Carl - Your Life Admin Agent

An intelligent iMessage bot that lives in your chats to bookmark what you want to remember, proactively remind you of what you need to do, and seamlessly process both texts and native voice memos.

Comment

Description

Every day, we juggle an endless stream of mental tabs: gift ideas, urgent chores, bills, and random thoughts. Traditional task-management apps add friction; they require you to open a separate app, fill out forms, and set rigid alarms, which disrupts your flow.

Life Admin Agent (affectionately known as "Carl") eliminates this mental load by living exactly where you already communicate: Apple iMessage. It acts as an intelligent safety net that ensures nothing important slips through the cracks. Instead of fighting with complex productivity tools, you can naturally brain-dump your life admin tasks as regular text messages or voice memos, trusting the agent to proactively follow up with you exactly when needed.

How It Works

Built using Node.js and the Photon SDK, the agent intercepts and responds to iMessages natively on macOS. You can interact with it just like a human contact.

  • Native Voice Support: If you send a native Apple voice memo (.caf), the system automatically transcodes the audio and transcribes it into text.
  • Intent Execution: Once the message text is processed, an AI engine categorizes your intent, whether you want to save a bookmark, set a reminder, or query your active tasks, and saves the structured data to a local SQLite database.
  • Proactive Scheduling: A continuous background scheduler runs every 30 seconds to check for due tasks, and when the time arrives, the agent proactively texts you.
  • Smart Linking: It even features context awareness. If you set a reminder (e.g., "Remind me to buy Mom's gift"), it cross-references your previous bookmarks and intelligently adds context (e.g., "🔗 You saved a Keurig Espresso machine for that").

How We Used Minimax APIs

We heavily utilized the Minimax Text API (MiniMax-M2) as the singular "brain" of the application. Because MiniMax-M2 is an advanced reasoning model, we tasked it with handling our entire NLP pipeline in a single API call.

We pass the user's message alongside the current local datetime into the model, and it leverages its internal reasoning to output a perfectly structured JSON object. In one fell swoop, the Minimax API:

  1. Classifies the user's intent.
  2. Converts natural language time expressions ("in 10 days") into absolute UTC ISO datetimes.
  3. Detects task urgency based on tone (e.g., inferring that "ASAP" means a 6-hour window).
  4. Extracts normalized entity tags for our Smart Linking database queries.
  5. Generates friendly conversational replies.

Relying on Minimax allowed us to replace an entire suite of fragile parsers with one highly capable, contextual AI prompt.

Built With

Share this project:

Updates