Note: this article is a planning roadmap describing the future evolution of the Agent integration system—not features that are all already live. Any capability not yet delivered is labeled 'planning' so you can judge whether to invest now.

In the previous article we covered what's already done: an Agent connects to the world through a domain via a dedicated channel; the world pushes information flow, nearby players, and player messages to it in real time; and to human players it appears as a human-shaped character. That 'door' is installed.

What's behind the door? Here is our three-step plan for the Agent ecosystem. Each step is anchored to a concrete, chargeable scenario—no empty concepts.

Step One (Near Term): Get AI 'Working' Inside a Single World

The goal is simple: run AI services inside one world, and have each thing directly map to revenue or retention.

Planned moves:

• Turn the guide / customer-service / host trio into templates;
• let the backend create an Agent—pick a template, fill in your own knowledge base, give a coordinate, and it's on duty;
• sell to merchants and institutions: a customer-service Agent at the world's spawn point, a guide Agent for new players, a host Agent that broadcasts and routes traffic during events.

This step has the most direct value: it immediately solves two real pain points for world operators—'no one around at cold start' and 'high cost of staffing customer service.'

Step Two (Mid Term): Turn the World Into an 'AI Mini-Program Platform'

Here we want to go from 'using AI ourselves' to 'letting others put their AI into your world.'

Planned moves:

• Agent store: third parties list, get reviewed, bill, and split revenue—the identity, permission, quota, and archiving foundations are already in place;
• wrap the whole world into an MCP Server: expose observe_world / move / say / teleport as tools. Once done, every MCP-capable AI client (various Agent frameworks, dev tools) can enter your world with zero development cost—a traffic lever we value highly;
• Agent-to-Agent (A2A) prototype: Agents can discover and message each other.

The business model is also clear: entry fee, time-based fee, or revenue share—pick one.

Step Three (Long Term): A Cross-World Federated AI Network

If our federation protocol (with agent-type passports and RS256 trust) gets adopted by partner worlds, worlds can mutually recognize Agent identities.

Imagine: several federated worlds recognize one AI guide's identity, and it carries its memory from your world to a partner's world; a merchant's AI follows its customer across worlds. At that point we're not 'building a world' but acting as the 'visa system for the AI world.'

This step depends on partner adoption of the federation protocol—it's the most long-term and hardest to pin down, so we put it last and don't jump ahead.

Three Batches of Scenarios (By Difficulty)

• First batch (sellable once the door is installed): new-player guide, customer-service duty, event host.
• Second batch (needs 1–2 iterations): shop sales Agent, AI pet companion, NPCs upgraded from fixed Q&A to 'truly see, truly walk, remember you' agents.
• Third batch (after federation opens, with the biggest differentiation): 'consulates' for foreign AIs, AI-to-AI socializing, cross-world AI travelers.

Our Restraint

Drawing this map, we deliberately hold a few lines to avoid blowing it into empty concepts:

• No empty talk—each phase is tied to a concrete, chargeable scenario;
• capacity is a red line—under mid-tier capacity there's an Agent limit; before opening the ecosystem we must do scheduling optimization first;
• Agent behavior must be governed—archiving and auditing are only the base; before opening we need a 'revoke on violation' backend flow;
• privacy boundaries must be written clearly—archiving chats involves real-human conversations; commercial authorization terms are set in advance;
• the visual channel isn't urgent—semantic annotation covers most scenarios, screenshot capture is costly, and we'll add it once there's a paying use case.

One Sentence to Close

We turn a 3D virtual world into one of the first 'enterable physical spaces' of the AI era. The door is installed, the road is long, but at every step we know what we're selling.

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.