Marathon is a media streaming service designed to binge watch all your stuff in a single session. A dream of a Whovian or just an everyday nerd that has time for a new short series.

Key Features

• True Marathon Playback: Treats an entire series as a single, continuous stream, splicing episodes together on-the-fly for seamless, back-to-back viewing without loading screens.
• Visual Interval Editor: A built-in editor to precisely mark custom time segments in any episode. Define openings, endings, recaps, or bonus content.
• Automatic Content Splicing: Dynamically removes the HLS segments corresponding to your marked intervals during playback, so you see the first opening and then just the story.
• Audio Fingerprinting: (broken) Mark an opening once, and Marathon will automatically detect and skip it in all other episodes. Uses acoustic fingerprinting (chromaprint) to find matches even with slight variations.
• Hardware-Accelerated Transcoding: Auto-detects NVIDIA (NVENC) or Intel/AMD (VAAPI) GPUs and uses them for encoding. Falls back to CPU gracefully. Your GPU finally does something useful.
• High-Performance Streaming: Powered by HTTP/2 multiplexing and parallel media indexing, ensuring instant seek times and fast library scans.
• Import / Export Intervals: Save all of a series' interval data to a single JSON file. Share your timings with friends or back them up.
• Keyboard-First Navigation: Designed for the "10-foot UI" experience. Features spatial navigation for all controls and advanced playback shortcuts.
• Variable-Speed Scrubbing: Hold down the seek keys to fast-forward or "rewind" at an accelerating speed, from 2x up to 64x.
• Persistent Watch History: Remembers your exact timestamp within the entire series marathon, not just a single file.
• Simple File-Based Library: Just point it at your media folder. No complex metadata agents or forced naming schemes.

Quick Start

1. Set the MEDIA_ROOT environment variable to point to your media library (e.g., export MEDIA_ROOT=./media).
2. Run make run (or cargo run --release -p marathon-server).
3. Navigate to https://localhost:3000.

⚠️ Security Warning: To support HTTP/2 (which dramatically improves streaming performance), Marathon runs over HTTPS by default.

On first run, it generates a self-signed certificate. Your browser will warn you that the connection is not private. Click "Advanced" -> "Proceed to localhost (unsafe)" to continue.

Configuration

Marathon is configured via environment variables:

Variable	Description	Default
MEDIA_ROOT	Path to the directory containing your video files.	./media
TLS_CERT	Path to a custom TLS certificate file (.pem or .crt).	(Auto-generated)
TLS_KEY	Path to the corresponding private key file (.pem or .key).	(Auto-generated)
RUST_LOG	Logging verbosity levels.	info,marathon_server=debug

Transcoding Options

All transcoding options are prefixed with MARATHON_. Most defaults are sane, but feel free to tweak:

Variable	Description	Default
MARATHON_CACHE_DIR	Where to store transcoded segments.	./cache
MARATHON_MAX_CONCURRENT_TRANSCODES	Max parallel FFmpeg jobs.	4
MARATHON_SEGMENT_READY_TIMEOUT_SECS	How long to wait for a segment.	120
MARATHON_SKIP_PREVIEW	Skip fast preview, go straight to native.	false
MARATHON_FAST_HEIGHT	Resolution for preview quality.	720
MARATHON_FAST_CRF	Quality for preview (higher = smaller).	28
MARATHON_NATIVE_CRF	Quality for native (lower = better).	23

Set MARATHON_SKIP_PREVIEW=true if you have a fast GPU - native quality is often faster than the two-stage preview -> native approach.

Hardware acceleration is auto-detected. If you have an NVIDIA GPU with NVENC support, it'll just work. Check the logs for [NVENC] or [VAAPI] to confirm.

Using Custom Certificates

If you want to get rid of the browser warning, you can provide your own trusted certificates (e.g., via mkcert):

export TLS_CERT=./certs/localhost.pem
export TLS_KEY=./certs/localhost-key.pem
make run

FAQ

Why?

I was mildly annoyed by my cheap projector's controls and the general clumsiness of streaming from various sites. I wanted to watch the whole story without the repetitive openings and endings, and manual labor (clicking "next episode") - just a never-ending stream of the show. That can't be too hard.

Why bother with playback controls for a potentially 1000-hour stream? Skipping a few seconds seems unbearable.

Big number go up hehe.

Seriously, it was a fun technical challenge. The player builds a virtual timeline of the entire series, so seeking from Episode 1 to Episode 50 is instantaneous.

Clumsy controls??

They were customized specifically for my workflow. Feel free to change the keybinds or even open a PR with a keybind configuration feature (I'll be very happy to take a look).

Why ffmpeg, and HLS instead of X and Y?

Idk, FFmpeg is cool and ancient, and HLS is a neat standard, so why not use them all at the same time? Also, if you do care enough - all components of the backend pipeline are replaceable in a Ports and Adapters manner. Feel free to create and share your alternative adapters for the pipeline components.

How do I know if hardware acceleration is working?

Check the server logs on startup. You'll see either:

• NVIDIA NVENC hardware acceleration available - you're golden
• VAAPI hardware acceleration available - Intel/AMD GPU is being used
• No hardware acceleration available, using CPU encoding - software fallback

During playback, transcode logs will show [NVENC] or [VAAPI] if HW accel is active.

Testing data?

Demo scripts in the root can generate some placeholder files. The system can handle timelines for over 1000 hours of video files just fine - well, at least my system can :p

Also, if it counts for anything, it was battle-tested on all of SPY x FAMILY (all episodes and the movie) and both seasons of Saiki K. All went fine.

Something is broken/doesn't work well

As said before - works on my machine. Open an issue, I'll be happy to look at it in my free time :)

License

This project is licensed under the GNU General Public License v3.0.

See the LICENSE file for details.