peter
8036251589
Two ways round 6's own fixes did not bound what their comments said they bounded. The reclaim budget's window anchor was written only where the window RESET, so it marked the window's start rather than the last loss — a tumbling window, not the rolling one the constant is named for. Any loss cadence slower than half the window satisfied the reset on every second loss, so the budget never reached its bound: measured over 200 loss cycles, 61/70/90/119s apart all gave 200 reclaims and never an error, while 5s and 30s apart correctly gave two and then the honest answer. The failing band is exactly the case the ERROR handler names — a transcode that falls behind realtime — and both committed tests spaced their losses 0s and 10min apart, i.e. outside it. The anchor now moves with every loss. And `finish`'s refusal handler lowered the play-through latch, which is the only gate on the 10s interval, the page-hide beacon and `settle`. All three post `absRef`, which on the play-to-the-end path IS the end of the media, and the backend reads a position within `_FINISH_MARGIN_S` of the duration as a finish — so within ten seconds of the toast saying the mark had failed, the item was marked watched, its position cleared and a scrobble sent to Plex. Round 6 closed the explicit rollback write and left the other three open; the test it added set `upNext: false`, which tears the effect down before any of them can fire. The refusal now takes back the CLAIM rather than the latch: a second flag stops `settle` re-asserting watched, and playing again clears both. `FINISH_MARGIN_S` duplicated the backend's constant in a second language with nothing relating them, and its comment had the asymmetry backwards — if the BACKEND's margin grows, this guard starts admitting a rollback the backend reads as a finish, i.e. re-opens the defect above. Pinned by a contract test that reads the Python, and the comment corrected.
Siftlode
Your YouTube subscriptions, the way a feed should work. Siftlode pulls every upload from the channels you follow into one clean, filterable feed — no algorithm deciding what you see, and no Shorts or livestream noise unless you want it. Self-hosted, multi-user, and private: your data stays on your own server.
Everything expensive (channels, videos, metadata) is fetched from YouTube once and stored locally, so filtering, searching and sorting are instant and don't burn API quota. Click a video to watch it in an in-app player that resumes where you left off — or open it on youtube.com so your own ad blocker and SponsorBlock keep working.
Features
- A readable subscription feed — sort and filter by channel, tag, language, topic, length, upload date or watch state; hide channels without unsubscribing; save filter setups as named views.
- Search all of YouTube from the feed — live results you can play, save or add to playlists like any other video; they stay temporary and out of your catalog unless you keep one.
- Channel pages & a channel manager — per-channel stats and uploads, priorities, and your own tags to slice the feed by.
- Playlists with two-way YouTube sync — build them locally, keep them in sync in both directions.
- In-app player with resume, plus keyboard/scroll controls.
- Download Center — save videos to the server with yt-dlp in a Plex-friendly layout (format presets, per-user storage quota), trim / crop / split & join them in a built-in editor, then save to your device, share with another user, or hand out a public watch link.
- Plex library (optional) — point Siftlode at your own Plex server to browse and play its movies and shows alongside your feed, with watch state kept in sync.
- Multi-user with per-user private state, a shared catalog, and a fair daily API-quota guard.
- Self-hosted & private, with a first-run web setup wizard and the interface in English and Hungarian.
Quick start (self-hosting)
You don't need to build anything — Siftlode runs from a prebuilt public image, and all configuration (your admin account, Google sign-in, email) happens in a first-run web wizard. You need Docker with the Compose plugin.
1. Get the files and run the installer:
git clone https://git.b1fr0st.eu/peter/siftlode.git
cd siftlode
./install.sh # Windows (PowerShell): ./install.ps1
The installer generates a private .env (secrets), pulls the image, starts the app + database, and
prints a one-time setup URL like http://localhost:8080/setup?token=….
2. Finish in your browser. Open that URL and follow the wizard:
- Admin account — the email + password you'll sign in with.
- Google sign-in (optional) — paste a Google OAuth client to enable "Sign in with Google" and pulling your YouTube subscriptions. Skip it to use email + password only.
- Email / SMTP (optional) — for verification/notification emails. Skip it and you (the admin) simply approve new accounts yourself.
Then sign in with your admin account. That's it. See docs/self-hosting.md for the full walkthrough.
Just trying it out? Press Enter at the installer's URL prompt to run on
http://localhost:8080.
Build from source (alternative)
Prefer to build the image yourself instead of pulling it:
git clone https://git.b1fr0st.eu/peter/siftlode.git
cd siftlode
cp .env.example .env
# generate the two secrets and paste them into .env:
python -c "import secrets;print('SECRET_KEY='+secrets.token_urlsafe(48))"
python -c "import base64,os;print('TOKEN_ENCRYPTION_KEY='+base64.urlsafe_b64encode(os.urandom(32)).decode())"
docker compose up --build -d # builds from the included Dockerfile
Open http://localhost:8080 and finish in the setup wizard as above. (Set a POSTGRES_PASSWORD in
.env too.)
HTTPS / public access
Port 8080 over plain HTTP is fine for a LAN or a quick trial. For public access put a reverse
proxy (Caddy, Nginx, Traefik…) in front to terminate TLS, and set the public URL (the installer
prompt, or OAUTH_REDIRECT_URL in .env) to your https://… address — this also marks the session
cookie secure. Add that same …/auth/callback URL to your Google OAuth client's authorized redirect
URIs. Behind a proxy, also set TRUSTED_PROXY_IPS so the rate limiters see the real client IP and
can't be bypassed via a forged X-Forwarded-For — see docs/self-hosting.md.
Updating & backups
docker compose -f docker-compose.selfhost.yml pull # or: docker compose pull
docker compose -f docker-compose.selfhost.yml up -d
That follows latest, which moves with every release. Set IMAGE_TAG=0.80.0 in .env to pin a
version and upgrade on your own schedule; released versions are listed in CHANGELOG.md
and in the app's Release Notes. The registry keeps only the newest few of those, so a pin decides
when you take the next hop rather than letting you stay put forever — see
docs/self-hosting.md.
Back up before upgrading. Migrations run automatically on startup and are not undone by pulling
the old image back. A backup is the database dump plus your .env — TOKEN_ENCRYPTION_KEY is
what decrypts the Google, SMTP and Plex secrets stored in that database, so a dump restored without it
is only half an instance. Use scripts/backup.sh (or backup.ps1 on Windows) and scripts/restore.sh;
the equivalent one-liners, the pinning recipe and the rollback steps are in
docs/self-hosting.md.
How it works
- Shared catalog, private state. Channels and videos are stored once and shared; each user's subscriptions, tags, playlists and watch/save/hide state are private.
- Cheap by design. Public reads are cached locally; a shared daily quota budget and a background scheduler keep unattended syncing within YouTube's free API limits. An optional API key lets backfill run without depending on a user's OAuth token.
Tech
FastAPI + PostgreSQL (SQLAlchemy, Alembic) backend; React + Vite + Tailwind + TanStack Query frontend; packaged as a single Docker image with Docker Compose.
License
GNU AGPL-3.0. You can run, study, modify and share Siftlode; if you modify it and let other people use it over a network, they are entitled to your modified source. Copyright © 2026 Siftlode contributors.
Note
This project is developed with AI assistance.