Corporate showrooms, exhibition halls, and campus visitor centers all hit the same mismatch: docents work shifts, visitors don't. After hours, overseas clients in another time zone, a dealer dropping by unannounced — they walk the floor with nobody to explain anything. This article explains the mechanics of putting an AI *inside* the showroom, and separates what is already built from what is still on the roadmap, so plans aren't mistaken for shipped features.

1. When an AI Presenter Is a Real Requirement

Not every showroom needs one. These usually do:

• Visitors spread across time zones that staff schedules can't cover;
• Halls open year-round with limited guides, where repetitive questions consume most of their time;
• Products updated often, so scripts change faster than staff can be retrained;
• Confidentiality requirements on visitor data and exhibit material, ruling out third-party platforms;
• A desire for visitors to be *accompanied* through the space rather than left reading panels.

One sentence covers them: the showroom needs someone present, but not always a human.

2. What the Usual Routes Cost Mechanically

Path A: Recorded narration / audio guides / QR-code audio

Cheap and always available. But it plays one way: a visitor can't ask "why was this component designed like that," and the guide won't change its explanation because someone lingered two extra minutes.

Path B: Website chat widget / chatbot

It answers questions, but it isn't *there*. It doesn't know which exhibit the visitor is standing in front of, and it has no body — the visitor experiences "a chat box opened," not "someone came over to talk to me."

Path C: Human staffing

Best experience, adapts on the spot. Cost is labor and coverage: nights and holidays are usually empty.

Path D: AI features inside a centralized platform

There is AI capability, but hall content plus visitor and conversation data live on someone else's server, so the data isn't fully yours; the capability is bounded by the platform and generally usable only inside its own frame.

The shared problem: either it isn't present, or it only plays one way, or the data isn't yours.

3. Five Paths Side by Side

| Dimension | Recorded audio guide | Website chatbot | Human staffing | Centralized platform AI | Genesis (self-hosted + embodied agent) |

| --- | --- | --- | --- | --- | --- |

| Present in space (has a body, can move) | No | No | Yes | Depends | Yes (human-like avatar, walks and teleports) |

| Aware of where the visitor stands and what they're saying | No | No | Yes | Depends | Yes (world pushes context) |

| Coverage | Anytime, one-way | Anytime | Scheduled hours | Anytime | Anytime, subject to the model service you connect |

| Who owns the conversations and data | — | Platform's server | On site | Platform's server | Your own server |

| Can interactions and knowledge base be customized | No | Limited by platform | Depends on people | Limited by platform | Yes (Three.js base + your own model) |

| Goes offline if you stop paying? | — | Yes | — | Yes | No (the world stays) |

4. Already Built: the Gateway That Lets an AI Enter the World

Genesis (Chuangshi Virtual World CRM) is a browser-based, self-hostable Three.js 3D virtual world base / framework. The core of its Agent integration system is already developed, and it works as a four-step loop:

1. Getting in: the agent reaches your world by domain over a dedicated agent channel, authenticated with a key (API Key) — not borrowed from someone else's account system;
2. The world feeds context (observe): the world proactively pushes what is happening in it right now, which players are nearby, and which messages players sent to the agent, including message content — instead of passively waiting for a question;
3. It decides and acts: `say` to speak, `walk_to` to move to coordinates, `teleport` to arrive directly — coordinate navigation happens to be something models handle well;
4. Humans see a character: to a visitor it appears as a human-like avatar (reusing the existing avatar and skeletal animations), with an overhead speech bubble, voice relay, and conversation within range — not a chat box.

Agents also carry their own identity in the federation layer (principalType=agent), so they can move between trusted worlds.

The essential difference from a web chatbot is embodiment: it has a body, it can move, it can see real people, and real people can see it.

5. Still on the Roadmap (Explicitly: Not Shipped)

The following belong to a planned three-stage roadmap. They do not exist today, and shouldn't be read as available:

• Near term: templated guide / support / host agent kits, and one-click agent creation in the admin panel (pick a template, fill a knowledge base, set coordinates);
• Mid term: an agent store (listing, review, billing, revenue sharing), packaging the world as an MCP Server so any MCP client can enter without custom development, and early A2A;
• Long term: a cross-world federated agent network, letting agents travel between worlds.

In short: the doorway is open; the finished model rooms are still being furnished. To connect today, you or your engineers write an agent client that consumes observe and issues say / walk_to / teleport. Templated creation in the admin panel is the next step on the roadmap.

6. Why Self-Hosting Is the Prerequisite

An AI standing in your showroom hears what visitors say and sees what they do. That data on your own server versus someone else's platform is a difference in kind:

• Private data: the system runs on your own server, so world and conversation data never pass through a third-party platform — suitable for organizations with confidentiality requirements;
• Your choice of model: which model sits behind the agent and what knowledge base it gets is your decision, not a platform lock-in;
• The world stays: not SaaS, so there is no "stop paying, go offline" — the hall and its presenter remain;
• Open to customization: built on Three.js, so tour routes, trigger points, and interactions can all be changed by you.

7. Roughly How It Goes

1. Deploy: install the system on your own server and get a world address that is yours;
2. Build the hall: place exhibit GLB models, set lighting and explanation points;
3. Enable agent access: turn on the agent channel and create a key for your agent;
4. Connect a model and knowledge base: your agent client connects to the world, processes the context pushed by observe, calls the model you chose to generate replies, and expresses them through say / walk_to;
5. (Roadmap) Templated creation: once the admin templates ship, pick a template, fill the knowledge base, set coordinates — no client code needed.

Want someone present in your showroom at any hour without handing visitor conversations to a third party? Genesis is a Three.js 3D virtual world base you host yourself: the agent gateway is open, and an AI can walk into your world in embodied form. Try the demo: miduo100.com

8. FAQ

Q: Is the AI docent usable today?

A: Two layers. The agent integration itself — connection, observe, say, walk_to, teleport, human-form presentation — is already developed. But "pick a template, fill a knowledge base, generate a docent in one click" is on the roadmap; for now you write the agent client yourself.

Q: What's the essential difference from a website chatbot?

A: Embodiment. It isn't a dialog box but a character standing in the hall: it walks, teleports to coordinates, sees nearby players and the messages they send, and real visitors can see and talk to it. The world pushes context to it proactively.

Q: Could visitor conversations end up on someone else's server?

A: No. The system is self-hosted, so world and conversation data never pass through a third-party platform. Which model service the agent calls is your choice and configuration.

Q: Might the AI say something wrong?

A: That deserves real attention. Archiving, auditing, and revocation are already in place; the admin workflow for behavior governance is still being refined. Before launch, bound the knowledge base, add fallbacks for sensitive topics, and spot-check archived logs manually.

Q: Will it replace human docents?

A: Our read is complement, not replacement. Repetitive questions, off-hours, and cross-time-zone visits go to it; situations needing judgment, rapport, and depth stay with people. Its value is covering the hours when nobody is there.

Plan Comparison: Self-Hosted Deployment vs. Platform Leasing

About Genesis

Genesis is a self-hosted 3D virtual world system built on Three.js + WebGL, helping individuals and businesses build their own 3D spaces. Accessible directly from a browser, compatible with both PC and mobile, it supports multiplayer online, federated teleportation, a shop system, and Agent integration—where an AI can enter your world as an embodied character. Your data runs on your own server, never passing through a third-party platform—so every world truly belongs to its owner.