I Built an Open-Source AI Wardrobe App. It's Now on iPhone.
I've wanted this app to exist for years.
Every morning, it's the same problem for everyone. Staring at a closet full of clothes, thinking "I have nothing to wear." I know there are good outfits in there somewhere but I can't see them. I check the weather on one app, mentally filter what's clean, what fits the occassion, what I haven't already worn three days in a row. All before coffee.
I kept waiting for someone to build the thing I wanted: an AI that knows my wardrobe, checks the weather, and just tells me what to wear. Something I could self-host. Something that doesn't sell my data to advertisers or lock basic features behind a paywall.
So after years of this small idea sitting in the back of my head, and after getting sidetracked by a dozen other things, I finally sat down and built it myself.
Wardrowbe is now live on the App Store, the open-source repo is growing, and the Android app is weeks away.
How It Started
The first version of Wardrowbe was honestly just for my wife. She always keep saying "I have nothing to wear"... which is ... not true because we have wardrobe full of her clothes lol. So, I thought, what if an app could actually solve that? Not with generic fashion advice from Pinterest, but with suggestions from clothes she actually owns. So I built it.
She loved it. Started using it daily, which is rare for anything I build. That was the signal that this might be more than a side project.
I posted about it on Reddit at some point, in a thread where people were discussing wardrobe apps. The response caught me off guard. People started asking for features, wash tracking being one of the most requested ones. "I want to know when something needs washing based on how many times I've worn it." Made total sense. So I built that in.
Then more people started reaching out. Some wanted analytics on what they actually wear. Others wanted family sharing so they could coordinate outfits with their partner. Feature by feature, what started as a quick thing for my wife turned into a full wardrobe management platform.
What Wardrowbe Does Today
You photograph your clothes. AI tags everything: type, color, material, pattern, formality, fit. No dropdowns, no manual entry. Point your camera, tap, done.
Every morning the app checks your local weather and suggests a complete outfit from your actual wardrobe. Rain coming? No suede. Feels like 30 degrees? Breathable layers. You pick the occasion (casual, office, date, formal) and the AI puts something together that makes sense for it.
Accept the suggestion or skip it. Either way, the AI learns. Over weeks it builds a style profile that reflects how you dress, not how some fashion algorithm thinks you should.
| Feature | What it does |
|---|---|
| AI tagging | Snap a photo. Type, color, material, formality detected in seconds |
| Weather-aware outfits | Daily suggestions adapted to your local forecast |
| Virtual try-on | See how clothes look on you before getting dressed |
| Style learning | AI adapts to your feedback. Accepts, skips, corrections all count |
| Wardrobe analytics | Track what you wear, what you skip, spot gaps and redundancies |
| Family sharing | Share outfits with household members, rate each other's looks |
| Smart pairings | AI learns which items work together from your feedback |
The web app has been running since launch. The iPhone release puts the full experience in your pocket with the same features, same AI, same sync.
iOS Is Live, Android Is Coming
The app hit the App Store on March 5th. Not a companion app, the full thing. Everything you can do on the web, you can do from your phone.
The Android build is in final testing. Same codebase (React Native via Expo), same features, same backend. Play Store release is coming very soon.
If you're on Android and can't wait, the web app works fine on mobile browsers. But the native app with push notifications for daily outfit suggestions, camera intergration, and haptic feedback is worth holding out for.
Virtual Try-On: The Feature That Almost Didn't Ship
Virtual try-on was always on my list. People were asking for it too, and I was keen to figure it out. But generating a realistic image of a garment on someone's body is genuinely hard. Lighting, proportions, fabric draping, pose preservation. Get it wrong and it just looks like a bad Photoshop job.
I spent weeks testing different workflows in ComfyUI. Most of them were not great. Some were decent but way too slow. Then I found a combination that actually produced accurate results consistently.
These are actual raw outputs from the ComfyUI workflows I was testing. The garment gets transferred onto the person while preserving their pose, body proportions, and the original lighting. It genuinely works.
