AI Agent (orchestrator)

The AI Agent is the operational core of the modular agent. It runs inside the flow like any node, but it also has four role ports on its bottom edge where the pieces that give it capabilities connect.

What it does when it runs: it discovers the connected pieces → assembles the system prompt (adding relevant skills and RAG context) → calls the LLM → if the model requests a tool, it runs it and returns the result → repeats until the model answers. That’s the classic function calling loop (request tool → run it → return result → repeat).

Ports

Port	Cardinality	Connects to
Flow input (left)	1	The previous workflow node.
Flow output (right)	1	The next node.
`brain` (purple)	exactly 1	A Brain (LLM).
`tools` (blue)	0..N	One or more Tools.
`rag` (amber)	0..N	One or more RAG (retrieve).
`skills` (pink)	0..N	One or more Skills.

Configuration

Field	Type	Required	Description
system_prompt	textarea	No	Base instructions for the agent (personality, rules, goal).
input	text	No	User message. If empty, it uses what arrives from the previous node (`data.input`, `data.message` or the whole `data`). Supports templates.
temperature	number	No	Creativity. `0` = deterministic, `1` = creative. Default `0.7`.
max_iterations	number	No	Max rounds of the tool loop. Default `5`.

Output: data.response (the agent’s final text) and data.iterations (how many tool rounds it ran).

Example 1 — Simple assistant

An agent that answers questions, with brain only.

Start → AI Agent → End.
AI Agent — system_prompt: “You are a Floogos support assistant. Answer briefly.”
Connect a Brain (LLM) (openai / gpt-4o) to the brain port.

If the agent receives { "input": "What is a workflow?" }, it answers in data.response.

Example 2 — With a tool

By adding a tool, the agent can act, not just answer.

To the previous example, add a Tool that checks the weather (tool_kind: node, node_type: http) connected to the tools port.
system_prompt: “You help with the weather. If asked about a city, query the tool.”

For “will it rain in Madrid?”, the LLM decides to call the tool, the agent runs it, passes the result to the LLM and it composes the final answer. data.iterations will be 1 (one tool round).