FRIEND.OS: An AI Memory That Remembers What Was Waiting for What

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built FRIEND.OS, a local-first AI memory system for a close friend who tends to remember individual things but sometimes forgets the connection between them.

The problem is simple:

“I’ll do X after Y happens.”

We remember X.

We remember Y.

But when Y finally happens, we may forget that X was waiting for it.

FRIEND.OS is designed to catch that missing connection.

For example:

“I’ll submit my internship application after I receive my certificate.”

Later:

“I received my certificate today.”

FRIEND.OS recognizes that the certificate was the dependency for the internship application and proactively surfaces:

Something just clicked

You can now submit your internship application.

But FRIEND.OS also handles the opposite situation: when you are about to do something before completing what it depends on.

For example:

“I’ll eat dessert after I finish rice.”

Then:

“I’ll eat dessert now.”

FRIEND.OS can detect that the dependent action is being attempted before its prerequisite has been satisfied and surface:

You're forgetting something…

Eat rice first.

You said you'd eat dessert after finishing rice.

So FRIEND.OS isn't just storing memories. It tracks the relationships and state between memories.

The core idea is:

MEMORY → CONNECTION → REALIZATION

FRIEND.OS doesn't wait for you to ask what you forgot. It notices when something you've been waiting for finally happens — or when you're about to skip something you said you needed to do first.

I built it around a real problem experienced by someone close to me rather than starting with a generic AI chatbot and searching for a use case afterward.

Demo

Demo video: https://youtu.be/M7XgCTY7Xig

Windows download: https://github.com/kusalar/FriendOS/releases/tag/v1.0.0

The demo shows two sides of FRIEND.OS.

1. Something just clicked

I create:

“I'll submit my internship application after I receive my certificate.”

Then:

“I received my certificate today.”

FRIEND.OS identifies the new evidence, connects it to the waiting commitment, and proactively surfaces the next action.

2. You're forgetting something…

I create:

“I'll eat dessert after I finish rice.”

Then:

“I'll eat dessert now.”

FRIEND.OS detects that the dependent action is being attempted before its prerequisite has been satisfied.

It surfaces:

You're forgetting something…

Eat rice first.

The user can also inspect why FRIEND.OS produced the insight and see the relevant memory relationship.

The project also includes a 60-second Demo Mode with three realistic scenarios:

• Internship application
• VLSI project report
• Research internship

Code

GitHub: https://github.com/kusalar/FriendOS

FRIEND.OS is structured as a local web application with:

• Next.js
• React
• TypeScript
• Tailwind CSS
• FastAPI
• SQLite
• Ollama
• Qwen3-4B
• Deterministic dependency matching

A packaged Windows installer and portable build are also provided through the GitHub Release.

How I Built It

The AI at the core of FRIEND.OS is Qwen3-4B, running locally through Ollama.

The architecture is:

text
Browser
   ↓
Next.js / React / TypeScript
   ↓
FastAPI
   ↓
SQLite
   ↓
Qwen3-4B via Ollama
   ↓
Structured Memory
   ↓
Deterministic Dependency Engine
   ↓
Proactive Insight