Can humanoid robots talk? What the X2 needs
Yes, within limits worth knowing before you book one. The AGIBOT X2 carries a microphone array, a wireless microphone and a speaker, and AGIBOT lists speech among its interaction modes, so it can hear a visitor and answer out loud. Out of the box it says what somebody scripted. Holding an open conversation takes a language model, and that account is yours, not the robot's.
| Layer | What it does | Who supplies it |
|---|---|---|
| Microphones and speaker | Hear the question, speak the answer | AGIBOT, built into the robot |
| Scripted speech | Greetings and fixed answers, the same every time | Set up with the robot |
| Language model | Answers questions nobody scripted | Your account: OpenAI, another provider, or a local model such as Ollama |
| Your documents | What the model may answer from | You, kept current by someone on your side |
What does the X2 hear and speak with?
AGIBOT's store lists a microphone array, a wireless microphone, a speaker, an interactive screen and lighting effects on both X2 versions. The published table for the X2 Basic names four interaction modes: speech, gesture, facial expression and touch. The larger A2 Ultra lists a microphone array among its sensors and carries a facial screen, and the wheeled G2 lists a four-microphone array.
None of these tables publishes how many languages the robot understands or how quickly it replies, so we do not quote either. Both depend on the software running on it, which is the next point.
What does a language model add?
Without one, the robot says what you scripted, reliably and forever. That suits a greeting at a door or a short introduction on a stand. With one, it answers questions you did not anticipate, sometimes well. The microphones take the question, the model writes an answer, and the speaker says it.
What decides whether the answers are any good is what the model reads. Our Client LLM Integration sets up a backend where you upload documents and folders, indexes them, and has the robot answer from those sources rather than from general knowledge. A hotel's check-out time, a shop's opening hours and this week's promotion come from your files, not from the model's guess.
What do you have to supply?
Three things. First, the model account: OpenAI, another cloud provider, or a local deployment such as Ollama on your own hardware. Usage is billed to you directly, and no token or usage cost passes through us. The integration takes different providers, so which model runs stays your decision and stays reversible.
Second, the documents, and a named person who keeps them current. A model answering freely about prices will sooner or later quote one that changed last week.
Third, a decision on what the robot does when it does not know. A robot that hands over to a person too readily is better than one that improvises, and that choice belongs in the requirements document before anyone builds it.
Which X2 do you need for it?
AGIBOT's table marks customer-side development as supported on the X2 Ultra and not supported on the X2 Basic. Connecting your own model is development work on the robot, so we plan a conversation project on the X2 Ultra. It rents at 2,500 EUR per week, net of VAT and shipping, and an in-stock unit is with you inside 48 hours of a confirmed booking. Engineering time to set up the integration is 825 EUR per engineer per day on site, travel and accommodation included within Europe.
A week of rental is a cheap way to learn the two things that matter most: whether your visitors talk to it at all, and what they ask.
If you want the X2 to hold a conversation in your lobby or on your stand, ask us to scope it: the requirements document first, then your language model on a rented X2 Ultra.
