Configure the AI Model (ai.properties)

Before using AI Agents, you must configure the connection to an AI provider. This is done via an file (click on filename to download an example), which must be uploaded to the resource folder of the Super Admin user.

Parameter	Description	Required

Parameter	Description	Required
`ai-apikey`	The API key for your chosen AI provider.	REQUIRED
`ai-endpoint`	The API endpoint URL of the AI provider.	REQUIRED
`ai-model`	The specific model to use (e.g. `claude-3-5-sonnet`).	REQUIRED
`ai-connector`	The connector class identifying the AI provider. See supported values below.	REQUIRED
`ai-provider-label`	A display label for the AI provider shown in the UI.	OPTIONAL

Supported Connectors (ai-connector)

The following connector values are available, each referencing a supported AI provider:

OpenAIModelManagerImpl — Open AI / OODA AI
AzureOpenAIModelManagerImpl — Azure OpenAI
AnthropicModelManagerImpl — Anthropic (Claude)
GoogleAIGeminiModelManagerImpl — Google Gemini
MistralAIModelManagerImpl — Mistral AI

TIM Flow ships with an file (click on file name for download) that defines the available AI agent templates. You can modify existing agents or define entirely new ones. This file can be uploaded to the tenants resource folder and will overwrite the product standard.

Customizing Agents (agents.json)

File Structure

The agents.json file contains two top-level arrays:

actionAgents — Agents that execute a task and write results to process variables.
decisionAgents — Agents that evaluate a situation and select a process transition (e.g. at an XOR Gateway).

Each agent entry follows this structure:

{
  "name": "My Agent",
  "type": "ActionAgent",
  "key": "my_agent",
  "metadata": {
    "description": "What this agent does.",
    "tags": ["Tag1", "Tag2"]
  },
  "instruction": {
    "prompt": "Your prompt / instruction for the AI."
  },
  "output": {
    "variables": ["variable_one", "variable_two"]
  }
}

Field	Description

Field	Description
`name`	Display name shown in the Workflow Designer.
`type`	ActionAgent or DecisionAgent
`key`	Optional unique identifier. Recommended for custom agents.
`metadata.description`	Shown as a tooltip or description in the UI.
`metadata.tags`	Used for filtering and categorization in the UI.
`instruction.prompt`	The AI prompt. Can be left empty for fully custom agents.
`output.variables`	Process variable names the agent will populate. Decision Agents select a transition instead and do not use this field.

Modifying an Existing Agent

Locate the agent by name in the appropriate array and edit the instruction.prompt or output.variables as needed. For example, to make the AI Text Extraction Agent also extract an email address, add email to its output.variables array and update the prompt accordingly.

Adding a New Agent

Add a new object to either actionAgents or decisionAgents.

For an AI Action Agent you need to provide name, type, key, metadata.description, metadata.tags, instruction.prompt and output.variables.

For an AI Decision Agent you need to provide name, type, key, metadata.description, metadata.tags and instruction.prompt.

For a fully flexible starting point, the pre-built AI Action Agent and AI Decision Agent ship with empty prompts — you can duplicate one as a template for your custom agent.

After editing agents.json, upload the updated file to the Super Admin resource folder and restart TIM Flow for the changes to take effect.