Skip to content

thatcooperguy/nvremote

BranchesTags

Repository files navigation


NVRemote

Stream your GPU-powered desktop to anything, anywhere.

Website · Downloads · Docs · API Status

Version NVIDIA WebRTC Built with Claude

This is a personal passion project built collaboratively by a human developer and AI (Claude by Anthropic). Every line of code across 7 languages, 6 platforms, and a production cloud backend was pair-programmed in iterative sessions. This project is not created, sponsored, endorsed, or affiliated with NVIDIA Corporation. See Legal Disclaimer below.

Downloads

Grab the latest release for your platform. No account required — just download and run.

🖥️ Windows — One App, Client + Host

Download Notes
NVRemote-0.5.1-beta-Setup.exe Includes client AND host. Stream to or from this machine. Auto-updates.

On Windows, NVRemote is one unified app. Go to Settings to switch between Client, Host, or Both mode. No separate host agent needed.

📱 Other Client Platforms

Platform Download Notes
macOS NVRemote-0.5.1-beta-universal.dmg Universal binary (Intel + Apple Silicon)
Linux x86_64 nvremote_0.5.1-beta_amd64.deb Recommended — sudo dpkg -i (Debian/Ubuntu)
Linux x86_64 NVRemote-0.5.1-beta-x86_64.AppImage Portable AppImage — chmod +x and run
Linux ARM64 nvremote_0.5.1-beta_arm64.deb Recommended — ARM64 Debian/Ubuntu package
Linux ARM64 NVRemote-0.5.1-beta-arm64.AppImage Portable ARM64 AppImage
Android NVRemote-v0.5.1-beta.apk Android 7.0+, sideload APK

🖧 Linux Host Agent (headless servers)

For headless Linux servers (Jetson, DGX, Docker) where Electron doesn't make sense, use the standalone Go agent:

Platform Download Notes
Linux x86_64 NVRemoteHost-v0.5.1-beta-linux-amd64.tar.gz GPG signed (.asc)
Linux ARM64 NVRemoteHost-v0.5.1-beta-linux-arm64.tar.gz Jetson / DGX Spark — GPG signed (.asc)

📋 SHA256 checksums · All releases

The Pitch

You have a beefy NVIDIA GPU sitting in your device. You want to use it from your couch, your phone, or a browser tab halfway across the world.

NVRemote installs a tiny agent on your GPU machine and streams the entire desktop — with hardware-accelerated encoding — to any device over a peer-to-peer WebRTC connection. Low latency. Adaptive quality.

Your RTX 3090 at home  ----WebRTC P2P---->  Your laptop at a coffee shop
Your Jetson Orin       ----WebRTC P2P---->  Your phone on the train
Your DGX Spark at work ----WebRTC P2P---->  A Chrome tab anywhere

What Makes It Different

Problem NVRemote
VPN + RDP looks terrible P2P WebRTC with NVIDIA hardware encode at up to 240fps
Existing solutions are one-size-fits-all 7 streaming profiles: Competitive gaming to CAD precision
Limited to desktops Native apps for Win, Mac, Android, Linux ARM + Chrome browser
No edge GPU support First-class Jetson Orin and DGX Spark support
Complex setup Install agent, sign in, click "Stream"

Platforms

Stream FROM (Host)

  • Windows 10/11 with any NVIDIA GPU (GeForce, Quadro, RTX)
  • Linux x86_64 with NVIDIA GPU
  • NVIDIA Jetson Nano / Xavier NX / Orin Nano / Orin NX / AGX Orin
  • NVIDIA DGX Spark (Grace Blackwell GB10)

Stream TO (Client)

  • Windows (Electron)
  • macOS (Electron + Swift/Metal)
  • Linux x86_64 & ARM64 (Electron AppImage)
  • Android 7.0+ (Kotlin/Compose)
  • Chrome browser (zero-install WebRTC)

Streaming Specs

The QoS engine runs a continuous control loop, adapting in real-time:

What Range
Resolution 720p → 8K (matches your display)
Frame Rate 30 → 240 fps
Bitrate 2 → 300 Mbps (adaptive)
Codecs H.264, HEVC, AV1 (auto-selected)
Color YUV 4:2:0 or 4:4:4 (for color-critical work)
Latency target sub-frame on LAN, network-dependent on WAN

Profiles

Profile What It's For Resolution FPS
Competitive Esports, fast shooters 1080p 240
Balanced General gaming & desktop 1440p 144
Cinematic Single-player, movies up to 8K 60
Creative Photo/video editing (4:4:4) up to 8K 60
CAD SolidWorks, Fusion 360 (4:4:4) up to 8K 60
Mobile Saver Phone on cellular 720p 60
LAN Same network, max everything up to 8K 240

Architecture

                    Your Device                              Your GPU Machine
               ==================                      =======================

          [NVRemote Client]                        [nvremote-host]
           Electron / Android /                      NvFBC capture
           Chrome / Swift                            NVENC encode (H.264/HEVC/AV1)
           Hardware video decode                     Opus audio encode
                  |                                         |
                  |      WebRTC P2P (encrypted)             |
                  |  <-------- Video/Audio/Input -------->  |
                  |                                         |
                  +------- STUN/TURN fallback -------+      |
                                                     |      |
                                              [NVRemote API]
                                               NestJS on GCP Cloud Run
                                               OAuth + JWT auth
                                               Signaling relay
                                               Session management
                                                     |
                                              [PostgreSQL 15]

Connection Modes

Mode How Latency When
P2P Direct STUN NAT traversal Lowest Default for most networks
TURN Relay NVRemote relay server Low Auto-fallback for strict NAT
WireGuard VPN Encrypted tunnel Medium Corporate firewalls
Zero-Trust Per-session auth proxy Medium Enterprise compliance

Repo Map

nvremote/
├── apps/
│   ├── server-api/          # NestJS API         (TypeScript)
│   ├── website/             # Next.js site        (TypeScript)
│   ├── client-desktop/      # Electron app — client + host (TypeScript + C++)
│   ├── android/             # Android client      (Kotlin)
│   ├── mac-client/          # macOS client        (Swift)
│   ├── host-agent/          # Host agent — Linux headless (Go)
│   └── gateway/             # TURN + VPN gateway  (Go)
├── libs/
│   ├── nvremote-host/       # GPU capture + encode (C++17)
│   ├── nvremote-viewer/     # Video decoder addon  (C++17)
│   └── nvremote-common/     # Shared networking    (C++17)
├── infra/
│   └── terraform/           # GCP IaC
└── .github/workflows/       # CI (6 jobs) + Release (13 jobs)

7 languages: TypeScript, C++17, Go, Kotlin, Swift, SQL, HCL

Tech Stack

Technology
API NestJS 10, Prisma 5, PostgreSQL 15, Socket.io
Website Next.js 14, React 18, Tailwind CSS, Framer Motion
Desktop Electron 31, React 18, Zustand, C++ N-API addon (unified client + host)
Android Kotlin, Jetpack Compose, Hilt, WebRTC
macOS Swift 5.9, Metal, VideoToolbox
Host Agent Go 1.22
Streaming C++17, NvFBC, NVENC SDK, OpenSSL, Opus
Infra GCP Cloud Run, Cloud SQL, Terraform
CI/CD GitHub Actions (13-job release pipeline)

NVIDIA ARM Platform Support

Platform GPU Encode Host Client
Jetson Nano Maxwell 128-core H.264 Yes Yes
Jetson Xavier NX Volta 384-core H.264, HEVC Yes Yes
Jetson Orin Nano Ampere 1024-core H.264, HEVC Yes Yes
Jetson AGX Orin Ampere 2048-core H.264, HEVC Yes Yes
DGX Spark Blackwell H.264, HEVC, AV1 Yes Yes

Security

  • JWT auth on every endpoint (default-closed)
  • Org-scoped data isolation
  • Token rotation on refresh
  • 30-min idle timeout, 24-hr max session
  • Rate limiting, Helmet headers, strict CORS
  • Non-root Docker containers
  • GPG-signed Linux releases
  • DTLS/SRTP encrypted streams

Getting Started

Stream your GPU desktop in 3 steps. Your host and sessions are private to your account — no one else can see or connect to your machine.

