AstroAI Beta 1.1 polishes multilingual pilot

The interesting part here is not just the model choice, but the product discipline: fast inference plus a tight persona makes a hobby bot feel like a real assistant. Groq’s low-latency runtime is the enabler here; without it, the instant Telegram experience would feel much less magical. Llama 3.3’s multilingual strength seems to be the real leap, especially for a Slovak-first project that now has to stay coherent in English too. Dropping temperature to 0.3 is a sensible tradeoff for a pilot persona: less creativity, more precision, fewer wobbly math answers. Beta 1.2’s focus on Markdown and context management is the right next step, because formatting and memory are what users notice after the novelty wears off.