IBM agent types

Simple reflex agent

If this, then that: rules on the current reading only.

A simple reflex agent maps what it senses right now to an action with fixed condition-action rules. It keeps no memory and has no goal beyond its rules: the same reading always gives the same action.

A thermostat is the classic example. Change the target with the slider and run it again.

Use it when: The current reading is enough to act; Rules are known and stable; Speed and predictability matter most
Skip it when: The world is only partly visible (it cannot remember); Situations the rules do not cover

Run it

No model needed · CriogenAI engine in your browser

Ready

The run

Press Run to watch the pattern work, step by step.

GoalTarget °C: 21

This run is the TypeScript code below, running on the CriogenAI engine in your browser (WebAssembly). Without a key, a scripted model stands in for MODEL; with your key, your provider answers, called from this tab. The tools return made-up data.

Build it with CriogenAI

Built from primitives: Workflow.route (a router is a rule)

target = 21.0

def rule(state: Dict[str, Any]) -> str:  # the condition-action rules
    t = state["temperature"]
    return "heat" if t < target - 1 else "cool" if t > target + 1 else "idle"

def act(action: str) -> Callable[[Dict[str, Any]], Dict[str, Any]]:
    return lambda state: {"action": action}

wf = Workflow()
wf.add_node("sense", lambda state: None)
for action in ("heat", "cool", "idle"):
    wf.add_node(action, act(action))
    wf.edge(action, END)
wf.start("sense").route("sense", rule, {a: a for a in ("heat", "cool", "idle")})
app = wf.compile()
for reading in (17.5, 21.0, 23.6):
    print(reading, (await app.run({"temperature": reading})).state["action"])

const target = settings.target; // the thermostat's set point
const rule = (state: State) => (state.temperature < target - 1 ? "heat" : state.temperature > target + 1 ? "cool" : "idle");

const wf = new Workflow().node("sense", () => undefined).start("sense");
for (const action of ["heat", "cool", "idle"]) wf.node(action, () => ({ action })).edge(action, END);
wf.route("sense", rule, { heat: "heat", cool: "cool", idle: "idle" }); // the condition-action rules
const app = wf.compile();
const readings: string[] = [];
for (const temperature of [17.5, 21, 23.6]) readings.push(`${temperature} °C → ${(await app.run({ temperature })).state.action}`);
console.log(readings);