torsync

Cross-seed tooling for a qBittorrent library: take torrents that are already seeding on one private tracker and seed the same data on another, without re-downloading a single byte.

Built around a specific setup - qBittorrent on a Windows host, Prowlarr on a Linux Docker box - but the scripts are generic.

How cross-seeding actually works

Two torrents, one copy of the data:

1. The existing torrent keeps seeding to tracker A.
2. The .torrent from tracker B is added as a separate torrent.
3. Its save path points at the same folder as the existing data.
4. A force-recheck confirms 100%, and it starts seeding.

Both trackers count you as a seeder. One copy on disk.

Never merge announce URLs. Adding tracker B's announce URL to tracker A's torrent leaks your passkey into tracker B's swarm. Both sites ban for it. Two separate torrent entries is the only safe method.

Keep Auto Torrent Management off for cross-seeded torrents - it can relocate files and break the other torrent.

A match only works if the remote release is byte-identical: same file layout, same sizes. A name match means nothing. That is why match.py filters on exact size rather than title similarity.

Layout

scripts/scan-qbt.py          inventory local torrents (bencode, no WebUI needed)
scripts/match.py             search an indexer per torrent, keep exact-size hits
scripts/add-indexers.sh      register TorrentLeech / TorrentBD in Prowlarr
scripts/flaresolverr-tag.sh  attach a Cloudflare solver to one indexer
stack/docker-compose.yml     Prowlarr + Byparr + Solvearr
stack/byparr/                Byparr image patched to accept request cookies

Setup

cp .env.example .env      # fill in credentials; .env is gitignored
cd stack && docker compose up -d
bash scripts/add-indexers.sh

Inventory what you have, then look for candidates:

python3 scripts/scan-qbt.py --tracker torrentbd
python3 scripts/match.py --indexer 1 --tracker torrentbd --json tl-matches.json

match.py caches to .match-cache.json. Trackers rate limit hard - TorrentLeech declares requestDelay: 4.1, so a ~90 torrent sweep is several minutes of deliberate waiting. Do not lower --delay below the indexer's declared value.

Cloudflare solvers

Private trackers sit behind Cloudflare, and the three common solvers each fail differently. Measured, not assumed:

Solver	Clears Cloudflare	Forwards cookies
FlareSolverr	no - times out on Managed Challenge	yes
Solvearr	no - 403	yes
Byparr (upstream)	yes	no - silently dropped
Byparr (patched here)	yes	yes

Upstream Byparr declares cookies on its response model only, so cookies sent by a client are dropped by pydantic and every request goes out anonymous. stack/byparr/add_cookie_support.py adds the field to LinkRequest and injects into the browser context before navigation. The patch is applied at image build and fails loudly if upstream's layout changes.

Solvers are expensive and flaky. None is attached to any indexer by default - opt in per indexer only when one actually returns a Cloudflare error:

scripts/flaresolverr-tag.sh on TorrentBD

TorrentLeech does not need one; it serves plain requests fine.

Known limits

• Cookie-auth trackers may bind sessions to a browser fingerprint. TorrentBD sets a ufp cookie and rejects a valid x_auth presented by a different browser, so a solver-based proxy cannot authenticate even with fresh cookies.
• Music rarely cross-seeds. General trackers have one flat audio category and little overlap in piece layout with dedicated music trackers.
• cf_clearance is bound to IP and User-Agent. Harvest cookies on a machine behind the same egress IP as the solver, or they will not work.