OpenClaw adds Multiplayer so teams can work with the same AI agent

OpenClaw is promising to break down workplace silos with something more concrete than another agent for every employee: a shared one. The Foundation released Multiplayer on November 5, 2026, letting multiple people work with the same agent, workspace and conversation context. Its demonstration shows a way to preserve context between coworkers—not a cure for organizational friction.

Two laptops, one changing scene

In the launch post, Joe Pishgar described a demonstration at GitHub Universe in San Francisco. Two laptops connected to one OpenClaw agent, while a large display showed a live 3D rendering of Fort Mason. Attendees could choose a landmark to add to the scene; the Statue of Liberty was a popular choice.

As someone typed the request on the first laptop, the prompt appeared on the second before it was submitted. After the agent added the landmark, a person at the second laptop could ask it to make the object three times bigger. According to Pishgar, the agent enlarged the statue without anyone copying the earlier prompt or explaining which object they meant.

The important part was not the low-poly landmark. Both participants were changing the same thing through the same conversation. The second person’s instruction built on work already visible to the agent, rather than starting a separate exchange and trying to reconstruct how the first person got there.

Sharing instructions, not passing summaries

OpenClaw’s diagnosis is that individual agents can create another layer of separation between teams. In Pishgar’s account, coworkers develop different instructions and context, then must summarize those exchanges when they need to collaborate. Important details can disappear in that handoff, leaving another person’s agent with an incomplete version of the work.

Multiplayer addresses that handoff by keeping participants in one environment. Engineering, product, QA, operations and nontechnical colleagues can contribute to the same agent and watch one another’s instructions develop. The proposed benefit is continuity: people do not have to relay an approximation of a conversation before a colleague can take the next step.

There is also a coordination mechanism beyond seeing live text. Multiplayer routes requests through a shared queue, and participants can see when the agent is busy. Pishgar presents that visibility as a way to avoid blindly sending conflicting instructions. The queue coordinates requests; OpenClaw itself acknowledges that the feature will not cure bad management.

Free software, with your own model

The Foundation says the feature grew from its own coordination problem. It has more than a dozen full-time maintainers and a contributor community producing large volumes of pull requests—proposed code changes. It says around 50-plus people have collaborated in one shared environment. That is an account of internal use, not a measured productivity result.

Multiplayer is included in OpenClaw’s free, MIT-licensed software. Users supply their own model rather than receiving a new OpenClaw model with the feature. The Foundation lists OpenAI, Qwen, Claude and local models as options, and says OpenClaw can run on a user’s machine or a virtual private server.