Pick the languages you're happy to speak and hear, and randos
pairs you with a compatible stranger: a short voice call, then
you both disappear. No accounts, no profiles, no video, no pressure.
How it works
- Choose every language you're happy to speak and hear. You can practise both ways (I speak Spanish, you speak English),
both speak the same language, or just listen.
- Confirm you're 18 or older. Calls are anonymous and unmoderated.
- randos finds someone who speaks a language you're happy to hear and is
happy to hear one of yours, and connects the two of you. It chooses one
speaking language for each of you and shows it beside your waveform.
Whoever has waited longest is matched first, and you won't be matched with the same person
again straight away.
- You talk for ten minutes. When time is up, you both choose whether to
keep going, ten more minutes at a time, up to thirty. Either of you can
end the call or move on to the next rando at any moment. If your
network changes or drops for a moment (Wi-Fi to mobile data, a tunnel),
the call says "Reconnecting audio…" and comes back on its own, within
fifteen seconds, or ends if it can't.
You can also bring someone you know: "Invite someone" makes a private,
one-time link to send them (it works for 15 minutes); whoever opens it
joins a call with you, the same ten minutes and the same vote, and you
both go back to the form afterwards, never to a stranger.
That's all there is. Nothing about the call is kept. The site speaks
all 20 languages, with a switcher in the navigation. Its translations are
machine-made, so
/translations lists every
string next to the English: "Suggest a fix" opens a GitHub issue with
everything filled in but your fix, and a pull request follows on its own.
Why these 20 languages, and how to add one:
docs/languages.md.
randos.slowinput.org/status shows
whether calls can connect right now and how many there have been today:
calls, not people.
In your browser
Open randos.slowinput.org and allow the
microphone when asked. Tested in Chrome and Safari, on desktop and mobile.
Before a call, expand Voice masking in the setup card. It offers a changed tone, with or without a gentle
pitch change. Choose Fuller or Brighter and use Hear myself with headphones
to try it locally. Preview stays on as you change styles; choosing Off lets you
hear your original voice. Nothing is recorded or uploaded. Masking starts off in each
new session; it changes how you sound, but people may still recognize you.
Press ? or click Keyboard shortcuts below the homepage form to show
the shortcut bar and activate
F Find, I Invite, P microphone preview, and M mute during a call.
C opens/closes “I will speak in.” X stops looking while you’re queued. A toggles the acknowledgment checkbox and B opens/closes Voice masking.
During a call, E ends it and N moves on to the next rando (except invitation calls).
P opens the voice panel if needed; with it open, S cycles Style and T
cycles Tone. O opens About, R Server Status, and L the interface-language
menu (arrows move, Enter applies, Esc closes the menu first). Letters work in either case. ? or Esc hides the bar and disables shortcuts. Server Status
shows connection health and call counts. Once you open shortcuts, this browser
remembers to show them when you return or change language. Esc hides them for
the current visit.
Speaking/hearing choices carry across tabs and return visits during the browser
session. They do not restore microphone preview, masking, or the 18+ checkbox
on a fresh visit.
In your terminal
The same calls, from a terminal (macOS on Apple Silicon, or Linux x86_64
with glibc 2.39 or newer, e.g. Ubuntu 24.04). Terminal users and browser
users can be matched with each other.
brew install wasichu/randos/randos
randos
In the terminal, press b before finding someone to open Voice masking.
Choose a style with 1/2/3, change tone with left/right, and press p
to hear yourself or stop. Preview stays on while you change style or tone;
Off lets you hear your original voice. Esc closes the preview. Voice masking on
and the selected style/tone stay visible in the main screen’s header when enabled,
including during calls. Use −/+ in the panel to adjust Preview volume;
it changes local listening only. Choices last for this
session and apply to your next call. Press l to choose the interface
language from a list; arrows move, Enter applies, Esc cancels. See the voice panel.
On Linux without Homebrew, download the archive from the
latest release, unpack
it anywhere, and run bin/randos (it needs ALSA, OpenSSL 3 and the system C++
standard library—libstdc++6 on Ubuntu—which most desktop distributions have).
- The essential keys for each screen are listed along the bottom; press ?
for all shortcuts available on that screen (Esc closes help, ↑/↓ scroll).
Use arrows to move through languages and space to check speaking or hearing choices
and Enter to find a rando. During a call,
m to mute, n for the next rando, e to end the call, q to
quit.
v invites someone you know: it shows a one-time link to send them
(y copies it, c withdraws it). j joins through a link you were
sent: paste it and press Enter (pasting on the start screen opens the
prompt too). Browser and terminal users can invite each other.t shows a mic level meter (silent: nothing is played back); t again
hides it and closes the mic. i and o choose the mic
and speaker. s shows the server's status, as on
/status; i there says what's
counted and what isn't.- Your languages, the 18+ confirmation, and your mic and speaker are
remembered on your computer:
randos --settings shows where,
--edit-settings opens them in your editor, --reset-settings forgets
them.
- Echo is cancelled, so speakers work, but headphones sound best.
- On macOS, the first call asks for microphone access for your terminal app.
randos --check tells you whether the audio part works.- Logs go to
~/Library/Caches/randos/tui.log (macOS) or
~/.cache/randos/tui.log (Linux). They never contain audio.
Privacy
- No accounts and no history. There is no database: the waiting room and
the calls exist only in the server's memory and vanish when they end.
- The server never hears you. Audio is encrypted between the two of
you (WebRTC) and passes through our relay, which forwards it without
being able to decrypt it.
- No recording, no ads, no tracking. The site loads nothing from other
companies: no analytics, no fonts, no CDNs.
- Your IP address stays between you and our servers. Like any
website, the server sees it while you're connected. The person you're
talking to never does: every call goes through our relay, and the server
strips direct addresses from everything it passes between you.
What each part sees, logs, and keeps, with the configuration behind every
claim: privacy.
Why randos is like this
randos is intentionally constrained:
- no accounts
- no profiles
- no followers, feeds, streaks, ratings, or scores
- no video
- no recording
- no server-side audio access
- no permanent social graph
- no ads, no tracking, no charging
It is not a social network, a content platform, or a gamified dopamine
delivery system. It's closer to public conversational infrastructure: a
calm place where adults can briefly meet, talk, and then disappear. It
supports crosstalk, language exchange, listening practice, and casual
conversation without forcing people into rigid modes or progress loops.
The product should feel like a doorway, not a platform.
randos is free infrastructure for the language-learning community, run as a
service. Donations may be welcome someday.
Working on randos or running your own
- Development: setting up, running it locally, the
tests, and how the code is laid out.
- Self-hosting: running your own instance, its
configuration, and the TURN relay.
License
randos is free software under the GNU Affero General Public License,
version 3 or (at your option) any later version: see LICENSE.
Anyone may use, change, and share it; if you run a modified randos for
others to use over a network, you must offer them its source too.
