Hylon AI Assistant
Hylon AI Assistant
Your intelligent companion inside OrbVPN. Hylon is a hybrid AI: a private model that runs directly on your device for instant, offline answers, backed by a powerful cloud engine for complex questions, with human support one tap away when you need it.
What is Hylon AI?
Hylon is the AI assistant built into the OrbVPN app. Internally the engine is called OrbAI; in the app you meet it as Hylon. It helps you troubleshoot connection issues, choose the right server and protocol, understand security features, and get answers about your account, all through a natural chat interface.
What makes Hylon different from a typical support bot is its architecture. Hylon is hybrid. A compact AI model runs entirely on your device for the most common questions, so you get answers instantly and privately, even with no internet. When a question needs deeper reasoning or live information, Hylon escalates to a cloud engine powered by Azure OpenAI. And when only a person will do, it hands off to a human support agent with full context.
On-Device First
A small TensorFlow Lite model and a local knowledge base answer common questions directly on your device. Fast, private, and available offline.
Cloud When Needed
Complex or unusual questions are handled by a cloud engine running Azure OpenAI, with answers streamed back to you in real time.
Human Handoff
If confidence is low or you ask for a person, Hylon escalates to a human agent and passes along a summary so you never repeat yourself.
One Assistant, Three Tiers
You never have to choose a tier. Hylon decides automatically: it tries the on-device model first, reaches for the cloud when a question is harder, and offers a human agent when that is the right call.
How Hylon Works
Every message you send flows through Hylon's three tiers in order. Most questions are resolved at the first tier without ever leaving your phone.
On-Device Understanding
Hylon detects what you are asking about (a connection problem, slow speed, billing, your account, protocols, or censorship bypass) and answers using the on-device model and your locally stored knowledge base.
Confidence Check
Hylon scores how confident it is in the on-device answer. If confidence is high, you get an instant local reply. If it falls below the threshold, Hylon moves up a tier.
Cloud Reasoning
For harder questions, Hylon connects to the Azure OpenAI cloud engine over a secure WebSocket and streams the answer back word by word, grounded in OrbVPN's help knowledge base.
Human Escalation
When the issue is sensitive (refunds, cancellations, account deletion), when sentiment turns negative, or when you simply ask, Hylon hands the conversation to a human agent along with an AI-written summary and your diagnostics.
The On-Device Layer
The heart of Hylon is a TensorFlow Lite model that ships inside the OrbVPN app and runs locally. It is paired with an offline knowledge base of OrbVPN help articles, synchronized from the OrbVPN help center, that lives on your device.
Instant Answers
Because inference runs on your own hardware, common questions are answered immediately, with no round trip to a server.
Works Offline
The on-device model and local knowledge base keep working even when you have no connection, which matters most in restricted networks.
Private by Design
Questions answered on-device stay on your device. Nothing about them is sent to OrbVPN servers to produce the reply.
Intent Aware
Hylon recognizes the type of request and offers the right next action, such as retrying a connection, switching servers, or enabling mimicry.
What Intent Detection Means
Hylon classifies each message into a topic such as connection issue, slow speed, billing, account, protocol, or blocked region. That topic drives both the answer and the suggested actions you see, like Switch Server, Change Protocol, or Enable Mimicry for censored regions.
The Cloud Layer
When a question is beyond the on-device model, Hylon connects to OrbVPN's cloud AI engine. This layer is built on Azure OpenAI and adds deeper reasoning, broader knowledge, and live, streamed responses.
Real-Time Streaming
Cloud answers stream back over a secure WebSocket and appear word by word as they are generated, so you start reading immediately instead of waiting for a full reply.
Grounded in OrbVPN Knowledge
The cloud engine uses retrieval over OrbVPN's documentation so answers reflect how OrbVPN actually works, not generic guesses.
Six Languages
The cloud assistant understands and replies in English, Persian, Arabic, Russian, Chinese, and Turkish, tuned for OrbVPN's global community.
Suggested Actions
Responses can include one-tap actions the app can perform for you, such as switching server, changing protocol, resetting the connection, or clearing DNS.
Resilient Connection
The cloud chat keeps itself alive with periodic heartbeats and reconnects automatically if your network drops mid-conversation, so a brief signal loss does not lose your place.
The Human Layer
Some problems need a person. Hylon never traps you in a loop with a bot.
Smart Escalation
Hylon hands off to a human when it is not confident, when a request is sensitive such as a refund or account deletion, when sentiment turns negative, or whenever you ask for a person.
Full Context Carried Over
An AI-generated summary of your conversation and your VPN diagnostics travel with the handoff, so the agent already understands your situation.
Routed to the Right Agent
Live chat routes you to an available agent matched by topic and language, with your place in the queue shown while you wait.
Live and Transparent
Human chat supports typing indicators and read receipts, so the conversation feels immediate and you always know it is being seen.
What Hylon Can Help With
Hylon is tuned specifically for OrbVPN and online security. It is at its best with the topics it was trained to recognize.
Connection Help
Diagnose why a connection failed or keeps dropping, with guided steps to retry, switch servers, or change protocols.
Speed and Performance
Understand why speeds drop and what to try, from choosing a closer server to switching to a faster protocol.
