Open Research Platform

About Voidly

The censorship intelligence layer for the open internet

...
Contributors
99.8%
Classifier F1
16
Global Nodes
...
Uptime

What We Provide

Ground truth for internet censorship — structured, verified, ML-classified data that turns raw network measurements into actionable intelligence.

  • 16 globally distributed probe nodes generating proprietary measurement data
  • 5,356 verified censorship incidents with evidence chains
  • ML classifier (99.8% F1) trained on 37K labeled samples
  • Shutdown forecast model (74.6% AUC) — 7-day risk predictions across 119+ countries
  • Multi-source correlation: our probe network + OONI + CensoredPlanet + IODA
  • Real-time VPN accessibility testing across major providers
  • 10-year historical archive (1.6M records, 119+ countries)

Research & Open Data

Mission

3.2 billion people live under internet censorship. Traditional monitoring is too slow— blocks are detected hours or days after they happen.

We measure censorship continuously using AI trained on ... measurements from 16 nodes across 6 continents.

Research contribution model: You use our free tools. Your anonymous performance data trains our AI. Everyone gets better censorship evasion.

Free Tools

Privacy

What we collect (for AI training):
  • Connection success/failure rates
  • Node performance metrics
  • Censorship pattern detection
What we never collect:
  • Browsing history or DNS queries
  • Real IP addresses (stripped after routing)
  • Personally identifiable information

Exit nodes run in RAM-only mode. Even if seized, no user data exists.

Verify Everything

Open Source
MIT licensed. Fork it, audit it, run your own nodes.
github.com/voidly-ai →
Blockchain Proofs
Zero-knowledge privacy proofs. AI training attestations. All on Base L2.
View contracts →
Warrant Canary
0 warrants received. Updated daily. Cryptographically signed.
Verify status →

Technical

AI Models
  • • Censorship classifier: 99.8% F1 (GradientBoosting)
  • • Shutdown forecast: 74.6% AUC (XGBoost)
  • • <50ms inference time
  • • Federated learning (privacy-preserving)
Infrastructure
  • 16 nodes, 6 continents
  • 99.8% uptime (self-healing)
  • • WireGuard + Stealth mode
  • • Zero manual intervention

Join ... contributors

Use the tools. Contribute data. Explore the research.