOCX, EML, or All Types.

Open

Agent Template & Instruction

Select from the predefined agent templates or create an individual one to fit your needs (AI Action Agent). Each agent has an Agent Instruction field (the prompt) that can be customized to suit your specific use case.

Open

Output Variables

Define which process variables the agent is expected to populate. They limit the agent's output — only the specified variables will be written back to the process.

Open

Testing an Agent

Each agent configuration includes a Test Prompt option that lets you preview the agent's behavior before deploying. In the test view you can:

• See the agent's output in the defined Output Variables format.

• Inspect the raw underlying JSON output structure.

• Upload test documents to simulate how the agent handles attachments.

Open

How Agents Execute

An AI Agent behaves like an Action Handler in the process automation. At the point where the agent is defined (Node Enter or Node Leave of an Activity or on a XOR Gateway), it executes automatically as part of the process flow — reading inputs, running the configured instruction against the AI model, and writing results to the specified output variables.

Example: Automated Onboarding Welcome Letter

As part of an employee onboarding process, an AI Agent can be used to automatically generate a personalized informational letter for the new employee's future colleagues. Once the onboarding instance is started and the relevant employee data is available as process variables, the agent is placed on the appropriate activity and instructed to compose a short, professional announcement.

This is how the Agent is confiugured:

Open

This is how the form / input variables look like for our example:

Open

On execution of the Agent, it reads the employee's name, department, and start date from the process, and generates a ready-to-use letter informing the team about their new colleague — without any manual writing effort:

Open