My Life Isn't an OS — So I Built One

Another LifeOS? There are numerous services labeled LifeOS. Some are straightforward, such as ChatGPT with scheduled tasks. Others involve Obsidian templates and manually curated knowledge bases—something I explored in an earlier article on knowledge management. Enhanced versions, like Compass, integrate AI into an existing setup.

However, none of these perfectly met my needs. While ChatGPT can remind me of tasks, I prefer not to store my life records within a chat. Obsidian templates are versatile, but for automated updates or automatic daily to-do collection from my phone, I needed to build that functionality myself. I didn't want to install a full suite of tools and clutter my machine.

My goal was to create a lean, dependable, and flexible system that could adapt as my habits evolved. What I envision is a system that allows me to log simple occurrences like noting a meal or recording project progress quickly. At night, the system can compile these logs into a daily review and update project notes. I should be able to interact with this system from my phone or work computer, while the majority of my personal data remains on the server at home.

Additionally, I should be able to convert source material into Wiki notes and use those notes for reference while writing. During the day, the system should pull tasks from my existing projects, consider recent health records when suggesting training or rest, and integrate into my TickTick list. The state of projects should be stored in the Vault.

A particularly useful feature is when I mark tasks in TickTick as done, dropped, or postponed; the system can be notified of this change. The evening review can then utilize this information to update relevant projects, eliminating the need to rewrite project files after each checkbox is ticked during the day. Each evening, the system should also generate a summary and daily report.

My system comprises two main pathways for data and work. During the day, logging an event triggers an event creation, which is then captured and stored in SQLite. Later, the system reads relevant records and Vault files to compile a daily review and update project documents. n8n can automate these tasks on a schedule or through webhooks.

The separation of these processes is crucial, as daytime logging needs to be lightweight, while heavier tasks occur at night. For the agent layer, I chose OpenClaw, which offers multiple model connections, a dedicated chat interface, and seamless integration with n8n. Logging a single event doesn't require loading all project, health, and Wiki rules.

When I ask the agent to write a review or a knowledge note, it loads the relevant skill, then writes the content. This design allows me to ask the agent for specific tasks without overloading it with unnecessary information. While ChatGPT's custom MCP also provides a conversational AI experience, it operates on a different network path, ensuring my interactions remain secure and efficient.

The core of LifeOS consists of two paths: a fast path for everyday logging, and a slower path for more complex tasks. The fast path involves the agent understanding my input, executing the corresponding Python script note.py to write to SQLite, retrieving an ID, and replying. I avoid having the model generate SQL queries itself.

If I need to edit an entry, I can use the edit command followed by the ID. If I simply state that I've completed a task, it's registered as a new completion event, not an edit to an existing entry. While task status could be stored in SQLite, I prefer to keep it as a log of events rather than a live table of unfinished tasks. Long-term progress, project goals, and task lists are stored in Markdown files in the Vault.