Frequently asked questions
Everything you need to know about TensorLay.
What is TensorLay?
TensorLay is a Windows desktop application that connects your home GPU to a remote VPS via a secure SSH tunnel. This allows AI agents running on the VPS (like Claude Code) to use your local GPU for tasks like image generation, LLM inference, text-to-speech, and more.
Is it really free?
Yes. TensorLay is free and open source under the MIT license — no trials, no paid tiers, no feature gates. If the app is useful to you, you can support development on GitHub Sponsors, but it's entirely optional.
What GPU do I need?
TensorLay requires an NVIDIA GPU with CUDA support. Most modern NVIDIA cards (GTX 1060 and newer) will work. More VRAM means you can run larger models — 8 GB is a good starting point, 16 GB is recommended.
What AI services are supported?
TensorLay currently supports 6 services: SD Forge (Stable Diffusion image generation), ComfyUI (node-based image pipelines), Ollama (LLMs), AllTalk TTS (text-to-speech and voice cloning), MusicGen (AI music), and TripoSR (3D model generation). More services are planned.
Do I need a VPS?
Yes. TensorLay is designed to bridge your local GPU to a remote VPS. You need a Linux VPS where your AI agents or workflows run. Any provider works — DigitalOcean, Hetzner, Linode, AWS, etc.
Is my connection secure?
Yes. All traffic goes through an encrypted SSH tunnel. SSH keys are generated locally and never leave your machine — they are not synced to any cloud account. Pairing codes are one-time use.
Do I need to configure port forwarding?
No. TensorLay uses reverse SSH tunneling, so no router configuration or port forwarding is needed. It works behind NAT and firewalls out of the box.
Can I use it with multiple VPS servers?
Yes. Connect to as many VPS servers as you want — there are no limits.
What's the optional account for?
Registering an account is optional. It lets you sync non-sensitive settings (VPS host/port/user, install directory, UI preferences) across machines. SSH private keys and pairing codes are never synced — they stay only on the device that generated them.
What operating systems are supported?
The desktop app runs on Windows 10 and Windows 11. The VPS relay component runs on any Linux distribution with Python 3 installed.
Can I read or modify the source code?
Yes. TensorLay is MIT-licensed. The source is on GitHub at github.com/papergallery/tensorlay. Pull requests and forks are welcome.