What I Built

Every time my friend Archi and I tried to make a plan, it went the same way:

"Where do you want to eat?"
"I don't mind. You pick."
"No, you pick."

Twenty minutes later we'd still be going back and forth. It wasn't a big problem, but it happened all the time, and it quietly took the fun out of making plans.

So I built Dozee for Archi: a personal AI decision buddy that plans everything. Food, cafés, date ideas, movies, late-night plans, even what to wear. Tell it your mood, your budget and how many people are coming (or just type what's on your mind) and it gives you a real pick with a reason and a couple of backups. No more "you decide."

My favorite part is the Cute Nudge 💌. After Dozee suggests a place or a plan, I can send Archi a sweet message about that exact suggestion, like "you two should try this place," either right away or as a reminder to send a few days later.

Stop debating. Start vibing.

Code

alakaloops / dozee-ai

🔮 Dozee AI — Personal AI Decision Assistant

Your Mood. Your Budget. Your Next Move.

Dozee is a 100% local, privacy-first personal AI assistant built to solve decision paralysis. Powered by Next.js 16, Tailwind CSS, Framer Motion, and Ollama + Gemma 3 4B, Dozee helps you decide what to eat, watch, wear, or do—without sending any of your private data to external servers.

✨ Features

🤖 100% Local AI Execution: Connects to local Ollama API using Gemma 3 4B / Ollama models.
🎨 Personal Assistant UI: Sleek dark obsidian glassmorphism aesthetic inspired by modern AI assistant apps.
🍕 Interactive Suggestion Chips: Quick prompt chips for Food, Movies, Going Out, Date Ideas, Late Night, Celebrations, Outfits, and Surprises.
💌 Friend Nudge Integration: Generate cute custom nudge messages, copy/share links, download .ics calendar files, or schedule local browser reminders.
⚡ Local Persistence: Saves…

View on GitHub

How I Built It

Open-weight model: Gemma 3 4B, running locally through Ollama. All inference happens on my own laptop.

App: Next.js, TypeScript and Tailwind CSS, with Framer Motion for the animations.

How the AI fits in: a Next.js API route sends the user's mood, budget, group size and activity to Gemma through Ollama's /api/chat endpoint. The reply becomes a result card with a headline pick, a reason and alternatives. The Cute Nudge uses the same local model to write a short message about the real recommendation, which can be shared through WhatsApp, saved as a calendar reminder or scheduled in the app.

Local-first: preferences and scheduled nudges are stored in the browser's localStorage, so nothing is saved on a server or sent to a cloud AI.

DEV Community

plans for two, picked by dozee