Operating in India
Two facts shape everything here. Commercial SMS in India requires DLT registration with the telecom operators — entity, header and template — and the carrier on the live account carries no outbound SMS at all. Voice and WhatsApp are therefore the working channels, and the Phone screen states the messaging position rather than offering a thread that cannot send.
What DLT registration involves#
DLT is a process with the telecom operators, not a setting in Connect. Three registrations, in order, each with its own approval:
- Entity registration — the business itself, registered as a principal entity with an operator.
- Header registration — the sender ID that appears on the message. Headers are approved per use, and a promotional header is not a transactional one.
- Template registration — the exact message text, with variable fields marked. A message that does not match an approved template does not go, and a template is registered per language variant.
dlt.py holds the registration data once it exists, and the SMS · DLT screen is where it lives in the app. Nothing in Connect shortens the approval process, and nothing in Connect can send around it.
Why outbound SMS still does not work#
What does work on SMS is the inbound half, and it works properly: inbound messages are received and deduplicated, attached to the right relationship, and STOP is honoured and recorded as a suppression like any other opt-out. A business can therefore receive and respect text messages today; it cannot originate them on this carrier.
The channels that do work#
| Channel | Position | What to know |
|---|---|---|
| Voice, inbound and outbound | Available | A carrier handles the PSTN leg; the realtime path adds SIP and a live speech model. The reply floor is 3.3 s median on the best measured call |
| Available | Per-workspace provider credentials. Templates, session windows, media and delivery reporting are provider-dependent and are not simulated | |
| Available | Gmail, Microsoft Graph or any IMAP/SMTP server; the least constrained channel of the four | |
| SMS inbound | Available | Received, deduplicated, attached to a person; STOP honoured |
| SMS outbound | Foundation | Provider-dependent, and none on the live carrier |
For most Indian businesses the practical substitution is WhatsApp for what would have been SMS, and voice for what would have been a call centre. Both carry their own constraints, and the WhatsApp ones come from the provider rather than from Connect.
Voice, locally#
A regional speaking-style layer shapes how a reply is delivered — pace, phrasing, the shape of a greeting — while the human voice profiles set character. Both are workspace configuration in Voice Lab rather than something a caller selects, and both are worth testing with real callers before being fixed.
Keep a profile prompt block under roughly 2,500 characters. Instructions of about 7,500 characters produced a first token in 1.2–1.8 s; at 9,600 characters the same measurement was 2.3–3.4 s, and that cost is paid on every turn of every call.
What it costs here#
Voice is the expense, as it is everywhere: audio tokens metered at several times the rate of text, priced per call, with a thirty-minute ceiling and a budget check that stops a call the remaining allowance cannot cover. There is no separate charge for operating in one country rather than another, and this documentation quotes no prices at all.
The cost people forget is DLT itself: time, operator approvals and a template discipline that outlives the registration. Budget that as project work before assuming SMS is a cheap channel.
Questions#
If we complete DLT registration, can Connect send our SMS?
Only if a provider that actually carries SMS is configured. DLT and carriage are two separate requirements, and the live carrier fails the second one. Completing DLT with no SMS-carrying provider produces approved templates and no way to send them.
Can Connect receive SMS from customers today?
Yes. Inbound messages arrive, are deduplicated, and attach to the right relationship, and STOP is honoured and recorded as a suppression. The channel is described as foundation because of the outbound half, not because inbound is unfinished.
Which language does the voice answer in?
The one the caller is speaking. This documentation carries no supported-language list and will not invent one; the regional speaking-style layer and the human voice profiles are what is recorded, and a real test call on your own workspace is the dependable check.