Creating a Smart Home Assistant by Combining ChatGPT with Siri via iOS Shortcuts

The author explains that after experimenting with ChatGPT and GPT‑3, existing voice assistants like Siri, Alexa, and Google Home feel insufficient, prompting the creation of a more intelligent smart‑home system that leverages GPT‑3 to interpret natural‑language commands and Siri to execute them.

A HomeKit‑based smart‑home setup is described, containing dozens of devices (lights, thermostats, floor heating, ventilation, cameras, etc.) across multiple rooms. The goal is to let GPT‑3 parse user requests and return structured JSON that Siri can act upon.

Four request categories are defined:

Command : control a device (e.g., turn a light on).

Query : retrieve a device state (e.g., temperature).

Answer : respond to a non‑home‑automation question.

Clarify : ask the user to rephrase an ambiguous request.

Each JSON response must contain specific fields. For example, a command response includes action , location , target , value , comment , and optionally scheduleTimeStamp . Queries contain property , and answers contain answer .

Sample command response:

{
  "action": "command",
  "location": "office",
  "target": "light",
  "value": "on",
  "comment": "Turning the light on for you.",
  "scheduleTimeStamp": ""
}

Sample scheduled command (turn off bedroom light after 20 minutes):

{
  "action": "command",
  "location": "bedroom",
  "target": "light",
  "value": "off",
  "comment": "Switching the lights off in your son's bedroom.",
  "scheduleTimeStamp": "Mon Jan 16 2023 12:16:31 GMT+0000"
}

Sample query response for kitchen thermostat temperature:

{
  "action": "query",
  "location": "kitchen",
  "target": "thermostat",
  "property": "temperature"
}

Sample answer response for a travel recommendation:

{
  "action": "answer",
  "answer": "St Albans has a great selection of historical sites and attractions..."
}

The article then shows how to set up an iOS Shortcut that captures user input, sends it to the OpenAI API (model text-davinci-003 ), and processes the JSON response with a series of if statements to trigger the appropriate HomeKit action via Siri.

API request details include headers ( Content-Type: application/json and Authorization: Bearer <YourAPIToken> ) and a prompt format of <Text> Request: <Provided Input> Response: . The response is parsed, the action field determines the branch (command, query, answer, clarify), and relevant fields ( location , target , value , etc.) are used to control devices.

Finally, the author encourages readers to experiment with simple English prompts to create their own intelligent assistants, highlighting the futuristic potential of combining GPT‑3 with Siri for personalized smart‑home automation.