SlipStream

Low-latency remote desktop streaming over WebRTC with NVIDIA hardware-accelerated encoding.

Overview

SlipStream captures the Windows desktop using the Windows Graphics Capture API and encodes frames in real-time using NVIDIA NVENC (H.264, H.265, or AV1). Video and audio are streamed to web browsers via WebRTC data channels with DTLS encryption. The browser client renders frames using WebGL2 and decodes video/audio using the WebCodecs API.

Architecture

+------------------------------------------------------------------------------+
|                              SERVER (Windows)                                |
+------------------------------------------------------------------------------+
|  Screen Capture         Encoder             WebRTC              Audio        |
|  +--------------+    +--------------+    +--------------+    +----------+    |
|  | WGC API      |--->| NVENC        |--->| Data Channel |<-->| WASAPI   |    |
|  | D3D11 Fence  |    | AV1/H265/264 |    | libdatachan  |    | Opus     |    |
|  | 6-tex pool   |    | FFmpeg       |    |              |    | 96 kbps  |    |
|  +--------------+    +--------------+    +--------------+    +----------+    |
|                                                 |                            |
|                          +----------------------+----------------------+     |
|                          |       HTTPS Server (port 443)               |     |
|                          |    cpp-httplib + OpenSSL                    |     |
|                          |    JWT Auth + Rate Limiting                 |     |
|                          +----------------------+----------------------+     |
+---------------------------------------------+--------------------------------+
                                              |
                          WebRTC (UDP 50000-50020)
                                              |
+---------------------------------------------+--------------------------------+
|                              CLIENT (Browser)                                |
+------------------------------------------------------------------------------+
|  +--------------+    +--------------+    +--------------+    +----------+    |
|  | VideoDecoder |--->| WebGL2       |    | AudioDecoder |--->| Worklet  |    |
|  | HW or SW     |    | Letterbox    |    | Opus         |    | RingBuf  |    |
|  +--------------+    +--------------+    +--------------+    +----------+    |
|                                                                              |
|  Input Handler -----> Mouse/Keyboard -----> Data Channel -----> Server       |
+------------------------------------------------------------------------------+

Features

• Triple Codec Support - AV1, H.265, and H.264 via NVIDIA NVENC
• Hardware Decoding - WebCodecs API with automatic HW/SW fallback
• System Audio - WASAPI loopback capture with Opus encoding
• Multi-Monitor - Live monitor switching with tabbed UI option
• Clipboard Sync - Bidirectional text clipboard (Ctrl+C/V)
• Relative Mouse - Pointer lock mode for games and full-screen apps
• Cursor Sync - Remote cursor shape updates (13 standard cursor types)
• Session Auth - PBKDF2-SHA256 password hashing with JWT tokens
• Rate Limiting - Brute-force protection with IP lockout
• Auto SSL - Self-signed certificate generation on first run
• Debug Stats - Real-time performance overlay

Requirements

Server

Component	Requirement
OS	Windows 10/11 64-bit (build 1903+)
GPU	NVIDIA GPU with NVENC support
IDE	Visual Studio 2022 with C++20 Desktop workload
Package Manager	vcpkg

Client

Component	Requirement
Browser	Chrome 94+, Edge 94+, Firefox 98+
APIs	WebRTC, WebCodecs (VideoDecoder, AudioDecoder), WebGL2, AudioWorklet

Quick Start

1. Install vcpkg

git clone https://github.com/microsoft/vcpkg.git C:\vcpkg
cd C:\vcpkg
.\bootstrap-vcpkg.bat
.\vcpkg integrate install

2. Build

build.bat

Output: build\bin\Release\SlipStream.exe

3. Run

run.bat

On first run:

• Prompts for username (3-32 alphanumeric characters, underscores, hyphens)
• Prompts for password (8+ characters with at least one letter and one digit)
• Generates self-signed SSL certificate (server.crt, server.key)
• Saves hashed credentials to auth.json

4. Connect

Open browser to https://<HOST_IP>:443 and log in with your credentials.

