βββ βββ βββββββ ββββββββββ βββ βββ βββ
βββ ββββββββββββββββββββββββββ ββββ ββββ
βββ ββββββ βββββββββ ββββββ βββββββ
ββββ βββββββ βββββββββ ββββββ βββββ
βββββββ βββββββββββββββββββββββββββββββ
βββββ βββββββ ββββββββββ βββββββββββ
AI-powered VPN that learns to defeat censorship
Experimental β’ Open Source β’ Privacy-First
Free access. No signup. No logs. A smarter way to fight network restrictions.
We're building an AI that learns from real-world censorship attempts to keep you connected.
The Exchange:
- You get: Free VPN, AI-optimized routing, privacy protection
- We get: Anonymous telemetry to train the AI
- The world gets: A tool that gets smarter at defeating censorship every single day
No profit motive. No data selling. Research-first.
- Website: https://voidly.ai
- Get VPN: https://voidly.ai/vpn
- Status: https://voidly.ai/status
- Metrics: https://voidly.ai/metrics
- AI Deep Dive: https://github.com/voidly-ai/voidly/blob/main/docs/AI_DEEP_DIVE.md
- API Docs: https://github.com/voidly-ai/voidly/blob/main/docs/API.md
- Architecture: https://github.com/voidly-ai/voidly/blob/main/docs/ARCHITECTURE.md
- GitHub Org: https://github.com/voidly-ai
- X (Twitter): https://x.com/Voidly_ai
- Telegram: https://t.me/+4laJpSoUooY5MWEx
USER (Iran) β Requests VPN Connection
β
API β Intelligence Service: "Best route for Iran?"
β
XGBoost Model Analyzes:
β’ Censorship intensity (HIGH)
β’ Recent block events (DPI/RST patterns)
β’ Node performance (health, latency, load)
β’ Historical success rates (per country/node)
β’ Time-of-day patterns
β’ Protocol selection (obfuscate? TLS wrapping?)
β
AI Predicts: Amsterdam + WireGuard-over-TLS + Port 443
Confidence: 95% | Reasoning: "High censorship detected; HTTPS mimicry optimal"
β
Config Generated β User Connects β β SUCCESS
β
Telemetry (anonymous) β Training Data Updated
β
Next Model β Even Smarter
Current stats:
- Model: XGBoost v2
- Training data: 8,429+ samples
- Accuracy: 92% AUC-ROC (target: 95%+)
- Inference time: ~47ms
- Features: 13 dimensions (location, threat level, node health, time patterns)
13 Active Exit Nodes β’ 5 Continents β’ <50ms Average Latency
π EUROPE (3) π NORTH AMERICA (5) π ASIA (3)
ββ Frankfurt, Germany ββ New York, USA ββ Tokyo, Japan
ββ London, UK ββ Chicago, USA ββ Singapore
ββ Amsterdam, Netherlands ββ Silicon Valley, USA ββ Bangalore, India
ββ Toronto, Canada
π OCEANIA (1) ββ Mexico City, Mexico π SOUTH AMERICA (1)
ββ Sydney, Australia ββ SΓ£o Paulo, Brazil
Smart routing:
- Real-time health monitoring (CPU, latency, bandwidth)
- Automatic fallbacks (seamless switching)
- Obfuscation (WireGuard-over-TLS/443) for high-censorship regions
- Geographic optimization (proximity when relevant)
- Threat-adaptive selection (protocols/ports based on risk)
Uptime: 99.8% β’ Status: https://voidly.ai/status
We collect (anonymous, aggregate):
- Connection success/failure and performance metrics
- Node health (CPU, load, reachability, bandwidth)
- Censorship events by country (aggregated patterns)
We never collect:
- Browsing history
- Per-user DNS queries
- Post-connection IP addresses
- Personally identifiable information
Nodes operate privacy-first:
- RAM-only posture (no persistent traffic logs)
- Ephemeral session configs (discarded after disconnect)
- Zero-knowledge architecture (we cannot see your traffic)
Example telemetry record (aggregate only):
ββ Connection Event βββββββββββββββββββββ
β Country: [IR] β
β Node: [AMS] β
β Outcome: SUCCESS β
β Latency: 45ms β
β Protocol: wireguard_obfs β
β Censorship: DPI_DETECTED β
βββββββββββββββββββββββββββββββββββββββββ
β Used for training; never tied to a person
For users:
- Visit https://voidly.ai/vpn
- Generate keys and a config (client-side)
- Scan QR code with WireGuard
- Connect β
Requirements:
- WireGuard app (iOS/Android/Desktop). That's it.
