TeaSIP calls your customers from a script, voices it with a neural voice and understands their answers — in Russian, English and Tajik. Entirely on your server. No cloud. No external APIs.
01 · Outbound marketing
Upload a list of numbers, pick a scenario — TeaSIP dials, delivers your offer in a lifelike neural voice, listens to the answer and logs the result in your CRM. One server replaces an entire calling team.
Script · payment-reminder.flow
ActiveDialing
3 attempts · 09:00–18:00
Greeting
AI voice · EN · {name}, {amount}, {date}
Question + keypress
1 — confirm · 0 — operator
Branch “1”
thank you → hang up
Result
CRM webhook · transcript · report
+992 93 ··· ·· 42
waiting…
94%
connect rate
0:47
avg. call
100%
calls in report
Name, amount and due date are pulled from your database. Confirmation takes a single keypress.
The robot reads out the order and delivery time, then records the customer's answer.
A 0–10 score by voice or keypad; comments are transcribed and saved as text.
Thousands of calls per hour. Press 1 — the lead lands in your CRM for a manager.
Confirm, reschedule or cancel an appointment — no receptionist needed.
A tactful scenario with transfer to an operator and a full call recording.
Scheduled retries, 09:00–18:00 calling windows, time zones, per-carrier rate limits.
{name}, {amount}, {date} — the AI voice fits them into the sentence and pronounces numbers, dates and currencies like a person would.
“Press 1” or just say “yes” — speech recognition understands the answer in three languages.
The customer wants a person — the call goes straight to the call center queue along with the conversation context.
Every call result — status, transcript, recording — goes straight to your CRM via webhook or API.
Connection, duration, script branch, recording and transcript — for every number in the campaign.
02 · Artificial intelligence
Speech recognition, the language model and voice synthesis run locally — right next to your telephony. Call recordings and customer data never leave your infrastructure. Ever.
Customer speaks
call · any phone
Hearing
speech recognition · streaming
Mind
local language model
Voice
neural speech synthesis
Customer hears
latency < 1 second
The whole pipeline stays inside your perimeter · GPU or CPU · not a single request to the internet
Streaming speech recognition
The customer is heard and understood from the first word: streaming transcription in Russian, English and Tajik, robust to 8 kHz phone-line audio.
Local language model
The AI agent runs the conversation by your scenario: answers questions, clarifies details and triggers actions — create a ticket, reschedule an appointment, transfer to an operator.
Neural speech synthesis
Voices you can proudly put in front of customers: natural intonation, correctly spoken numbers, dates and currencies. Script voiceover is instant — no studio, no voice actors.
When the customer interrupts, the AI stops instantly and listens. Voice activity detection tells speech from street noise, and end-of-utterance detection never leaves the customer waiting in silence. The conversation feels human.
03 · Call flow builder
Every scenario compiles to native telephony-engine code and runs in milliseconds — no interpreters, no lag between steps.
promo-campaign.flow · v7 · published
✓ validatedAnswer, play, ask, dial, recognize, send to a queue — build a scenario from ready-made blocks.
“No answer — retry in an hour, at most 3 times” is a single arrow. Conditions, variables, counters — all executed faithfully.
Every publish is an immutable version. Something went wrong? Roll back in one click — calls in progress are unaffected.
Place a test call and watch the path light up through the scenario — node by node, with live variable values.
04 · Call center
AI takes over the routine, people handle the hard cases. TeaSIP is also a complete contact-center platform: queues, agents, quality control and all of your company's telephony.
Distribution strategies from “least busy” to “ring all”. Hold music, priorities, agent statuses.
Who is talking, to whom and for how long — a live dashboard with no page reloads.
A supervisor listens in, whispers to the agent (the customer can't hear) or joins as a third party — in one click, no SIP phone required.
Stereo recording: customer on the left, agent on the right. Automatic transcription — search conversations like documents.
Agents don't need a phone — just a headset and a browser tab. Transfer, hold, DTMF.
Every call is logged: who, where to, how long they waited, how it ended. Reports by agent, queue and campaign.
05 · Security and privacy
Cloud dialers listen to your customers through someone else's ears. TeaSIP installs entirely on your infrastructure — AI included.
A single Docker Compose stack on your server or in your data center. Internet access is only needed for your carrier.
Recordings, transcripts, the customer database — everything is stored and processed inside your perimeter. Not a single byte goes to someone else's cloud.
Administrator, developer, operator, observer — everyone sees only their part. Every action lands in the audit log: who, what, when.
Protection against toll fraud to premium destinations — the biggest source of telephony losses: rate limits, blocklists, an expensive-call guard.
Subscriber passwords and carrier keys are encrypted at the application level. The database holds only hashes and ciphertext.
IP filters, automatic brute-force blocking, isolation of the engine's internal networks.
06 · Languages
Global cloud services don't understand spoken Tajik. TeaSIP recognizes it, speaks it and holds conversations in Tojikī — as well as in Russian and English. One scenario, three languages.
Салом!
Тоҷикӣ
recognition · synthesis · AI agent
Здравствуйте!
Русский
recognition · synthesis · AI agent
Hello!
English
recognition · synthesis · AI agent
07 · Questions and answers
You upload a list of numbers and choose a script — the conversation scenario. The platform dials the numbers itself, reads the text in a neural voice with the customer's name and data filled in, listens for the answer (keypress or voice) and records the outcome of every call: status, scenario branch, recording and transcript.
Yes. Speech recognition, voice synthesis and the language model are deployed locally, on your server, next to the telephony. Only the call itself goes out, through your carrier. Recordings and customer data never leave your infrastructure.
Russian, English and Tajik — in all three roles: speech recognition, voice synthesis and the AI voice agent. Tajik support is unique: cloud platforms don't offer it.
No. The robot handles typical campaigns on its own — reminders, confirmations, surveys, promos. If a customer wants to talk to a person, the call is instantly transferred to the call center queue along with the conversation context.
Around 500 concurrent conversations and 10,000 registered subscribers per installation. For campaigns, that means thousands of connected calls per hour.
With a ready-made script — in 10–15 minutes: upload the numbers, pick a voice, press “Start”. Custom scenarios are drawn in the visual builder, no programmer required.
A Linux server with Docker (a GPU for the full AI stack; a CPU profile is available too) and a SIP connection to your carrier. Installation and setup are done turnkey by the Hizmatrason team.
30-minute demo · a live call from the AI agent to your phone
Tell us about your use case — we'll build a script for your business and call you in the TeaSIP voice. Turnkey implementation by the Hizmatrason team.