AI that thinks and speaks Turkish
Start chatting with Olgu AI right away, no account needed.
You're currently chatting with our first model — a 1-billion-parameter proof of concept. Larger, more capable models are in development.
Why Olgu AI?
We're building a language model designed for Turkish from the ground up — not a general-purpose model with Turkish bolted on afterward.
A Turkish-specific tokenizer
60% lower token costOlgu AI's Turkish-specific tokenizer avoids needlessly splitting suffixes, cutting token usage by 60%.“geliyorlardı ("they were coming")”
General-purpose model
Olgu AI
Illustrative example — a tokenizer trained from scratch for Turkish doesn't needlessly split inflectional suffixes.
Native-level nuance
Generic model (translation-flavored)
"Eline sağlık" → attempted word-for-word, context-blind translation.
Olgu AI
"Eline sağlık" is an expression of thanks/appreciation — produces a natural, context-aware response.
Domestic, independent, uninterruptible
Runs without depending on a foreign provider — instead of trusting an outside system that could be cut off overnight.
Built for institutions and defense
SECUREFrom banking to public institutions to the defense industry — a homegrown technology base built to be trusted with critical infrastructure, prioritizing data sovereignty.
Sensitive Financial Data Security
Customer data is processed with high accuracy on local servers, never leaving the institution.
How it works
Get started in three steps, no account required.
Type your question
Write anything on your mind into the chat box on the homepage.
The model processes it
Olgu AI analyzes your question with Turkish context in mind and forms a response.
Get your answer
A natural, fluent response appears within seconds.
What will become possible
Not for you in general — for exactly who you are. Pick an example.
Step-by-step homework and exam support
3 min
Time to solve and explain a complex problem
- Shows formulas and solution steps in clean, structured form
- Explains using examples matched to Turkish curricula
- Generates extra practice questions to reinforce the topic
Why a national language model?
AI will be one of the most critical technologies of the next decade — and no country can afford to stay entirely dependent on foreign providers for it.
- Especially in defense and critical infrastructure, relying on a system that could be cut off from outside overnight is a real risk.
- Thanks to our Turkish-specific tokenizer, our models process Turkish more efficiently and run more cheaply.
- More affordable service for individuals; trustworthy infrastructure that protects data sovereignty for institutions.
- Long-term technological independence for the country.
3x
Faster Turkish processing
1B+
Custom-parameter model
100%
Domestic infrastructure assurance
Frequently asked questions
The model available today is our first, from-scratch 1-billion-parameter model. It was our first full end-to-end run — from data collection through tokenizer to training — proving out our whole approach in practice. We're now building larger, more capable models on that foundation.
To be the strong, homegrown AI company Turkey needs, and to reduce the country's reliance on foreign AI technology. By building models that speak Turkish at a native level, we aim to serve individuals in daily life, institutions as trustworthy infrastructure, and — over the long term — national security as a domestic technology base.
We aim to serve three main groups: individuals who want to benefit from AI in daily life, Turkish companies and public institutions that want to process their data without it leaving the country, and, over the long term, national-security-related institutions.
Core chat features are free to use without an account. You can create a free account to save your chat history.
Yes. We use KVKK-compliant infrastructure and explain exactly how your data is processed in our KVKK Disclosure Notice. Hosting your data entirely within Turkey is a long-term goal we're actively building our infrastructure toward.
Our first model is best suited to simple conversation for now. Broader capabilities like correspondence, summarizing, and coding help are on our roadmap as we scale to larger models.
We validated our approach with our first 1-billion-parameter model. Next, we're scaling to several billion parameters; over the long term, our goal is an internationally competitive, national-scale model.
In the short term, we'll keep shipping a chat experience with steadily expanding capabilities. Over the long term, our goal is a fully homegrown AI platform you can rely on across everyday tasks and professional use cases alike.
Yes — the platform fully supports English today. The model's primary language is Turkish, but it can already hold conversations in English as a secondary language; on a few specific topics its English responses may be a bit weaker than its Turkish ones.
Create an account for more
Save your chat history and pick up where you left off.