# Zenith

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Bharanidharan V](https://cerebralvalley.ai/u/Bharanidharanv26)
- **GitHub:** https://github.com/bharanidharanv26/LifeOS.git
- **Demo video:** https://youtu.be/vKNjyoiRJlY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4404

LifeOS is a personal AI work assistant that remembers how you work, plans around it, asks before it acts, and learns from what you decide.

I built it because every AI assistant I use forgets me. I have to explain my habits again each time: no meetings before 11am, keep emails short, protect Friday afternoons. And when assistants do take actions, they rarely check with me first.

LifeOS works in a simple loop: Remember, Plan, Ask, Act, Learn.

You type a goal like "Prepare for my Friday investor call." LifeOS recalls what it knows about you, then proposes a short plan: a prep block on the calendar, an agenda email draft, and a note. Each step shows which memories shaped it. Every step is a card you can approve, edit, or reject (with a reason), and nothing runs until you say so.

Trust is built into the design. A rule-based policy checker, not the AI model, blocks any step that breaks one of your strong preferences, so the model can't overrule you. For example, a 9:00 AM block gets flagged if you've said no meetings before 11.

The part I'm most proud of is the learning. When I reject that 9:00 AM block with "too early" and edit the email draft, LifeOS reflects on those decisions and writes new, traceable memories, such as "avoid morning blocks." Then I give it a new goal, "Get ready for next week's board meeting," and the new plan avoids mornings on its own. A side-by-side Plan v1 vs v2 view shows exactly what changed and which memories caused it.

You stay in control throughout. You can view, edit, delete, or export everything LifeOS knows about you, and a full "forget everything" option wipes it.

Tech: React, TypeScript, Node.js, Express, SQLite, and Zod-validated structured model output. The Meta model handles goal understanding, plan generation, and reflection through a single adapter, and every model call is visible in a Model activity panel.

To be transparent: email drafts, calendar blocks, and notes are local simulations by design. Nothing is sent to real email or calendar services. The next step would be real Gmail and Calendar integration, and running open-weight models locally so personal memory never has to leave the user's device.

## More from Meta Global AI Developer Hackathon

- [Curato](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4401)
- [jaguar](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4402)
- [The Magic Table](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4403)
- [Innovian](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4405)
- [Stardive](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4406)
- [Ishan's Team](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4407)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4404. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt