AI Assistant
YunoBot: an on-device chatbot without an LLM
YunoBot runs in your browser with no language model: a C++ core compiled to WebAssembly classifies your question and answers by quoting the matching passage from this site, in English, Turkish or both. In a network test, none of a message's text left the page. Say hello, ask about my work, or explore a project or journal topic, and follow the source links.
How does YunoBot work without a language model?
A C++ core compiled to a 913,776-byte WebAssembly module turns your message into hashed word and character features, scores 82 conversational intents with a 96-unit neural layer (401,363 parameters) and, for factual questions, ranks passages from this site with BM25-style term weights. It selects, composes or quotes replies; it does not generate free text. How it works, with its measured accuracy.
Does YunoBot send my messages to a server?
In a network test on 3 October 2026, no request made while loading the page or chatting contained any word of the test message. The reply is computed by a WebAssembly module inside the page. The site's analytics tags still load on every page. The test, and how to repeat it yourself.
More quick answers
- Which languages does YunoBot understand?
- English, Turkish and mixed-language messages.
- What can I ask it?
- Questions about this site's projects, writing, background and places, plus light small talk. When nothing matches well, it says it is not sure instead of guessing.
- How big is the download?
- The WebAssembly core is 913,776 bytes and the bundled source pack is 405,655 bytes. Both are fetched when the page loads, and the service worker can keep them for offline use.