Skip to content

About

Self-hosted recommendations from everything you've actually watched, on every service. Netflix, Prime Video, Apple TV, Disney+, Max and Plex history; Claude, Gemini, OpenAI or a local model.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Streamline

Streamline

Stop scrolling. Start watching.
A self-hosted recommendation engine that learns your taste from everything you've actually watched,
then answers "what should I watch tonight?" like a friend who has seen it all.

Documentation • Quick start • Screenshots

Python 3.10+ Claude, Gemini, OpenAI, or a local model Self-hosted MIT license

Streamline home page: ask for anything, and see the tastes it learned from what you loved


Every streaming app recommends from its own catalogue, to keep you on its own service. None of them know you also watched three seasons of a British cop show somewhere else, or that you gave up on the drama everyone loved.

Streamline does. It reads your real history from Netflix, Prime Video, Apple TV, Disney+, Max, and Plex, learns what you love, and recommends across every service, with a reason for every pick.

You: gritty British crime drama

Streamline: Unforgotten · IMDb 8.4 · Prime Video A methodical, understated British police drama set on grey London streets, with the patient slow-burn tension, emotional depth, and layered cold-case mysteries this user loves from Broadchurch.

Why it's different

  • It knows your history, from every service. Not one app's view of you, but all of it in one place.
  • It learns from what you loved, not just what you finished. Tap through your history in Rate It, and your taste becomes named rows like Proper British Telly or Sci-fi that stays with you.
  • It doesn't recommend what you've already seen, whichever service you saw it on.
  • It explains itself. Every pick says why it fits you. Ask "why not Slow Horses?" and it tells you exactly why it was left out.
  • It's yours. It runs on your laptop or home server, your history lives in one SQLite file, and you choose the AI: Claude, Gemini, OpenAI, or a local model.
  • It's cheap to run. With the default Claude models, about $1 to describe 1,000 titles once, a few dollars to build a large taste profile, and roughly 6 cents a search. Find, On Deck, and the rest of the app cost nothing. Details

A look around

Search results with explanations
Ask anything. Plain English in, explained picks out, with where to stream them.
Mood Match
Mood Match. Can't decide? A few quick questions about tonight, then picks.
On Deck
On Deck. Every show you follow, and which ones have new episodes ready.
Find
Find. The best-rated titles you haven't seen, filtered your way. Instant, no AI.
Rate It
Rate It. Tap the posters you loved. That's how it learns.
Archive
Archive. Everything you've ever watched, from every service, in one wall of posters.

Screenshots come from a demo library of famous titles, not anyone's real history.

Also

  • Watchlist across every service, with CSV export
  • Plex plays and ratings arrive as they happen
  • IMDb ratings on search results, with TMDB as the fallback
  • Home Assistant sensors and a calendar of upcoming episodes
  • Docker images for x86 and ARM, Raspberry Pi included
  • A full command line, with an interactive mode that remembers the conversation ("more like #2", "but lighter")

Quick start

With Docker:

git clone https://github.com/abhichandra21/streamline.git && cd streamline
mkdir -p data recommender/cache logs
touch config.local.yaml                             # Settings saves your changes here
docker compose up -d                                # open http://localhost:5051

Or with Python 3.10+:

git clone https://github.com/abhichandra21/streamline.git && cd streamline
python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt
./recommend-web start                               # open http://localhost:5051

Then follow Get started in the browser: add your keys, point at your exports in data/, and build your library. Prefer the terminal? Put the keys in .env, set export paths in config.local.yaml, and run ./recommend setup.

The Getting started guide walks through every step, including where to download each service's history.

How it works

Once: Streamline matches every title you've watched on TMDB and writes a short description of each. From the titles you mark Loved, it builds a taste profile of named themes.

Every search: an LLM turns your words into a structured request. Candidates come from TMDB and from the model's own suggestions, anything you've seen is removed, and the rest is checked for where it streams and ranked against what you asked for and what you love.

The architecture page has the full picture.

Documentation

Everything else is on the docs site, streamline-docs.pages.dev:

Contributing

See CONTRIBUTING.md for how to report bugs, propose changes, and run the tests.

License

MIT

About

Self-hosted recommendations from everything you've actually watched, on every service. Netflix, Prime Video, Apple TV, Disney+, Max and Plex history; Claude, Gemini, OpenAI or a local model.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages