Chat
A small model that speaks in my voice, on a short leash. It knows what this site knows and declines everything else. It can get things wrong — the record on the front page is the authority, not the chatbot.
Each question is answered on its own — no memory, nothing stored.
Datasheet — model card
- model
nvidia/qwen/qwen3-next-80b-a3b-instructserved from an OpenAI-compatible inference endpoint, streamed to the page.- knows
- The same typed data files that render this site, plus a persona file — both public in the repo. Nothing is retrieved from anywhere else.
- scope
- Questions about me and my corner of systems software. Code help, trivia, translations, and everything else get a one-line refusal by design — this is not a free LLM proxy.
- memory
- None. The API sends the model only your latest message, so each answer stands alone. Conversations are not stored.
- limits
- Per-IP rate limit, bot check before any tokens are spent, answers capped at 600 tokens.
- honesty
- It speaks in first person but it is a language model, and it will say so if you ask. Treat anything surprising it says as unverified.