How to Build an AI Personal Operating System

TL;DR

Build a personal operating system by combining a custom dashboard with Supabase storage, an Anthropic API key, Vercel functions, Google Calendar, and a Telegram bot that sends voice notes through OpenAI Whisper. The resulting system can classify and prioritize tasks, maintain persistent memory, display financial data, and centralize calendars, habits, journals, health information, goals, and CRM records.

Transcript

I spent 2 weeks building a custom personal operating system which runs my life. It manages all my tasks, finances, calendar, and automatically prioritizes my day based on the things that matter. It allows me to send voice notes from anywhere and have AI automatically populate my to-do list instead of what I used to do, which was going into notion a... Read More

Key Insights

• The system is a custom dashboard connected to a cloud memory layer, allowing its interface to change as the user's interests and priorities evolve while the stored personal data remains available on the back end.
• Voice capture works by sending a Telegram message to a bot, routing it through a Vercel webhook, transcribing voice notes with OpenAI Whisper, classifying the resulting text with AI, and saving it in the correct Supabase category.
• Supabase serves as the system's central memory database, storing logged information that appears through the dashboard. The creator selected it after considering Firebase and Supermemory, describing Supabase as the most sovereign configuration for the project.
• Daily prioritization starts with reviewing the CRM and starring three to five important tasks. Those selected items then appear on the home dashboard as key tasks, keeping attention on the work that should be completed first.
• Financial tracking uses a Google Sheet with an API connection that feeds the dashboard. The finance pulse can display net worth over time, monthly earnings, daily earnings, and current net worth, while remaining hidden by default for privacy.
• The dashboard brings together a live Google Calendar, habit tracking, journaling, CRM records, nutrition estimates, health information, goals, finances, and prioritized tasks, creating one interface for information that had previously been distributed across several locations.
• The CRM and nutrition components required repeated rebuilding because data did not always save correctly. The CRM was described as the hardest component, with refresh and logging problems demanding multiple debugging stages before it worked reliably.
• The development process depends on trial and error after initial implementation. The creator used the system for several days, added desired features as practical needs became clear, and fixed problems whenever existing functions failed during real daily use.

Questions & Answers

Q: How do you build an AI personal operating system?

Start by mocking the desired dashboard design, then implement its architecture with Claude Code or another capable coding system. Create a custom schema for areas such as content, products, business operations, and finance. Use Supabase as the memory database, connect the required AI API, add integrations such as Telegram and Google Calendar, protect access with a password, and test every component through regular use.

Q: How can Telegram voice notes automatically create tasks?

Create a Telegram bot through BotFather and point its webhook to a Vercel function. When a voice note is sent to the bot, OpenAI Whisper transcribes the recording into text. The AI then reads that text, determines its category and priority, and saves it in Supabase. The dashboard reflects the saved information without requiring manual entry into a task application.

Q: What is the role of Supabase in the personal dashboard?

Supabase acts as the back-end memory system and stores the information logged through the personal operating system. The dashboard is the front end used to view and interact with that stored data. Keeping memory separate from the interface allows the dashboard to evolve around changing interests and priorities while the underlying information remains centrally stored and transportable to another large language model.

Q: How does the dashboard prioritize daily tasks?

Incoming tasks can be categorized and prioritized by AI when they are captured through Telegram. Each morning, the user also reviews the CRM and stars three to five tasks that matter most for that day. Those starred entries appear as key items on the home dashboard. Once completed, they can be replaced with new priorities, keeping the main view focused on essential work.

Q: What tools are used to create the AI dashboard?

The demonstrated build uses Claude for the initial design and Claude Code for architecture and implementation. Supabase stores the data, an Anthropic API key supplies Claude access, Vercel hosts the function connected to the Telegram webhook, BotFather creates the Telegram bot, and OpenAI Whisper transcribes voice notes. Google Calendar supplies live calendar information, while Google Sheets supports the financial section.

Q: What information can the personal operating system manage?

The dashboard manages prioritized tasks, a live calendar, daily habits, journal entries, CRM records, nutrition information, health data, goals, and finances. Its home view includes starred priorities and recurring habits. The finance pulse can show net worth over time, current net worth, monthly earnings, and daily earnings, while the back-end memory system stores and organizes the underlying personal data.

Q: How does the dashboard track finances and protect privacy?

Financial information is collected in a Google Sheet that uses an API to pull in live data, which is then reflected on the dashboard. The finance pulse tracks net worth over time, earnings for the month, earnings for the day, and current net worth. It is hidden by default, so leaving the computer open does not immediately expose the displayed financial figures.

Q: Why does an AI dashboard require testing and debugging?

Connected components can fail to store or refresh data correctly, even after the basic architecture is working. In this build, the CRM produced glitches where information did not log or appeared lost after refreshing, and the nutrition section also required several rebuilds. Using the system daily exposed these problems and revealed useful feature changes, making repeated testing and adjustment a necessary part of development.

Summary & Key Takeaways

• The personal operating system replaces fragmented task management with a dashboard backed by persistent cloud storage. Telegram accepts text or voice notes from anywhere, OpenAI Whisper transcribes recordings, and AI classifies each entry before saving it in Supabase and presenting it within the appropriate dashboard category.

• The build began with visual mockups in Claude, followed by architecture and implementation in Claude Code. Its supporting services include Supabase for stored data, an Anthropic API key for AI functionality, Vercel for the Telegram webhook, Google Calendar for live events, and Google Sheets for financial information.

• The dashboard combines financial metrics, prioritized tasks, daily habits, journaling, CRM records, nutrition estimates, health data, goals, and calendar information. Building it took a couple of weeks and required repeated debugging, especially where CRM and nutrition data failed to save reliably. Continued daily use revealed additional improvements.