Live interpreter: two-way translation on a call

The interpreter puts a simultaneous translator between two people. Each side speaks their own language and hears the other's, on a phone call or in a browser tab.

It is not an agent. There is no prompt, no knowledge base, no tools — it never contributes anything of its own. If you want something that answers on your behalf, see voice agents.

What it looks like in practice

Both directions run at once. You speak Turkish, they speak Arabic; each of you hears the other in your own language, with their real voice mixed in quietly underneath so the conversation still feels like a conversation rather than a relay.

Two ways to run a session

Over the phone

An outbound call, started from the dashboard or the API. The order matters: we ring you first, and only dial the other party once you have picked up. Nobody is left listening to silence while the other end is fetched.

Interpreters → your interpreter → Start call, enter the number.

In a browser

Every interpreter gets a permanent web address like https://call2me.app/t/k7m2p9x4w. Share it and the other person opens it — no account, no install. Video works too.

Optionally require a one-time passcode: a 6-digit code, valid 24 hours, consumed on first use. Good for "this link is for one meeting only."

The room holds two people. A third gets a clear refusal rather than a degraded session.

Setting one up

You need at least one purchased phone number first — the interpreter uses it as the caller ID for outbound calls. Without one, creation returns 402 and the dashboard tells you so up front.

Interpreters → New:

Languages

31, both directions, any pair:

Turkish · English · German · French · Spanish · Italian · Arabic · Russian · Chinese · Japanese · Korean · Hindi · Portuguese · Dutch · Polish · Swedish · Ukrainian · Indonesian · Greek · Romanian · Czech · Danish · Finnish · Norwegian · Bulgarian · Hungarian · Slovak · Vietnamese · Malay · Tamil · Croatian

Regional variants normalise to the base language — tr-TR is treated as tr.

Pricing

The owner pays, always — including when a guest joins your web link. Rates are per minute of session:

The phone rate is higher because a phone session carries two PSTN legs — yours and theirs — on one call record. Nothing is discounted in the browser case; there is simply no telephony to pay for.

Your balance is checked before a room opens: too low and the session does not start (402). During a session we watch it every 20 seconds. As you approach the floor, both sides hear a warning that the call will end shortly; at zero the session is announced and closed. Nobody gets cut off without notice.

Latency, honestly

Translation typically lands in well under a second, sometimes up to about a second and a half. The first phrase or two of a session can take several seconds while the model warms up.

Turkish, Japanese and Korean are structurally slower in both directions: the verb comes last, so a sentence cannot be translated faithfully until it is finished. This is a property of the languages, not something a faster model fixes.

The room shows a live latency indicator so you can see what you are getting.

Limits

• Two participants. One language pair per session.
• No maximum session length. Sessions end when someone hangs up or the balance runs out.
• No inbound service line. Sessions are started by you — an outbound call or a web link you share. A number cannot be dialled into an interpreter today.
• No server-side transcripts. The live transcript appears in the browser and can be downloaded there; we keep nothing afterwards.

API

Copy# Create an interpreter
curl -X POST https://api.call2me.app/v1/interpreters \
  -H "Authorization: Bearer $CALL2ME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Clinic — Turkish/Arabic",
    "source_language": "tr",
    "target_language": "ar",
    "phone_number": "+905551112233",
    "callback_number": "+905554445566"
  }'

# Start a phone session (rings you first, then the target)
curl -X POST https://api.call2me.app/v1/interpreters/$ID/call \
  -H "Authorization: Bearer $CALL2ME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target_number":"+971501234567"}'
# → {"call_id":"call_...","room_name":"interp_call_...","status":"pending"}

# Enable the permanent web address
curl -X POST https://api.call2me.app/v1/interpreters/$ID/web \
  -H "Authorization: Bearer $CALL2ME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"enabled":true}'

# Issue a one-time passcode for that link
curl -X POST https://api.call2me.app/v1/interpreters/$ID/web/passcodes \
  -H "Authorization: Bearer $CALL2ME_API_KEY"

# Who is in the room right now
curl https://api.call2me.app/v1/interpreters/$ID/web/live \
  -H "Authorization: Bearer $CALL2ME_API_KEY"

# Session history
curl https://api.call2me.app/v1/interpreters/calls \
  -H "Authorization: Bearer $CALL2ME_API_KEY"

A read-only API key can list and read; creating interpreters or starting calls needs a full key.

What's next

• Voice agents — something that answers for you
• Phone numbers — buying the number an interpreter dials from
• Pricing — full rate card
• Voices — choosing the voice that speaks the translation