PlexMind

Your Plex server, with a brain. On your own GPU.

Live demo Unraid License: MIT

• 🎯 Picks people actually want. Each user gets movies and shows matched to everything they watch, not just their last title, explained in one line, and delivered straight into Plex.
• 🎙️ Subtitles for everything. Whisper writes subtitles for videos that have none.
• 🌍 In your languages. A local model translates subtitles, slang and all, checked line by line.
• 🧹 A tidy library. Duplicate and replaced subtitles are quarantined, never deleted.

No cloud, no subscription, nothing leaves your server.

Install

Unraid

1. Add the template: https://raw.githubusercontent.com/FineChAInesium/PlexMind/main/templates/PlexMind.xml
2. Set your Movies and TV folders, and where your llama.cpp and Whisper containers are.
3. Open http://[unraid-ip]:8000 and Sign in with Plex.

No model yet? ./setup.sh --model /mnt/user/appdata/llama-cpp/models downloads one that fits your GPU.

Docker Compose

git clone https://github.com/FineChAInesium/PlexMind.git && cd PlexMind && ./setup.sh

setup.sh signs you in with Plex, finds your libraries, downloads a model sized for your GPU and starts everything. Then open http://<server>:8000 and sign in with Plex.

No tokens to copy. No API keys to make.

Bring any model

PlexMind works with any GGUF llama.cpp can run. The default is Gemma 4 12B, which made a third of the translation errors of the runner-up in blind tests. Want to try another? Model Trial runs real translation and recommendation work on it on your GPU, next to the current model, and won't let you switch to one that doesn't fit or breaks. More on models →

Needs

• Plex Media Server
• An NVIDIA GPU with 8 GB or more (12 GB for the default model)
• Docker, or Unraid

Docs

Models · Configuration · Architecture · API & command line · Upgrading · Security · Changelog

Keep port 8000 on your LAN. PlexMind writes to your media folders and starts its own containers, so don't forward it to the internet.

Made by @FineChAInesium. MIT licensed.