Downloadeight ways.

FetchDeck is a free download manager. It pulls one file over eight connections and resumes from the exact byte it stopped at.

Connections8
ubuntu-24.04.1.iso8 ranges
2.31 of 5.78 GB40%
Time left9.3 MB/s
6 m 23 s

One row, start to finish.

Arrives
blender-4.2.3-windows-x64.zip
Next in line download.blender.org
Asking the server…
Runs
ubuntu-24.04.1-desktop-amd64.iso
releases.ubuntu.com · 8 connections
4.06 / 5.78 GB
10.5 MB/s
2 m 48 s
Lands
node-v22.11.0-x64.msi
Done at 21:12 · nodejs.org
30.3 MB

First, it asks for one byte.

A range request for byte zero tells FetchDeck the real size, the filename and the ETag. A 206 reply means the server can resume. The Accept-Ranges header alone is only a hint, so it checks.

releases.ubuntu.comRequest and reply
GET /24.04.1/ubuntu-24.04.1-desktop-amd64.iso
Range: bytes=0-0

HTTP/1.1 206 Partial Content
Content-Range: bytes 0-0/6205243392
ETag: "171d1a00-171e50000"
Resumable5.78 GB8 connections

4×

faster with four ranges than with one.

That was the first test: the same file from a local server that caps every connection. Each bar here is the ceiling for that many connections, 1.5 MB/s apiece. Green is the run we measured.

1.5 MB/s1
3.0 MB/s2
4.5 MB/s3
6.0 MB/s4
7.5 MB/s5
9.0 MB/s6
10.5 MB/s7
12 MB/s8

When the download goes wrong

It resumes at the byte it stopped on

Pause, close the lid, come back tomorrow. Each range keeps its own offset, so nothing downloads twice.

gimp-2.10.38-setup.exe
Paused at 132.8 of 324.0 MB
Range 1picks up at 38.9 MB
Range 2picks up at 110.0 MB
Range 3picks up at 195.2 MB
Range 4picks up at 274.6 MB

It checks the file is still the same file

Before resuming it asks the server again. If the size or ETag moved, it starts over instead of gluing two versions together.

Saved ETag
"171d1a00-171e50000"
Server ETag
"171d1a00-171e50000"
Size
5.78 GB, unchanged
Last-Modified
Thu, 29 Aug 2024 16:41
Checked
21:05, on Resume
Ranges kept
9 of 9
Same file, continuing from 4.06 GB

It waits out a dead connection

Wi-Fi drops are normal, so it keeps trying from the last byte, waiting a little longer each time, never more than 30 seconds.

android-studio-2024.2.1.11.exe
Reconnecting in 8 s
1 s2 s4 s8 s16 s30 s30 s
21:03:12 Range 4: connection reset
21:03:13 Retry from 812.4 MB, failed
21:03:15 Retry from 812.4 MB, failed
21:03:19 Waiting 8 s before the next try

It fetches a fresh link when the old one dies

GitHub release links point to a signed URL that expires in about an hour. On resume FetchDeck asks the original link again.

You pasted
github.com/…/releases/download/v1.4/app.zip
Resolved to
objects.githubusercontent.com/…
Expired
21:04, 403
Re-resolved
21:05, 206
Range 3
back at 412.6 MB
Ranges 1, 2, 4
done before it expired
Fresh link, same file

It tells you when it can’t resume

Some servers ignore range requests. FetchDeck falls back to one stream and says so on the row, before you pause and lose it.

rainfall-district-2025.csv
Can't resume: the server doesn't support it
18.6 MB, starts over
Probe answered 200 OK with no Content-Range, so this file comes down as one stream. Pausing it means starting again.
Not resumable

It never writes over a file you have

Same name in the folder already? The new one gets a number. Half-finished files stay as .part until every range lands.

report.pdfYesterday
report (1).pdfMonday
report (2).pdfJust now
fedora-41-x86_64.iso.part61%
Git-2.47.0-64-bit.exeMonday

Free, and it stays free.

No pro tier, no trial, no ads between your downloads. The code is on GitHub under the MIT licence, so you can read exactly what it does with your connection.

₹0forever
  • Eight connections per file
  • Pause, resume and crash recovery
  • Three downloads at once, the rest wait in line
  • Windows installer and a Linux AppImage
Download for Windows

Fair questions

Is it faster than my browser?

Only when the server caps each connection, and most mirrors and file hosts do. On a server with no cap, one connection already fills your line and eight won’t help. The side panel shows each connection’s speed, so you can tell which case you’re in.

What if my PC crashes halfway?

Progress goes to disk about once a second, and only after the bytes it describes are flushed. Open FetchDeck again and the download waits, paused at its last saved byte. Press Resume.

Why stop at eight connections?

Past a point, extra connections mostly annoy the server, and some servers start refusing them. Eight is the ceiling in v1. Going higher waits until there are real benchmarks to justify it.

Does every link work?

Any plain HTTP or HTTPS link. If the server ignores ranges, FetchDeck downloads it as one stream and marks it as not resumable. Links behind a login don’t work yet.

Will it hammer the server?

No. A 429 or 503 makes it back off, honour Retry-After and run fewer ranges until things calm down. After five refusals it stops and tells you why.

Is there a browser extension?

Not yet. It’s the next thing after v1. For now, copy the link and press Ctrl+V anywhere in the window.

Which systems does it run on?

Windows, with an installer, and Linux as an AppImage or a .deb. There’s no macOS build.