Bypassing Censorship
Get guidance for restricted regions, including VLESS with mimicry, plain-TCP transports, and region-optimized servers for Iran, Russia, and China.
Account and Billing
Find answers on subscriptions, payments, device limits, and account settings, with escalation to a human for sensitive changes.
Example Conversations
Speed Optimization
'My VPN slows down on the US server. Which protocol and server give the fastest speeds from Europe?'
Staying Safe
'Is public Wi-Fi safe with OrbVPN connected, and which protocol should I use for the strongest protection?'
Censorship Bypass
'I am in Iran and nothing connects. How do I use VLESS with mimicry to get back online?'
Troubleshooting
'My VPN disconnects every few minutes on Android 14 with WireGuard. How can I fix it?'
Be Specific
The more detail you give, the better Hylon can help. Instead of "why is my VPN slow", try "my speed drops on the US server using VLESS over Wi-Fi". Specifics let Hylon pick the right answer and the right suggested action.
Honest About Capabilities
Hylon is genuinely useful, and we want you to know exactly what it does so you can trust it.
Focused, Not General-Purpose
The on-device model is a compact assistant trained on OrbVPN support topics. It excels at OrbVPN and security questions rather than acting as an open-ended general chatbot.
Knowledge-Grounded
Answers draw on OrbVPN's real help content. When the on-device model is unsure, the cloud engine adds deeper reasoning grounded in the same knowledge base.
Context From Diagnostics
With your permission, Hylon can read your current VPN status, such as whether you are connected and which server and protocol you are using, to make its suggestions relevant.
Always a Path to a Human
If Hylon cannot resolve something, it does not guess endlessly. It tells you and offers a human agent.
Privacy
Hylon is built so that the most sensitive processing happens on your device, and so you always know where your data goes.
On-Device Stays On-Device
Questions that the local model answers are processed entirely on your phone. They are not transmitted to OrbVPN to produce the reply.
Escalations Use the Cloud
When a question is escalated to the cloud engine or to a human agent, that conversation is sent to OrbVPN's backend and stored so we can answer you and keep continuity, in line with our privacy policy.
Privacy-Preserving Feedback
If you rate an answer to help improve Hylon, your feedback is anonymized by default, aggregated on your device, and shared as statistics only, never as raw messages.
You Stay in Control
From OrbAI settings you can choose your feedback privacy level, turn anonymous analytics off, and clear your stored data at any time.
Transparency by Design
We are deliberately clear about this: instant on-device answers never leave your device, while cloud and human escalations are processed and stored on OrbVPN servers to deliver the help you asked for. See the AI Settings page to manage exactly what is kept.
Helping Hylon Get Smarter
Hylon improves over time with your feedback, and it does so without compromising your privacy.
Rate an Answer
After a response, you can mark it helpful or not. That single signal is the most valuable thing you can give Hylon.
Anonymized Locally
Your rating is stripped of identifying details and protected with privacy-preserving techniques on your device before anything is shared.
Aggregated, Not Individual
Ratings are combined on your device over time and shared only as aggregate statistics, so no single message or person can be singled out.
Everyone Benefits
These aggregate signals show us which answers work and which to improve, making Hylon better for the whole community.
The Road to Decentralised AI
Hylon today is a hybrid support assistant. Its longer-term vision, set out in the OrbAI technical specification, is far bigger: a privacy-first, decentralised intelligence that the OrbVPN community helps build and owns.
A Roadmap, Not Today's Product
The capabilities in this section describe where OrbAI is heading. They are a forward-looking roadmap. The hybrid on-device plus cloud assistant described above is what ships in the app today.
A Network of Personal AIs
The vision runs a Personal AI on each OrbMesh node that learns locally from its owner's data. A global model then learns from the network of Personal AIs, never from raw user data.
Two-Layer Privacy
Raw data stays on the user's device with the Personal AI. Only learned patterns, never the underlying content, are shared with the network, keeping the global model two layers removed from personal data.
Federated Learning
Devices share only encrypted model updates, never data. Updates are compressed and aggregated hierarchically across the network, with no single point of failure.
Safety-Bound Self-Improvement
A self-improvement engine can extend the system's capabilities, but only within an immutable foundation of Truth, Consistency, Honesty, Safety, and Integrity, with human oversight for significant changes.
Earn by Powering the Network
The decentralised vision is also an economy. OrbMesh is designed so that people who contribute compute and connectivity are rewarded.
Run a Node, Earn Rewards
OrbMesh devices are designed to host a Personal AI and contribute to the network. Node operators are the largest share of the planned token rewards, turning spare compute into a decentralised data center.
Community-Owned Intelligence
Instead of one company owning a giant central cluster, intelligence is distributed across many user-owned nodes, with the community sharing in the value it creates.
Why This Matters
This is how OrbVPN intends to build powerful AI without harvesting personal data: keep raw data on the owner's device, learn only from patterns, reward the people who provide the infrastructure, and bind the whole system to safety rules that cannot be edited away.
Meet Your On-Device AI Advisor
Hylon is built into OrbVPN: instant, private answers on your device, deeper help from the cloud, and a real person when you need one. Always available, always on your side.