For business · commercial licence
rlibtorrent is the engine inside rbittorrent. In our races that client gets to the swarm first, finishes first when the tracker is slow and uploads the most, on a fraction of qBittorrent's memory. Put the engine in your own client, appliance or service and your product starts with that lead.
Why it pays
7.02 s vs 14.6 s
A private tracker hands a new upload's first peers to the clients it has accepted, and the ratio goes with them. With the tracker holding the torrent back for 5 s, the client built on it was let in at 7.02 s; rTorrent took 14.6 s, and qBittorrent and Deluge were not let in at all within the 60-second race.
SSD, tracker lag 5 s, 5 runs, medians, from each client's own fetch of the torrent.
8.12 s vs 17.1 s
Customers pay for the download to be there. When the tracker was slow to accept, the client built on it had every piece in and hash-checked in 8.12 s, against 17.1 s for the fastest other client, rTorrent; qBittorrent and Deluge did not finish within the race. On a 50 GiB download, 16.9 min against qBittorrent's 22.2 min.
SSD, tracker lag 5 s, 5 runs, medians; the large download: 50 GiB, the spinning disk, one run each, 2026-10-13.
69% vs 22%
Ratio is what a private-tracker customer pays a seedbox for. In the first minute the client built on it sent 69% of everything the racing clients sent each other, against rTorrent's 22%, and it uploaded the most in every one of the four setups.
SSD, tracker lag 0 s, 5 runs, medians; the seeders' own upload not counted.
83 MB vs 1,022 MB
Memory a client does not take is room for the next customer. Downloading 50 GiB, rbittorrent peaked at 83 MB, 12× less than qBittorrent's 1,022 MB. Holding 1,100 torrents at once it used 40.1 MB against 64.5 MB. And less processor time for the whole download: 159.2 s against 219.8 s.
50 GiB, the spinning disk, one run each, 2026-10-13, one seeder on the same machine. The catalog: 1,000 paused and 100 active torrents, 2026-10-16.
22 of 22
Private trackers ban clients whose reports do not add up. In every race rbittorrent finished, its copy matched the original byte for byte, and the upload it reported matched the tracker's own count every time. On a private torrent it never publishes to the DHT or shares peers by exchange, and announces to one tracker tier at a time.
22 races, scorecard 208; the private-torrent rules are checked by the bench's own probes.
26 crates, 1 name
The rlibtorrent front door re-exports the whole engine, so your product depends on one name. A C ABI with a generated header serves C and C++, and the daemon crate gives you a qBittorrent-compatible Web API without writing one.
The crate map is on the home page.
15 fuzz targets
Its behaviour is compared with two pinned libtorrent builds, rqbit and Transmission, every parser is fuzzed, and a deterministic lab replays scripted peers on a virtual clock. Fewer surprises in your product, fewer support tickets for you.
How it is checked is on the home page.
The licence
An active slot is a person who used it in the last 24 hours, counted across every daemon on one licence. A daemon nobody used counts as one. An idle seat costs nothing.
No download, seed, add or API call is ever refused for the licence. Over your count, the software says so, gives you a grace period, and asks you to upgrade.
The licence check carries the licence id, the instance id, the active-slot count and the version: never a torrent name, a file, a hash or a customer.
Who needs one
Commercial terms are not published yet. Tell Lunar Works what you run and for how many people, and we will come back with terms.
Every number on this page is read from the bench's result files by the script that builds the site, under the conditions on the method page.
