SalamAI
Bring every voice into the conversation.
Speak naturally. Connect across languages with live voice translation and readable captions.
One to One
Two people. Two languages. A shared conversation.
One to Many
Reach your audience in the languages they understand.
Listen on your own device
Join through a shared link or QR code. No app to install.
Need access? Ask your workspace administrator for an invitation.
Translate
Choose a conversation type, connect your audio, then start translating.
AI translation may contain errors. Verify critical information before acting.
Conversation type
Translation type
Console settings apply to this login only. Before Start, check Save SRT/MD and Keep awake; both are off by default.
About this translation output
Starting sends audio through our translation relay and external service. Original and translated text are generated even when subtitles are hidden.
Advanced translation options
Choose the voice used for the primary translated output. Additional languages and One to One participants can choose their own voice.
One to One uses voice activity detection. Choose each person's language; automatic language selection is not available for this option.
One source = translation mapping per line, up to 64. Additional channels use normal translation. Terms are saved with this session's settings.
Visual translation — tab or camera
Optional: selected images are sent to an external service and incur additional usage. This app does not save screenshots. Audio sharing alone never shares images.
Visible text
Translated visual text
AI translation may contain errors. Verify critical information before acting.
What you're translating
This browser cannot capture another app or tab. Use the microphone or a remote speaker.
Set up an audio cable
One to Many does not support using this phone’s microphone and loudspeaker together. Connect headphones and select them in your phone’s audio controls, or open Companion on another device. Audience audio and recordings are unchanged.
The browser cannot verify your output route. Confirm again after reconnecting audio or returning to this page. If headphones still pop, use a separate listening device.
Give each speaker their own microphone. Whoever is speaking is switched through automatically; the others stay muted, so only one voice reaches the translator at a time.
Speaking:
Create a one-off speaker login and share the QR — they speak from their phone and it's translated live for the audience.
Languages
Translation language and paths
Ready when this language starts
How it works
Hold your language, speak, then release to hear the translation. Keep this page open. Your phone controls the speaker or headset route and media volume.
Choose Push-to-talk or Auto, then press Start. In Push-to-talk, hold a person's button to speak and release; the floor changes only after the complete translation plays. Headphones are recommended to avoid echo.
Optional with separate outputs. Speech is sent live; translated playback waits until release or Auto detects speech end.
Continuous keeps both microphones open. Each person's speaker can be picked up by the other person's microphone and translated back, looping until neither side is audible. Headphones for both people prevent this — separate devices alone do not.
Sends a few seconds of each microphone through this server to be identified. The clips are not recorded or kept.
Participants
Person A
Devices
Preparing Person A's phone…
Enable Companion on the phone, or change Uses to console devices.
AI translation may contain errors. Verify critical information before acting.
Person B
Devices
Preparing Person B's phone…
Enable Companion on the phone, or change Uses to console devices.
AI translation may contain errors. Verify critical information before acting.
Both people hear on the same device. Choose a different Hears on for each person to send each translation to its own headset.
On this screen
Show speech as text in its original language. Translation channels remain saved but will not run.
Translation paused for this run. Your language settings are saved.
Locked while a session is running — stop first to change it.
More options Listening here, voice, sharing and session
Console monitor
Optional
Advanced speech Speech speed and repeating the target language
Locked while a session is running; speech speed can still change.
Slows translated audio for everyone who hears it and saves the same pacing in encrypted recordings, while preserving voice pitch. Long uninterrupted speech eases toward 100% before delay exceeds 2.5 seconds.
Companion Mode
Phone listening, links and audience controls by languageAutosleep is set per language in its ⋯ menu (on by default): an extra language shared only to phones starts when someone joins and stops a minute after the last listener leaves. Links and QR codes do not change. Your console language, recorded languages and languages on a screen always run.
Your organisation keeps every language running.
Stops a language whose listeners all have it paused or put away for this long. One tap on the phone (or a refresh) starts it again.
Turn on Companion Mode on any language above. That language gets its own persistent link and QR code.
Changes here apply to the link your audience already has — edit, then Save. Settings that affect what listeners agreed to will ask you to confirm.
Your organisation allows text-only sharing. Listeners follow subtitles without phone audio.
PNG or JPG — shown to listeners and in shared-link previews.
Each feed keeps that language's text on the server in plain text while it runs and deletes it when the feed ends. Links work only while the language is live and die when you stop, sign out or regenerate them.
Address restrictions rely on your proxy reporting real client addresses.
Pick a language above to offer a feed for it.
Session and sharing
Silence handling, listen-along and overlay links, qualityChoose a target language different from the one being spoken.
Name corrections
Fix names and phrases in subtitles and saved textThis table belongs to this sign-in. Save it with Back up & restore or Export CSV to use it after you sign in again.
AI only: as captions arrive, any part that may hold a name is sent to the AI service with your names (all of them up to 100, otherwise the 60 closest) and waits up to about 1.5 seconds for the answer. Text with no possible name is shown at once. Your exact “heard as” words are still matched on this device.
