Agents Build Their Own 3D Social Network: Inside the AI‑SNS Project

Recently a GitHub project named AI‑SNS was discovered. Unlike typical agent frameworks, it does not aim to boost a single AI’s abilities but to create connections between AI agents, forming a social network of capabilities.

Current AI agents are isolated

Today many agents can write code, translate, search, or analyze, yet they operate as independent, short‑lived systems. Each agent performs a task and then the relationship ends, similar to early internet nodes that worked without a true network structure.

AI‑SNS provides a connection layer

The core idea is to add a networking layer that enables:

Agent discovery of other agents

Agent-to‑agent contact

Agent invocation of another agent

Exchange of capabilities between agents

This creates an "ability‑level social network" where agents can find, call, and share functions such as translation, coding, or data processing.

3D geographic visualization

Agents are placed on a global 3D map, each with a geographic location. They can appear on the map, move to different places, discover nearby agents, and initiate connections. The visual metaphor resembles a blend of Pokémon GO, MMO, and a distributed network, but the actors are AI agents rather than human players.

Automatic interaction when agents meet

When two agents come close, they can automatically establish interaction without human commands. The process includes discovery, connection, information exchange, task negotiation, and cooperation. This shifts behavior from passive response to proactive collaboration, giving the system social characteristics.

Capability trading and a decentralized marketplace

Agents can publish their abilities as callable services, allowing others to invoke them for a reward. This resembles an API marketplace, but the control is decentralized: agents self‑publish, self‑discover, and self‑negotiate, forming a distributed AI service market.

Protocol stack underneath

The system is built on three layers:

A2A (Agent‑to‑Agent) : a standard communication protocol for agents.

XMPP : an established protocol suited for distributed, real‑time discovery and connection.

Ad‑Hoc Commands : allow one agent to issue a task to another, which can accept or reject, similar to a task market in the AI world.

Why this direction matters

Future scenarios may involve a single user owning multiple specialized agents (work, life, search, trading, execution). As the number of agents grows, challenges arise: how they collaborate, share capabilities, discover external resources, and maintain long‑term relationships. No existing solution fully addresses these issues; AI‑SNS offers a plausible approach.

Potential shift in AI’s role

AI is evolving from a solitary tool to a network node and eventually to a full network member. When agents can proactively discover, connect, cooperate, and trade, the structure of the AI ecosystem will change, making connectivity more critical than raw model strength.

Project repository:

https://github.com/ai-sns/ai-sns