On-premise
Your voice AI. Your servers.
We install our speech-to-text, text-to-speech and voice agent models directly on your infrastructure. Everything runs inside your network — no audio, text or transcripts ever leave your servers.
- Data never leaves your network
- Works fully offline
- Setup & support by our team
Available models
The same models that power our cloud API, packaged for your own hardware.
| Model | What it does | Languages |
|---|---|---|
| Speech to Text | Real-time and batch transcription, speaker diarization, timestamps | Uzbek, Russian, English |
| Text to Speech | Natural voices, plus voice cloning from a 10-second sample | Uzbek, Russian, English |
| Voice Agent | STT + LLM + TTS pipeline that talks to your customers, tuned to your domain | Uzbek, Russian |
| SIP Gateway | Connects agents to your PBX or SIP trunk (Asterisk, FreePBX and others) | — |
Server requirements
Pick a tier based on how many audio streams you need to process at the same time. We help you size it during the consultation.
| Tier | Streams | CPU / RAM | GPU | Storage |
|---|---|---|---|---|
| Starter | up to 10 | 16 vCPU / 64 GB | 1× NVIDIA L4 or RTX 4090 (24 GB) | 500 GB NVMe |
| Business | up to 50 | 32 vCPU / 128 GB | 2× NVIDIA L40S (48 GB) | 1 TB NVMe |
| Enterprise | 100+ | 64+ vCPU / 256 GB+ | 4× NVIDIA A100 or H100 (80 GB) | 2 TB+ NVMe |
Software
Ubuntu 22.04 LTS, Docker with NVIDIA Container Toolkit, CUDA 12. Internet is needed only during installation — after that everything runs offline.
How it works
- 1
Consultation
We learn your use case, call volume and security requirements, and recommend the right server tier.
- 2
Pilot
We test the models on your real audio so you see the quality before you commit.
- 3
Installation
Our engineers deploy everything on your servers and connect it to your systems.
- 4
Support
Updates, monitoring and model improvements — we stay with you after launch.
Terms
Every deployment is a separate agreement — here is what it usually includes.
- License — Annual subscription or perpetual license, priced by the number of concurrent streams.
- Hardware — You provide the servers, or we help you choose and procure them.
- Customization — Fine-tuning on your vocabulary — product names, addresses, medical or banking terms.
- Support — Updates and technical support are included for the first year, then renewed annually.
- Security — Signed NDA, no remote access without your permission, all data stays with you.
Ready to run it on your servers?
Message us on Telegram — tell us about your use case and we will get back to you with a proposal.