Note: Self-signed certificate will trigger a browser warning. Click through to proceed.

Network Configuration

Port	Protocol	Purpose
443	TCP	HTTPS server and WebRTC signaling
50000-50020	UDP	WebRTC media transport

Firewall

netsh advfirewall firewall add rule name="SlipStream HTTPS" dir=in action=allow protocol=tcp localport=443
netsh advfirewall firewall add rule name="SlipStream WebRTC" dir=in action=allow protocol=udp localport=50000-50020

Security

Password Storage

Parameter	Value
Algorithm	PBKDF2-HMAC-SHA256
Iterations	600,000
Salt	16 bytes (random)
Key Length	32 bytes

Passwords are never stored in plain text. Only the salt and derived hash are saved in auth.json.

Rate Limiting

Threshold	Action
5 failed attempts in 15 min	30-minute IP lockout
Successful login	Clears attempt counter

JWT Sessions

Parameter	Value
Algorithm	HS256
Issuer	slipstream
Expiry	24 hours
Storage	HttpOnly, Secure, SameSite=Strict cookie

SSL Certificates

Auto-generated on first run:

• 2048-bit RSA key
• 10-year validity
• Subject Alt Names: localhost, 127.0.0.1, 0.0.0.0

To use custom certificates, replace server.crt and server.key before starting.

API Endpoints

Endpoint	Method	Auth	Description
/	GET	No	Web client HTML
/styles.css	GET	No	Stylesheet
/js/*.js	GET	No	JavaScript modules
/api/auth	POST	No	Login with {username, password}
/api/session	GET	Cookie	Validate current session
/api/logout	POST	No	Clear session cookie
/api/offer	POST	Cookie	WebRTC SDP offer exchange

Video Pipeline

Capture

Component	Detail
API	Windows Graphics Capture (WGC)
Texture Pool	6 textures with D3D11 fence sync
Frame Buffer	4-frame ring buffer
Cursor	Optional capture in stream
Border	Disabled (if OS supports)

Encoding

Setting	H.264	H.265	AV1
Encoder	h264_nvenc	hevc_nvenc	av1_nvenc
Preset	P1 (fastest)	P1 (fastest)	P1 (fastest)
Tune	Ultra-low latency	Ultra-low latency	Ultra-low latency
Rate Control	VBR	VBR	VBR
CQ Level	23	25	28
Keyframe Interval	2 seconds	2 seconds	2 seconds
B-Frames	0	0	0
Color Space	BT.709	BT.709	BT.709
Color Range	Full (JPEG)	Full (JPEG)	Full (PC)

Bitrate Formula: 0.18085 × width × height × fps bps

Example: 1920×1080 @ 60fps ≈ 22.5 Mbps

Transport

Parameter	Value
Chunk Size	1400 bytes max (1379 payload + 21 header)
Header Size	21 bytes
Video Buffer	256 KB threshold
Max Queue	3 frames worth of chunks
Delivery	Unreliable, unordered (UDP semantics)

Video Packet Header (21 bytes)

Offset	Size	Field	Description
0	8	timestamp	Capture timestamp (microseconds)
8	4	encodeTimeUs	Encode duration (microseconds)
12	4	frameId	Frame sequence number
16	2	chunkIndex	Current chunk index
18	2	totalChunks	Total chunks in frame
20	1	frameType	1=keyframe, 0=delta

Audio Pipeline

Capture

Parameter	Value
API	WASAPI Loopback
Mode	Shared
Sample Rate	48,000 Hz (resampled if system differs)
Channels	Stereo (max 2)
Frame Duration	10 ms (480 samples)

Encoding

Parameter	Value
Codec	Opus
Application	Restricted Low Delay
Bitrate	96 kbps
Complexity	3
Signal Type	Music
FEC	Disabled
DTX	Disabled