Windows (one app does everything)

  1. Download NVRemote-0.5.1-beta-Setup.exe and install it
  2. Sign in with Google
  3. To stream FROM this machine: Go to Settings → switch to Host or Both mode → follow the setup wizard to register
  4. To stream TO this machine: Your hosts appear in the Dashboard — click Stream

That's it. One app. Client, host, or both at the same time.

Linux Host (headless / Jetson / DGX)

tar xzf NVRemoteHost-v0.5.1-beta-linux-amd64.tar.gz
cd NVRemoteHost-v0.5.1-beta-linux-amd64
sudo ./install.sh    # Installs as systemd service, prompts for token

Or run directly without installing: ./NVRemoteAgent (interactive setup on first run)

Other Clients (macOS, Android, Linux desktop)

Download from Downloads, sign in with the same Google account, and your hosts will appear.

🔒 Privacy: Your host is registered to your account and is only visible to you (or members of your organization if you create one). Sessions are encrypted end-to-end with DTLS/SRTP. No one — not even NVRemote servers — can see your stream content.

Building from Source

git clone https://github.com/thatcooperguy/nvremote.git
cd nvremote && npm install

# API server
cd apps/server-api && npm run start:dev

# Website
cd apps/website && npm run dev

# Desktop client
cd apps/client-desktop && npm run dev

# Host agent (Go)
cd apps/host-agent && go build -o NVRemoteAgent ./cmd/agent

Release Pipeline

Push a tag → 13 jobs build everything:

C++ native libs → Go host agents (Win/Linux/ARM64) → Host bundles → Electron installers (Win/Mac/Linux/ARM64) → Android APK → GitHub Release + GCS upload + SHA256 checksums + GPG signatures

Current Status

Component Status
API (NestJS) Live on Cloud Run
Website (Next.js) Live on Cloud Run
Desktop Client (Electron) Builds in CI
Android Client (Kotlin) Builds in CI
macOS Client (Swift) Skeleton
Web Client (Chrome) Built — WebRTC in browser
Host Agent (Go) Builds in CI (Linux + ARM64 headless)
C++ Streaming Core Builds in CI
Infrastructure GCP, Terraform, CI/CD all live

How This Was Built

This entire project — from bare-metal C++ GPU capture to a production Kubernetes-free cloud backend — was built through human-AI pair programming.

The human brings product vision, hardware expertise, architecture decisions, and real-world testing on NVIDIA GPUs.

The AI (Claude by Anthropic) brings rapid multi-language implementation, consistent patterns across a massive codebase, and the ability to context-switch between C++ memory management and Kotlin Compose animations in the same session.

The result: a multi-platform GPU streaming system spanning 7 programming languages, 6 target platforms, 13 CI/CD jobs, and production infrastructure on GCP — built iteratively, one feature at a time.

Legal Disclaimer

NVRemote is an independent, personal project. It is not created, sponsored, endorsed, or affiliated with NVIDIA Corporation or any of its subsidiaries. The author is employed by NVIDIA but developed this project entirely on personal time, using personal equipment, personal cloud infrastructure, and personal funds. This project does not represent NVIDIA in any way, and no proprietary NVIDIA information, internal tools, or confidential resources were used in its development.

Trademark Notice: "NVIDIA," "GeForce," "Quadro," "RTX," "Jetson," "DGX," "NVENC," "NvFBC," and related marks are trademarks or registered trademarks of NVIDIA Corporation. "WebRTC" is a trademark of Google LLC. All other trademarks are property of their respective owners. Use of these names is for identification purposes only and does not imply endorsement.

No Warranty: This software is provided "as is," without warranty of any kind, express or implied. See the MIT License for full terms. Use at your own risk.

License

This project is licensed under the MIT License.

Built with Claude by Anthropic

About

No description, website, or topics provided.

Resources

Readme

License

MIT license

Contributing

Contributing

Security policy

Security policy
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

6 tags

Packages

 
 
 

Contributors

Languages