During user testing, the accuracy surprised me. Upper body, lower body, or full outfit. All three categories produce useable results. What people actually end up doing with it is they try on AI-suggested outfits before commiting. Instead of pulling five shirts from the closet, they preview combinations on their phone and walk to the closet knowing what to grab. It turns the morning outfit routine from a guessing game into a 2-minute decision.
Three monthly credits come with the cloud subscription. But here's what I'm most proud of: if you have your own ComfyUI instance, you can connect it directly to Wardrowbe. Your hardware, zero credits consumed, unlimited generations. I didn't want to gatekeep a feature behind credits when people already have the infrastructure to run it themselves. The virtual try-on deep dive covers photo tips and provider options.
For the Community: It's Open Source, Fork It
This is the part I care about the most.
I know I'm not the only one who wanted this app. If you browse Reddit, Hacker News, or any dev community, you'll find people trying to build thier own wardrobe organizer. Vibe coding it over a weekend with ChatGPT. Getting the basic CRUD working, maybe some AI tagging, then hitting a wall on the hard parts. Outfit recommendation logic, weather integration, learning from user feedback, image processing pipelines. These things just take time to get right.
I've been there. The idea is simple. The execution is not.
The open-source version exists and holds nothing back. Full AI tagging, outfit suggestions, weather integration, analytics, family sharing, notifications. All of it. No crippled "community edition." No features locked behind a license key.
Deploy it with Docker Compose on a Raspberry Pi, NAS, laptop, or VPS. Bring your own LLM via Ollama or any OpenAI-compatible API. Your data stays on your machine.
If you've been building your own wardrobe app and got stuck somewhere, take a look at the repo. Use it as a base. Rip out the parts you don't need, add the parts you do. The recommendation engine, the AI tagging pipeline, the weather integration. Those are the hard problems that took me months to get right. You don't have to solve them again from scratch.
For details on running your own instance, the self-hosted setup guide covers hardware requirements, configuration, and LLM options. The entire stack runs on a Raspberry Pi 5 with 8GB RAM.
PRs are welcome. Issues get responses.
Cloud That Only Knows Your Clothes
Not everyone wants to maintain a server though. The cloud version exists for exactly that reason.
I want to be clear about what Wardrowbe's cloud stores and what it doesn't. There are no passwords in the system. None. Authentication is handled through OIDC and passkeys (biometric login via Face ID, fingerprint, or PIN). I don't have your credentials because they literally don't exist in my system. There's nothing to leak even if someone tried.
What the cloud stores is your wardrobe. Photos of your clothes, tags, outfit history, preferences. Just your clothes.
The one exception is virtual try-on. That feature needs a photo of you to generate the preview. You decide if you want to use it or not. If you do, those photos are stored in encrypted storage buckets. Not used for training. Not shared with anyone. They exist so the feature works, and they're gone when you delete them.
Cloud pricing is $10/month ($9/month if you go yearly). There's a 7-day free trial with no credit card required.
Read more about how Wardrowbe handles your data.
The Stack
- Backend: FastAPI, async SQLAlchemy, PostgreSQL, Redis + arq workers
- Frontend: Next.js 14, TanStack Query, shadcn/ui
- Mobile: React Native (Expo SDK 54), NativeWind, Expo Router
- AI: Any OpenAI-compatible API. Ollama for local, OpenAI for cloud
- Auth: OIDC, Apple Sign-In, Google Sign-In, passkeys. No passwords anywhere
- Try-on: Multi-provider. Replicate, Fashn, ComfyUI, RunPod, or bring your own
- Infra: k3s on AWS (cloud), Docker Compose (self-hosted)
What's Next
- Android release. Play Store submission coming very soon
- Smarter recommendations. Deeper learning signals, better outfit scoring
- More try-on providers. Expanding backend support
- Whatever the community builds. The repo is open, PRs are welcome
Try It
iPhone: Download from the App Store.
Self-host: Clone from GitHub, docker compose up, done in 5 minutes.
Cloud: Start a free trial at app.wardrowbe.com. No credit card, 7 days, full access.
I built the app I wanted to use every morning. Maybe you wanted it too.