Transport

Parameter	Value
Audio Buffer	128 KB threshold
Max Queue	8 packets
Delivery	Unreliable, unordered

Audio Packet Header (16 bytes)

Offset	Size	Field	Description
0	4	magic	0x41554449 ("AUDI")
4	8	timestamp	Capture timestamp
12	2	samples	Sample count (480)
14	2	dataLength	Opus payload size

Client Playback

Component	Detail
Decoder	AudioDecoder (WebCodecs)
Processor	AudioWorklet with ring buffer
Buffer Capacity	9600 samples (200ms)
Prebuffer Threshold	2400 samples (50ms)
Target Buffer	3360 samples (70ms)
Max Buffer	6720 samples (140ms)

Input Protocol

Mouse Move Absolute (12 bytes)

Offset	Size	Field
0	4	magic (0x4D4F5645)
4	4	x (float 0.0-1.0)
8	4	y (float 0.0-1.0)

Mouse Move Relative (8 bytes)

Offset	Size	Field
0	4	magic (0x4D4F5652)
4	2	dx (int16)
6	2	dy (int16)

Mouse Button (6 bytes)

Offset	Size	Field
0	4	magic (0x4D42544E)
4	1	button (0-4)
5	1	action (1=down, 0=up)

Button mapping: 0=left, 1=right, 2=middle, 3=X1, 4=X2

Mouse Wheel (8 bytes)

Offset	Size	Field
0	4	magic (0x4D57484C)
4	2	deltaX (int16)
6	2	deltaY (int16)

Keyboard (10 bytes)

Offset	Size	Field
0	4	magic (0x4B455920)
4	2	keyCode (JS keyCode)
6	2	scanCode
8	1	action (1=down, 0=up)
9	1	modifiers

Modifiers: Ctrl=1, Alt=2, Shift=4, Meta=8

Clipboard Data (8 + N bytes)

Offset	Size	Field
0	4	magic (0x434C4950)
4	4	length
8	N	UTF-8 text (max 1MB)

Cursor Shape (5 bytes)

Offset	Size	Field
0	4	magic (0x43555253)
4	1	cursorType (0-13, 255)

Cursor types: 0=default, 1=text, 2=pointer, 3=wait, 4=progress, 5=crosshair, 6=move, 7=ew-resize, 8=ns-resize, 9=nwse-resize, 10=nesw-resize, 11=not-allowed, 12=help, 13=none, 255=custom

Rate Limits (Server-Enforced)

Type	Max per Second
Mouse moves	500
Clicks	50
Keystrokes	100

Blocked Inputs

• Windows key (left/right)
• Ctrl+Alt+Delete

WebRTC Data Channels

Channel	Ordered	MaxRetransmits	Purpose
control	Yes	3	Commands, ping/pong, monitor list
video	No	0	Video frame chunks
audio	No	0	Audio packets
input	Yes	3	Mouse/keyboard events

Client Modules

File	Purpose
state.js	Shared state, constants, codec detection, metrics, clock sync
network.js	HTTP auth, WebRTC signaling, data channel handlers
renderer.js	WebGL2 rendering with aspect ratio letterboxing
media.js	VideoDecoder, AudioDecoder, AudioWorklet processor
input.js	Mouse/keyboard capture, RAF batching, pointer lock
ui.js	Settings panel, fullscreen, tabbed mode, stats overlay

Clock Synchronization

Parameter	Value
Ping interval	200 ms
Sample count	8
Offset calculation	Median filter
RTT estimation	Median of samples

Frame Handling

Parameter	Value
Max buffered frames	8
Frame timeout	200 ms
Max frame age (jitter)	50 ms
Decode queue limit	6 frames

UI Features

Settings Panel

Access by clicking the right edge of the screen:

• Logout - Disconnect and clear session
• Fullscreen - Enter fullscreen with keyboard lock (Escape captured)
• Audio - Toggle system audio playback
• Monitor - Switch between displays
• Frame Rate - 15/30/60/120/144 or custom (1-240)
• Codec - AV1, H.265, or H.264 (shows HW/SW status)
• Tabbed Mode - Monitor tabs at top of screen
• Debug Stats - Real-time performance overlay
• Relative Mouse - Pointer lock mode for gaming
• Clipboard Sync - Enable Ctrl+C/V synchronization

Tabbed Mode

When enabled, displays a tab strip showing all monitors with:

• Monitor name (friendly name from EDID)
• Resolution
• Primary indicator (star icon)

Debug Stats Overlay

Real-time metrics display:

Section	Metrics
Throughput	FPS (actual/target/efficiency), bitrate, resolution, codec
Latency	RTT, frame age
Jitter	Frame interval mean and std dev
Decode	Average decode time, queue size, HW acceleration status
Render	Average render time, frame count
Network	Packet count (total/video/audio), average packet size
Drops	Dropped, timeout, late frames, decode errors
Audio	Packets received/decoded, buffer health, underruns, overflows
Input	Mouse moves, clicks, keystrokes
Session	Uptime, total frames, total data

Server Threads

Thread	Priority	Purpose
Main	Above Normal	HTTPS server, process priority
Encoder	Time Critical	Frame encoding and sending
Audio	Highest	Audio capture and encoding
Cursor	Below Normal	Cursor shape polling

Dependencies

Managed via vcpkg (vcpkg.json):

Package	Purpose
libdatachannel	WebRTC implementation
cpp-httplib[openssl]	HTTPS server
nlohmann-json	JSON parsing
opus	Audio encoding
openssl	Cryptography, SSL/TLS
ffmpeg[nvcodec,avcodec]	Video encoding
jwt-cpp	JWT token handling
picojson	Required by jwt-cpp

Building

Standard Build

build.bat

Installer Build

build_installer.bat

Creates SlipStream-1.0.0-win64.exe (NSIS) or .zip (fallback).

File Structure

SlipStream/
├── include/
│   ├── common.hpp      # Common includes, types, auth utilities
│   ├── capture.hpp     # Screen capture with WGC
│   ├── encoder.hpp     # NVENC video encoding
│   ├── webrtc.hpp      # WebRTC server implementation
│   ├── audio.hpp       # WASAPI audio capture + Opus
│   └── input.hpp       # Input handling + clipboard
├── src/
│   └── main.cpp        # Application entry point
├── client/
│   ├── index.html      # Web client HTML
│   ├── styles.css      # Stylesheet
│   └── js/
│       ├── state.js    # Shared state and metrics
│       ├── network.js  # WebRTC and HTTP
│       ├── renderer.js # WebGL2 rendering
│       ├── media.js    # Video/audio decoding
│       ├── input.js    # Input capture
│       └── ui.js       # UI controls
├── vcpkg.json          # Dependencies
├── CMakeLists.txt      # Build configuration
├── build.bat           # Build script
├── build_installer.bat # Installer build script
├── run.bat             # Run script
└── LICENSE.txt         # License

Troubleshooting

Problem	Solution
vcpkg not found	Set VCPKG_ROOT environment variable or install to C:\vcpkg
NVENC unavailable	Requires NVIDIA GPU with NVENC support (GTX 600+ series)
Connection refused	Check firewall for TCP 443 and UDP 50000-50020
Black screen	Wait for keyframe or click canvas to focus
No audio	Click "Enable" in settings panel after first audio playback
Input not working	Click the video canvas to capture input focus
Certificate warning	Expected with self-signed certificate, click proceed
Kicked message	Another client connected (single client only)

Known Limitations

• Windows server only
• Single client connection (new client kicks existing)
• NVIDIA GPU required for encoding
• No file transfer support
• No custom cursor image sync (standard cursors only)

License

Business Source License (Personal-Online / No-Company) v1.1

Personal use permitted. Commercial use requires separate license.

See LICENSE.txt for full terms.

Copyright

(c) 2025-2026 Daniel Chrobak. All rights reserved.