Autonomous Agent Nodes

Agent nodes bring autonomous AI reasoning into your workflows. Unlike regular AI steps that run a single prompt, an Agent node gives an AI model a goal and a set of tools, then lets it figure out how to accomplish the task on its own.

How Agent Nodes Work

When a workflow reaches an Agent node, here is what happens:

1. Think — the AI model analyzes the task and the available data
2. Act — it decides which tool to use and executes it
3. Observe — it reviews the result
4. Repeat — it continues thinking, acting, and observing until the task is complete

This loop lets the agent handle tasks that require multiple steps, judgment calls, and adaptation based on intermediate results — things that would be difficult to hard-code as a fixed sequence of workflow steps.

When to Use Agent Nodes

Agent nodes are ideal when:

• The path is not predetermined — you do not know in advance exactly which steps need to run or in what order
• Judgment is required — the agent needs to interpret data and decide what to do next
• Multiple tools may be needed — the agent might need to search, create, update, and verify across different actions

Examples:

• "Review this support ticket and take the appropriate action — escalate it, respond to the customer, or close it"
• "Analyze this dataset, identify anomalies, and create a summary report"
• "Find all related records across multiple systems and compile a unified view"

For tasks with a known, fixed sequence of steps, regular workflow nodes are more efficient and predictable.

Available Tools

Agent nodes can use any action registered in your Sovereign deployment as a tool. This includes:

• Connector actions — interact with external services (create Jira issues, update Monday boards, search Confluence)
• Built-in actions — transform data, parse JSON, manipulate strings, do math
• Platform tools — manage workflows and executions within Sovereign itself:
  • Create and execute workflows
  • Search for workflows by status or tags
  • Create and manage workflow schedules
  • Create saved configurations
  • Navigate the UI

The tools available to each Agent node are configured in the workflow definition. You can limit which tools an agent has access to, controlling the scope of what it can do.

Configuration

When you add an Agent node to your workflow:

1. System prompt — define the agent's role and task. Be specific about what you want it to accomplish.
2. Allowed tools — select which tools the agent can use. Limiting tools improves focus and reduces unexpected behavior.
3. Max iterations — set a limit on how many think-act-observe cycles the agent can perform. This prevents runaway agent loops.

Agent Nodes vs AI Steps

License Requirements

Agent nodes require a license with AI/Agent capabilities enabled. If your license does not include this feature, Agent nodes will be rejected at dispatch time with a clear error message.

Next Steps

• AI-Powered Workflows — use simpler AI steps for predictable tasks
• AI Agent Access (MCP) — let external AI agents invoke your workflows
• Node Types — see all available node types