Put your computer's power to work for AI
The Onysoft AI Gateway Node desktop app runs open-source language models on your own graphics card. You use these models free of charge from the chat in your Onysoft dashboard and with your API keys. Participation is by application; individuals and organisations apply separately.
How does it work?
Four steps from application to your first chat.
Apply
Fill in the individual or corporate application form, then read and accept the program documents. The form is in your dashboard and takes a few minutes.
Approval and download
The Onysoft team reviews your application. Once approved, a download screen for Windows and macOS opens in your dashboard; you are also notified by e-mail.
Link your device
Install the app, click "Sign in with my Onysoft account" and confirm the 8-character code shown in your browser. Your password is never entered into the app; the app opens the connection and no ports are opened on your computer.
Run models
The app suggests models that fit your hardware, downloads the files directly from their source and verifies them with SHA256. Use a ready model in the dashboard chat or via the API with the node/ prefix.
What does the program offer?
Your own AI server
Install models that fit your hardware from open-source families such as Qwen, Gemma, Mistral, Phi, Granite and DeepSeek, right inside the app.
Free usage
Requests sent to your own device are not charged to your balance. It works with the dashboard chat and the OpenAI-compatible API using your existing keys.
You stay in control
Choose a sharing intensity of Low, Balanced or High and switch sharing off whenever you like. You decide which models are installed.
Future-ready infrastructure
The program is built so that approved devices may one day contribute capacity to the Onysoft network. Today, no other customer's requests are sent to any device; if this feature is enabled, a separate addendum and your explicit consent will be required.
Individual or corporate?
Both application types use the same program. The difference is who applies and the scale of the hardware operated.
| Topic | personIndividual | apartmentCorporate |
|---|---|---|
| Who it is for | People who want to run models on their own computer: developers, students and enthusiasts | Companies, agencies, universities and research teams; anyone operating several workstations or servers |
| Information requested | Full name, contact details, province, device, graphics card and internet connection details | Company title, tax office and number, address, authorised contact, hardware and hosting details |
| Typical scale | Usually 1–3 personal computers | Several workstations or servers |
| Identity verification | No national ID number is requested; the verified e-mail and phone on your account are used | The tax identification number is validated with its checksum algorithm |
| Accepted documents | Individual Participation Agreement, GPU Sharing Program Privacy Notice (KVKK), Acceptable Use Policy, Technical Conditions and Security Policy | Corporate Participation Agreement, GPU Sharing Program Privacy Notice (KVKK), Acceptable Use Policy, Technical Conditions and Security Policy |
| Review | Applications are reviewed in order | Additional information or a technical call may be requested |
System requirements
The app measures your hardware itself and suggests suitable models. Please check the following before you start.
desktop_windowsWindows
- Windows 10 or Windows 11, 64-bit
- An NVIDIA graphics card (CUDA) is recommended; AMD and Intel discrete cards are supported experimentally via Vulkan
- Without a suitable graphics card, CPU mode is available on processors with AVX2; it is slower and for your own use only
- Free disk space from a few GB to tens of GB, depending on the models you choose
laptop_macmacOS
- A Mac with Apple Silicon (M1 or later); Intel-based Macs are not supported
- macOS 13.3 Ventura or later
- At least 8 GB of unified memory; models run on the GPU with Metal
- Free disk space from a few GB to tens of GB, depending on the models you choose
network_checkInternet connection
- To enable sharing: at least 10 Mbps download, 5 Mbps upload and at most 150 ms latency
- The speed test runs only against Onysoft servers; no third-party speed test service is used
- Model files are large; an unlimited or high-quota connection is recommended for the first setup
Security and privacy
The app is designed to run without exposing your computer to the internet.
No open ports
The app only makes encrypted (HTTPS) outbound connections to the Onysoft server. Your computer does not accept inbound connections; no router or firewall changes are needed.
Password-free pairing
The device is linked to your account with a one-time code you confirm in the browser. The device key is stored in the macOS Keychain or in the Windows encrypted store (DPAPI); the server keeps only its hash.
Verified downloads
The runtime is downloaded from the official llama.cpp release on GitHub and models from Hugging Face, directly to your computer, and verified with SHA256.
Your requests stay yours
Requests you send to node/ models run only on your own device. Today no other customer's requests are routed to any device; devices in CPU mode never receive customer requests.
Transparent data processing
A hardware summary, app version, connection and speed test results, model states and request records (request and response contents, token counts, duration) are kept on the server to operate the service. Event logs are deleted after 90 days. See the privacy notice for details.
Electricity and heat
Running models uses your graphics card heavily, consumes electricity and warms up your computer. You choose the intensity and can switch sharing off at any time.
Program documents
During the application you read and accept the documents below. Each document shows its version and effective date; the version you accept is recorded.
Frequently asked questions
Does it cost anything to join?
No. The program is free, and requests you send to models running on your own device are not charged to your balance.
Will I be paid for sharing?
There is currently no revenue sharing or payment under the program. If such a model is introduced, its terms will be announced in a separate addendum and will apply only to participants who accept it.
Will other people's requests run on my computer?
No. Today only your own requests and short test requests that Onysoft sends to check that the device works correctly run on your device. Routing customer requests to devices is switched off; enabling it would require a separate addendum and your explicit consent.
When will my application be decided?
Applications are reviewed in the order received. You can follow the status on the "My GPU Sharing" page of your dashboard and you will also be notified by e-mail. If more information is needed, you can update the form and resubmit it.
Can I join without a graphics card?
On Windows the app also runs in CPU mode on processors with AVX2 and can run small models for your own use. This mode is slower than a graphics card.
Why do I see a security warning during installation?
The macOS build is signed with an Apple Developer ID belonging to Onysoft Veri Merkezi A.Ş. and notarized by Apple; on first launch you only get the standard “downloaded from the internet” confirmation. The Windows build is not code-signed yet, so SmartScreen may show a warning (More info → Run anyway). You can verify the SHA256 checksum of both files on the download screen.
How do I leave the program?
You can switch sharing off with the button in the app at any time, sign out in the app to unlink the device from your account, and uninstall the app. To end your participation or for requests about your data, write to info@onysoft.com.
How are updates delivered?
When a new version is released it appears on the download screen in your dashboard and the app notifies you. For security and compatibility, sharing is paused on old versions that are no longer supported; after updating you continue where you left off.
Ready to apply?
The application form is in your dashboard and takes a few minutes. If you do not have an account yet, simply create a free one first.
You will be taken to the sign-in page; after signing in, the application form opens directly.