Life Mentor

LifeMentor is a local-first AI life coach that turns your handwritten values and continuous phone/watch data into daily, personalized micro-adjustments with Gemini 3 Flash.

Comment

Inspiration

Most people are not short on advice; they are short on sustained attention to their real life. Work and daily chores consume energy, and it becomes hard to answer: What do I actually want? What is my current state? What should I adjust next?

Generic AI usually sees only the current prompt, often without stable priorities or user feedback loops. We wanted to build an AI mentor that stays with the user over time.

What it does

LifeMentor is a local-first AI life coach that combines:

  • User-handwritten Values / Goals / Projects in Markdown
  • Low-friction personal state data (currently phone + smartwatch)
  • Daily feedback loops (morning micro-adjustment, daytime records, nightly reflection)

The system aligns suggestions with the user's primary value, while still offering optional inspirations from other values so guidance is multi-dimensional, not one-axis.

How we built it

We built a FastAPI backend connected to Gemini 3 Flash (gemini-3-flash) through the OpenAI-compatible endpoint, and an Obsidian-based frontend for a markdown-native experience.

Gemini is central in four places:

  1. Alignment analysis: pattern + value_board from longitudinal data
  2. Morning micro-adjustment: one concrete value-aligned suggestion
  3. Record parsing: structure free-form user logs into usable context
  4. Evening reflection: summarize the day from records + task completion + auto-collected signals, then adapt next-day suggestion difficulty

All outputs are stored as user-owned, editable Markdown/JSON.

Challenges we ran into

The hardest part was balancing personalization, privacy, and friction:

  • Keeping data useful without making users do heavy manual logging
  • Turning multi-source signals into actionable, value-aligned suggestions
  • Designing a loop that is adaptive but not overwhelming

What we learned

A persistent loop beats one-off chat. When users define their own values and the AI continuously learns from daily signals, guidance becomes specific, practical, and behavior-changing.

What's next

Right now we start with phone and smartwatch signals. Next, we will introduce more digital and physical-world data from different devices, expand agent interactions in the record workspace, and let background agents continuously push value/goal execution forward while keeping data ownership fully user-first.

Built With

Share this project:

Updates