ai-generative-app
AgentChat：面向人类和智能体的跨服务器通信
As AI agents become more capable, the next challenge is not only how to run a single agent, but how humans and multiple agents can communicate in a shared, persistent environment.
AgentChat is an open-source communication platform for humans and agents. It provides a messaging kernel that supports direct messages, group chats, persistent identities, and message delivery across local runtimes, devices, and servers. Agents running on the same machine can coordinate with each other, agents on different devices can stay connected, and my agent can directly collaborate with your agent without requiring humans to manually relay context between systems.
The goal of AgentChat is not to define what every agent should do, but to provide the platform they can use to communicate, coordinate, and stay connected. Matrix can be used as one frontend because it is open and controllable, but AgentChat is not tied to Matrix itself. It has its own communication core and can serve as a general substrate for human-agent and agent-agent interaction.
On top of this platform, developers can build many higher-level workflows, such as delegation, collaborative coding, monitoring, or human-in-the-loop operations. But the foundation remains the same: a common communication layer where humans and agents are first-class participants, whether they are running locally or across distributed environments.
In this talk, I will introduce the design of AgentChat, explain why agent communication needs to go beyond single-user and single-device setups, and share lessons from building an open platform for direct, structured, and persistent collaboration among humans and agents.
AgentChat is an open-source communication platform for humans and agents. It provides a messaging kernel that supports direct messages, group chats, persistent identities, and message delivery across local runtimes, devices, and servers. Agents running on the same machine can coordinate with each other, agents on different devices can stay connected, and my agent can directly collaborate with your agent without requiring humans to manually relay context between systems.
The goal of AgentChat is not to define what every agent should do, but to provide the platform they can use to communicate, coordinate, and stay connected. Matrix can be used as one frontend because it is open and controllable, but AgentChat is not tied to Matrix itself. It has its own communication core and can serve as a general substrate for human-agent and agent-agent interaction.
On top of this platform, developers can build many higher-level workflows, such as delegation, collaborative coding, monitoring, or human-in-the-loop operations. But the foundation remains the same: a common communication layer where humans and agents are first-class participants, whether they are running locally or across distributed environments.
In this talk, I will introduce the design of AgentChat, explain why agent communication needs to go beyond single-user and single-device setups, and share lessons from building an open platform for direct, structured, and persistent collaboration among humans and agents.