The AI check isn't available on this server. Names are matched on this device only.
When a name is unsure, an AI service is asked whether it is one of your names. That name and the caption text around it are sent to the AI service, and that caption waits up to a second for the answer.
Recognition hints help Same-language captions and Transcribe only hear the names. Other modes are corrected as text.
Changes apply when you next press Start.
One line per name: what was heard, then => and the correct form. Put phrases in quotes. A line with only => Name still catches similar-sounding mishearings after a title or mid-sentence. Start a line with a language code (zh-Hans:) for that language only.
No names yet. Add them above, import a CSV, or restore a backup.
These words are never changed or hidden unless a title such as Mr or Encik comes right before them.
This preview uses matching on this device only. In AI only mode, live captions are checked by the AI instead.
Only text is corrected. The translated voice still says the name as it was heard.
Event messages
Pop-ups on phones before and after the eventGenerate a summary?
Create a summary, decisions, and action items from this session.
Generating summary…
AI translation may contain errors. Verify critical information before acting.
Recordings
Encrypted in your browser before upload. Playback requires authorised access to the recording key.
Reports
Usage across every conversation type — One to Many, One to One, subtitles, campaign stations and Companion audiences. Minutes and money are counted from recorded activity; transcripts are never read.
By translation experience
| Translation experience | Activities | Billed minutes | Audio minutes (in / out) | Cost / min | Provider cost | Bill / min | Billable | Margin |
|---|
| Day | Activities | Billed minutes | Active minutes | Audio minutes (in / out) | Provider cost | Billable |
|---|
Sessions
| Started | Person | Conversation type | Translation experience | Language | Minutes | Audio minutes (in / out) | Provider cost | Billable |
|---|
History
Your sessions and their transcripts, encrypted in your browser.
AI translation may contain errors. Verify critical information before acting.
Account
Magic Login
Sign in on another device without typing your password. Two-step verification still applies. Do not share your personal QR.
Two-step sign-in
Add an authenticator app for two-step sign-in. You'll get one-time backup codes — keep them somewhere safe.
Scan this QR with your authenticator app, or enter the setup key manually.
Keep this key private. Choose a time-based code (TOTP) when adding it manually.
Settings
Defaults saved in this browser and applied to your next session.
Back up & restore this console
Download an encrypted copy of this login’s console settings, branding and reusable QR references. Restore it here or in another browser with a separate backup password.
Includes saved defaults, languages, audio routing, captions and per-language sharing choices. Account security, team policies, campaigns, recordings, transcripts and one-use Magic Login links are not included. Save changes in Settings first.
Review restore
Reset this console
Puts this console back to its defaults: name corrections, Companion, participant and remote speaker links and QR codes, event branding, saved defaults, languages, audio routing and caption choices. Download a backup first if you may want them again.
Recordings, saved transcripts, campaigns, team settings, your theme and interface language are not affected.
Team
Phones and links
Set the maximum audio access for audience and participant phones.
Audience links switch within a minute. Participant and campaign phones switch when they next reconnect, within five minutes. Delegate station equipment is unchanged.
Autosleep: extra languages shared only to phones run only while someone is listening, to save cost.
Operators may choose a shorter wait, never a longer one or Off. Leave empty for no limit.
Live now
| User | Session | Started | Running | Status | Actions |
|---|
Live listeners
Members
| Member | Role | Daily cap | Status | Actions |
|---|
Glossary
Correct names and key terms in subtitles.
Vocabulary Augmentation
Preferred spellings for names, brands and technical terms in Same-language captions and Transcribe only. Up to 100 terms. Changes apply on the next Start; hints improve recognition but cannot guarantee a spelling. This does not replace the translation glossary.
| Preferred spelling | Actions |
|---|
Billing — last 30 days
| Day | Engine | Minutes | Bill |
|---|
Platform
Platform administration
Rate limits & recovery
Temporarily disable abuse limits for testing. Spending, storage and concurrency limits stay enforced. This switch ends with your login session.
| Date (UTC) | Limit type | Rejected requests |
|---|
Name matching
Which engine decides unsure names in captions. The fast engine falls back to the standard one on any failure, and uses only the standard one for 5 minutes after two failures in a row. Captions in scripts without capital letters always use the standard engine.
Tenants & limits
| Tenant | Sessions / hour | Sessions / day | Minutes / day | Retention (days) | Max session runtime (seconds; blank = platform default, 0 = clear) | Actions |
|---|
Rate card
A new version applies to activity that starts after it is added. Earlier activity keeps the price it started with.
Email delivery (SMTP / Postmark)
Used for invitations. Values saved here always override the server's .env settings.
Cross-tenant billing — last 30 days
| Tenant | Minutes | cost | bill |
|---|
Audit
Audit integrity
Recompute the tamper-evident audit chain for your team and confirm it hasn't been altered.
Audit trail
| Time | Actor | Action | Detail |
|---|