Hugging Face has become the GitHub of machine learning — over 900,000 models, 200,000 datasets, and a community that ships faster than most startups. But here's the thing: there's no single place to see what's actually happening across the ecosystem.
That's why I built HuggingFaceOS, an open-source operational dashboard that tracks the Hugging Face ecosystem in real time.
The Problem
If you work with ML models, you've probably experienced this:
- A new model drops on Hugging Face and you hear about it 3 days later on Twitter
- You're trying to pick between 15 different fine-tunes of Llama and have no idea which ones are actually gaining traction
- Your team is using a dataset that got updated last week and nobody noticed
- You want to know what's trending in the MLOps space but have to manually browse dozens of pages
Hugging Face's website is built for discovery of individual assets. It's not built for operational awareness of the ecosystem.
What HuggingFaceOS Does
The dashboard gives you a single-pane view of what matters:
Trending Models
Track which models are gaining stars and downloads right now — not yesterday's leaderboard, but live momentum. See emerging fine-tunes before they hit the top of HF trending.
Dataset Activity
Monitor new and updated datasets across categories. If you're training models, knowing when a dataset you depend on gets updated is critical.
Community Pulse
Aggregate signals from Spaces, discussions, and paper implementations. When a new paper drops and people start building on it, you'll see the activity spike.
Weekly Digest
Get a curated summary of the most important changes — new model architectures, dataset releases, trending Spaces, and community milestones. Sign up at the dashboard to receive it in your inbox every Monday.
How I Built It
The stack is intentionally simple:
- Express.js backend — handles API aggregation, caching, and serves the dashboard
- PostgreSQL — stores historical snapshots for trend analysis
- Hugging Face API — pulls model metadata, download counts, and community activity
- Cron jobs — scheduled data collection every few hours to build trend lines
The key architectural decision was snapshot-based tracking. Instead of just showing current state (which HF already does), we store periodic snapshots. This lets us calculate:
- Download velocity (downloads/day trend, not just total)
- Star acceleration (is this model gaining momentum or plateauing?)
- New model detection (first appearance in our tracking)
What I Learned About the ML Ecosystem
Building this taught me a few things:
1. The ecosystem moves insanely fast. In any given week, there are 500+ new models uploaded. Most are fine-tunes, but the signal-to-noise ratio matters. Automated tracking isn't optional — it's necessary.
2. Downloads don't tell the whole story. A model with 1M downloads might be plateauing while a model with 10K downloads is doubling every day. Velocity matters more than absolute numbers.
3. The Hugging Face API is solid but underdocumented. The model info endpoint gives you everything you need, but figuring out rate limits and pagination took some trial and error.
4. Developers want operational intelligence, not just search. Every ML engineer I talked to had the same reaction: "Why doesn't this already exist?" The answer is that Hugging Face focuses on being a platform, not an analytics tool. There's room for both.
Try It Out
The dashboard is live at huggingfaceos.polsia.app/dashboard.
Here's what you can do:
- Browse trending models and datasets
- See momentum indicators (rising/falling/stable)
- Sign up for the weekly digest email — a curated summary of what happened in the HF ecosystem that week
It's free and open source. If you work with Hugging Face models or datasets, this saves you hours of manual browsing every week.
What's Next
A few things on the roadmap:
- Custom watchlists — track specific models or orgs you care about
- Alerts — get notified when a model you use gets updated or when a new competitor appears
- API access — so you can integrate ecosystem data into your own workflows
- Deeper Spaces analytics — track which demos and apps are gaining traction
If any of this sounds useful, check out the dashboard and let me know what features would be most valuable for your workflow.
Built with Express.js, PostgreSQL, and the Hugging Face API. Follow @huggingfaceos for updates.
Top comments (0)