Projectionist

A self-hosted AI curation layer for Plex. Reads your library, applies a mix of rule-based filters and AI judgment, and writes the results back as Collections pinned to your Plex home screen as browsable rows — refreshed automatically on a weekly schedule.

---

Rows

Movies

Row	Logic	Description
Collecting Dust	Rule-based	Unwatched films sitting in your library for 30+ days
Easy Watch	AI	Warm, low-stress comfort picks
Existential & Atmospheric	AI	Philosophical, meditative films that linger after the credits
Second-Hand Adrenaline	AI	High-tension, propulsive thrillers and crime films
90-Minute Dash	Rule-based	Unwatched films under 90 minutes
Wildcard	AI	A surprise collection invented fresh each week — name, theme, and films all chosen by the AI

TV Shows

Row	Logic	Description
Collecting Dust	Rule-based	Shows started but not watched in 60+ days
Give it a Shot	Rule-based	Shows with zero episodes watched

Each run shuffles the results — rows feel different every week. No film appears in two movie rows (except 90-Minute Dash, which overlaps intentionally).

---

Requirements

• Docker Desktop
• A Plex server
• An AI provider — see AI providers below

---

Setup

1. Get the files

Download both files from the latest release:

curl -LO https://github.com/rafvasq/projectionist/releases/latest/download/docker-compose.yml
curl -LO https://github.com/rafvasq/projectionist/releases/latest/download/config.example.yaml
cp config.example.yaml config.yaml

2. Edit config.yaml

Fill in your Plex URL and token. To find your Plex token, follow this guide. Then choose an AI provider — see AI providers below.

3. Start

Gemini (no GPU required):

docker-compose up -d

Ollama (local, self-hosted):

docker-compose --profile ollama up -d

Projectionist runs once on startup, then on the configured cron schedule (default: Mondays at 3am).

Viewing logs

docker-compose logs -f projectionist

A successful run ends with Done — 7 collections updated and pinned to library views.

---

Configuration

All options with descriptions:

plex:
  url: "http://192.168.1.X:32400"       # Plex server URL; use host.docker.internal when running on the same machine as Plex
  token: ""                             # Plex auth token
  library: "Movies"                     # Name of your movie library
  tv_library: "TV Shows"                # Name of your TV library

ai:
  provider: gemini                      # gemini | ollama

  # Gemini
  model: "gemini-2.5-flash"
  api_key: ""

  # Ollama
  # model: "llama3"
  # base_url: "http://ollama:11434"     # use http://localhost:11434 outside Docker

rows:
  max_results: 15                       # max items per row; duplicates across rows are excluded
  collecting_dust:
    enabled: true
    min_age_days: 30                    # how long a film must sit unwatched before qualifying
  easy_watch:
    enabled: true
  existential:
    enabled: true
  adrenaline:
    enabled: true
  quick_watch:
    enabled: true
    max_minutes: 90                     # maximum runtime in minutes
  tv_collecting_dust:
    enabled: true
    idle_days: 60                       # days since last watched episode before a show qualifies
  give_it_a_shot:
    enabled: true
  wildcard:
    enabled: true                        # AI invents a new collection name, theme, and film list each week

schedule:
  cron: "0 3 * * 1"                    # every Monday at 3am

Any row can be disabled by setting enabled: false.

---

AI providers

Gemini (default)

No GPU required. The free tier of gemini-2.5-flash handles a typical home library with one API call per AI row per run.

1. Get an API key at aistudio.google.com.

2. In config.yaml, set provider: gemini and fill in your api_key and model.

Ollama (local, self-hosted)

No API costs, no data leaves your network. 7B–9B parameter models work well (llama3, mistral). Ollama runs as a sidecar in Docker — not exposed outside your machine.

GPU support requires the NVIDIA Container Toolkit.

1. In config.yaml, set provider: ollama and fill in the model and base_url under # Ollama.

2. Start with the Ollama profile:

   docker-compose --profile ollama up -d

3. Pull a model:

   docker-compose exec ollama ollama pull llama3