{"id":173,"date":"2026-06-27T23:35:21","date_gmt":"2026-06-27T20:35:21","guid":{"rendered":"https:\/\/agentura.ro\/blog\/?p=173"},"modified":"2026-06-27T23:42:36","modified_gmt":"2026-06-27T20:42:36","slug":"sensors-teach-your-ai-agent-to-watch-notify-and-act","status":"publish","type":"post","link":"https:\/\/agentura.ro\/blog\/sensors-teach-your-ai-agent-to-watch-notify-and-act\/","title":{"rendered":"Sensors: teach your AI agent to watch, notify, and act"},"content":{"rendered":"

Most AI agents wait to be asked. Sensors<\/strong> flip that \u2014 a tiny script that watches one condition on a loop, inside your agent’s own container, and pings your agent the moment something actually changes.<\/p>\n

A lot of useful work in a business is that shape: watch this, and tell me when it changes.<\/em> New orders. A price. A server. A car cooling down. One of our users recently told their agent to watch their Tesla’s cabin temperature and ping them when it cooled \u2014 then, over a single chat, turned it into a full thermostat<\/strong> that switches the AC on and off by itself. We’ll get there. First, let’s build one.<\/p>\n

What a sensor actually is<\/h2>\n

Three parts. A script<\/strong> you (or your agent) write at workspace\/sensors\/<name>.py<\/code> that checks one thing and prints a single line of JSON. A loop<\/strong> the platform runs for you \u2014 every interval_seconds<\/code> (minimum 10) \u2014 executing that script in the agent’s sandbox, with no model call per tick<\/em>, so it’s effectively free to run. And a delivery<\/strong> that fires only when the condition trips.<\/p>\n

The whole contract your script prints is just {\"condition_met\": true, \"message\": \"\u2026\", \"data\": {\u2026}}<\/code>. That’s it.<\/p>\n

You don’t need to read any of the code below to use sensors.<\/strong> The steps that follow show what’s happening under the hood \u2014 handy if you’re technical or want to hand-tune one. The everyday way to make a sensor is simply to ask<\/em>: tell your agent something like “watch X and tell me when Y”<\/em> and it writes the script, picks the trigger, and wires up the delivery for you. The Tesla thermostat further down was built exactly that way \u2014 entirely by chat, with no human touching code.<\/p>\n

Step 1 \u2014 Open the Sensors tab<\/h2>\n

Open any agent and go to the \ud83d\udef0\ufe0f Sensors<\/strong> tab. You’ll see the contract, a + Add sensor<\/strong> button, and any existing sensors with their last status.<\/p>\n

\"The
The Sensors tab. Each sensor shows its script, interval, trigger mode and last status.<\/figcaption><\/figure>\n

Step 2 \u2014 Write the script<\/h2>\n

In the Files<\/strong> tab, create a file under sensors\/<\/code> \u2014 say demo_check.py<\/code> \u2014 that prints the JSON contract. Here’s a self-contained example you can paste as-is:<\/p>\n

import json, os\n\n# Sensors hand the previous run's "data" back as $SENSOR_PREV_DATA, so a sensor\n# can compare against last time. This demo just counts runs.\nprev  = json.loads(os.environ.get("SENSOR_PREV_DATA", "{}"))\ncount = prev.get("count", 0) + 1\n\nprint(json.dumps({\n    "condition_met": count >= 1,            # <-- your real check goes here\n    "message": f"Demo check #{count}: all good",\n    "data": {"count": count},\n}))<\/code><\/pre>\n
The platform hands your script a couple of things for free as environment variables: $SENSOR_PREV_DATA<\/code> (the previous run’s data<\/code>, so you can diff against last time) and credentials to call your agent’s connectors.<\/p>\n
Step 3 \u2014 Add the sensor<\/h2>\n
Back on the Sensors tab, click + Add sensor<\/strong> and fill in the form: a name, the script filename, how often to run, and the trigger.<\/p>\n
\"The
The new-sensor form: point it at your script and choose how it fires.<\/figcaption><\/figure>\n