Kazakh, inside your application
Conversation, grammar and speech-to-text — Kazakh and Russian. The models run on our own GPUs in Kazakhstan and connect like OpenAI: change base_url, keep your code.
Try it
Ask right here
No sign-up, no key. The same request you will later make from your code.
Speech to text
An hour of audio in four minutes
til-asr turns Kazakh and Russian speech into text and restores punctuation and capitals itself. No need to name the language: it understands both and switches as it goes.
{
"duration_seconds": 3603,
"text": "Сәлеметсіздер ме, білім алушылар…",
"words": [
{"word": "Сәлеметсіздер", "start": 0.42, "end": 1.08},
{"word": "ме", "start": 1.08, "end": 1.31}
]
}Who it is for
For those who need Kazakh specifically
General models know Kazakh only on the surface: they confuse cases, lose the nine letters, invent facts about the country. Ours are trained on Kazakh and tested on it.
Transcribing lectures and meetings, helping students, processing Kazakh archives.
Kazakh in your app without your own infrastructure: same code, different base_url.
Data does not leave the country: the models run on GPUs in Kazakhstan.
Models
What the platform does
Answers, rewriting, error correction, document analysis. Works with the OpenAI SDK.
Kazakh and Russian, punctuation and capitals, a timestamp per word.
Embeddings for searching your own documents and knowledge bases.
Available to your key
Conversation and text work: Kazakh, Russian, English and code.
Kazakh grammar: fixes mistakes, summarises, changes style.
The platform picks the model for the task.
Speech to text with punctuation. Kazakh and Russian together, no need to name the language.
Getting started
Your first request in a minute
The task travels in the model name. A short "qazgramma" means grammar correction on the free tier.
- 1Create an organisation — one form, no emails.
- 2Issue a key in the portal: Developers → API keys.
- 3Put the key into the request and run it.
curl https://router.tilqazyna.kz/v1/chat/completions \
-H "Authorization: Bearer ВАШ_КЛЮЧ" \
-H "Content-Type: application/json" \
-d '{"model":"qwen","messages":[{"role":"user","content":"Сәлем!"}]}'Terms
Limits, honestly
A new organisation starts with 100 000 tokens per month.
Up to 30 requests per minute.
You can raise it to 200 000 yourself; more is a request to us.
Documents up to 15 MB, formats: txt, md, csv, pdf, docx.
Will we have to rewrite our code?
No. The platform is OpenAI-compatible: change base_url and the key, keep your SDK and the rest of your code.
What happens to our data?
The models run on our own GPUs and requests do not go to third parties. Organisations are isolated from each other. The request body is stored in the platform log — it is needed to investigate failures and count usage.
Where are the servers?
In Kazakhstan. That is the reason this platform exists separately from foreign services.
What does it cost?
A new organisation gets a free quota straight away, enough to get acquainted and run a small project. More is a request to us.