Frontend (voidly.ai)
- Next.js 14 (App Router), TypeScript, Tailwind
- RainbowKit + Wagmi (Web3 wallet connection)
- Vercel/Cloudflare Pages
API (api.voidly.ai)
- Cloudflare Workers (TypeScript)
- D1 (SQLite) for sessions, R2 for config backups
- SIWE (wallet authentication), ethers.js (NFT verification)
Intelligence (intelligence.voidly.ai)
- Node.js + Python FastAPI
- XGBoost v2 (production), TFJS/ONNX (planned)
- TimescaleDB (time-series telemetry), Redis (jobs/cache)
VPN Infrastructure
- WireGuard (protocol), Nginx (reverse proxy + SSL)
- 13 global exit nodes
- Python agents (telemetry + health checks)
- Systemd (service orchestration)
Web3 (optional premium tier)
- Base (Ethereum L2)
- Hardhat + Solidity
- OpenZeppelin (ERC-721)
Deep dive: https://github.com/voidly-ai/voidly/blob/main/docs/AI_DEEP_DIVE.md
We welcome contributions in:
- Traffic obfuscation and DPI evasion
- ML research (features, retraining, drift detection)
- Agent/node automation, monitoring, ops
- Security reviews and pen testing
- UX improvements for non-technical users
Quick start:
git clone https://github.com/voidly-ai/voidly.git
cd voidly
make install
make test
Workflow:
- Fork β Branch β Commit β PR with description
Resources:
- Contributing Guide: https://github.com/voidly-ai/voidly/blob/main/CONTRIBUTING.md
- Quickstart: https://github.com/voidly-ai/voidly/blob/main/QUICKSTART.md
- Architecture: https://github.com/voidly-ai/voidly/blob/main/docs/ARCHITECTURE.md
- API Docs: https://github.com/voidly-ai/voidly/blob/main/docs/API.md
Phase 2: Global Expansion (40% complete)
- 13 global nodes across 5 continents
- XGBoost ML model v2 (92% accuracy)
- 8,429+ training samples (grow to 20K+)
- AI-powered routing in production
- Traffic obfuscation research (GAN shaping) in progress
Phase 3: Decentralization (planned)
- Protocol obfuscation (HTTPS mimicry, TLS wrapping)
- GAN-based traffic shaping (production-ready)
- Community-run nodes, on-chain registry
- P2P mesh networking
- 50+ nodes globally
Phase 4: Autonomy (research)
- Continuous learning vs live censors
- Real-time DPI fingerprint adaptation
- Decentralized governance (DAO)
- Fully autonomous network
Dashboards:
- Network Status: https://voidly.ai/status
- Metrics: https://voidly.ai/metrics
- Intelligence Health: https://intelligence.voidly.ai/health
Current performance snapshot:
ββ NETWORK STATUS βββββββββββββββββββββββ
β Nodes Online: 13/13 β β
β Avg Latency: 47ms β
β Success Rate: 95%+ β
β Uptime: 99.8% β
β Active Sessions: [LIVE] β
β AI Model: XGBoost v2 β
β Training Samples: 8,429+ β
βββββββββββββββββββββββββββββββββββββββββ
Found a vulnerability?
- Critical: security@voidly.ai (PGP available on request)
- Non-critical: open an issue at https://github.com/voidly-ai/voidly/issues
We donβt run paid bounties (research project), but valid reports receive public credit.
MIT License (open source, permissive). Use it, fork it, improve it. No warranty.
EXPERIMENTAL SOFTWARE β USE AT YOUR OWN RISK
- Active research project; expect occasional breakage
- Not audited by third-party security firms (yet)
- No guarantees or warranties of any kind
For proven anonymity: Use Tor Browser + Tails OS
For advancing censorship evasion research: Use Voidly
Why Free?
- Privacy is a human right. Paywalls create a two-tier internet.
Why Open Source?
- Trust requires transparency. Audit it. Break it. Improve it.
Why AI?
- Censors adapt. Static VPNs get blocked. ML trained on real attempts stays ahead.
Why Telemetry?
- You canβt fight what you canβt measure. Data is anonymous, aggregate, performance-only.
Vision
- An AI that learns faster than censors can adapt. A network thatβs unstoppable.
Use it: Every connection trains the AI
Share it: Help people in censored regions
Build it: Contribute code, research, ideas
Break it: Find bugs, report issues
Fork it: Start your own network
[SYSTEM_STATUS]
ββ NODES: 13/13 OPERATIONAL β
ββ AI: LEARNING π§
ββ UPTIME: 99.8% π‘
ββ MISSION: ACTIVE π
END_TRANSMISSION