01
Caller taps talk
Voice starts right inside the chat widget — no phone number, no app install, microphone permission only.
AI voice agent
An AI voice agent that
talks back — and knows the answer.
Customers press one button and speak. Softsasi Agent listens, retrieves the answer from your own documents, and replies out loud in Bangla or English — logging the transcript and handing over to a human whenever the caller asks.
Test the voice agent right here
Tap the mic, ask a question in Bangla or English, and hear the agent answer out loud — the same speech pipeline your customers get.
Live voice test
Talk to the agent
Tap, ask a question, tap stop
What the voice agent does
- Speaks like a person
- High-quality neural speech with natural pacing, so callers get a conversation instead of a menu tree.
- Bangla and English, mid-call
- Set the languages your agent supports and switch during the call. The reply comes back in the language the caller is using.
- Same knowledge as the chat agent
- The voice turn retrieves from the identical document set, so spoken and written answers never contradict each other.
- Hands the call to a human
- One tap escalates: the agent goes quiet, the conversation appears in your inbox, and the call is logged as transferred.
How a call flows
02
We hear the turn
Speech is transcribed the moment the caller stops talking, using silence detection rather than a push-to-talk button.
03
The agent retrieves
The transcript is matched against your knowledge base and tools before a single word is spoken back.
04
It answers out loud
The reply is streamed as speech while the transcript is written to the conversation for your records.
Admins get the receipts
Each call is stored with its duration, language, outcome and transcript. Filter by outcome to find transferred or abandoned calls, read what the agent said, and check voice minutes against your budget on the billing dashboard. The same inbox holds chat, WhatsApp and Messenger threads, so one team works one queue. Want the written channel first? See the AI chat agent.
Questions callers' teams ask
Do callers need to install anything?
No. The voice call runs in the browser from the same chat widget you already embed. It asks for microphone permission and nothing else.
Which languages does it handle?
You choose the languages per agent — Bangla and English are configured out of the box, and callers can switch mid-call.
Can I read what was said?
Yes. Every call produces a record with duration, language, outcome and a full transcript in the agent's call log.
What happens when the agent doesn't know?
It says so plainly and offers to bring in a person. The escalated thread lands in the shared inbox with the transcript attached.
Is the voice agent grounded like the chat agent?
It uses the same retrieval pipeline over the same documents, so there is one source of truth for both channels.
How is voice usage billed?
Voice minutes are metered per organisation and visible on the billing dashboard alongside conversation counts, with budget stops you control.
Let customers hear your agent
Trial starts at $1/mo. Publish an agent, enable voice, and take your first call today.