A chatbot talks. This assistant also acts: it decides for itself which tool it needs, actually calls it, reads the real result — and only then writes an answer. Every step below is a real record of what actually happened; nothing is retold or made up.
Type a request
0 / 300
Or try
Under the hood: one language model (Groq, Llama 3.3) decides for itself whether and which tool it needs — looking up the weather via Open-Meteo, running a calculation through a safe calculator, or a fact via Wikipedia. Those tools are actually executed here on the server; the model gets the real result back and then decides the next step. That whole process takes a few seconds — this page shows it afterwards, slowed down, step by step, so you can follow along.
Honest about the limits: the model can sometimes skip a tool where it would have helped, or misread a question. And as with any language model: the final answer is a rephrasing by the AI, so for anything that matters, always check the raw tool results above.