OpenClaw 2.0: easier setup and shared cloud sessions
The OpenClaw Foundation folds more than 16,000 pull requests into version 2.0: setup detects existing plans, and teams share cloud sessions.
Illustration: an AI-generated scene showing empty workstations in a half-circle, their screens linked by thin threads of light.
OpenClaw 2.0 is the project's largest release so far: it detects existing subscriptions, API keys and local models during setup on its own, ships a rebuilt browser app, and lets several people work inside the same cloud session.
At a glance
- Version 2.0 bundles more than 16,000 pull requests, per The Decoder — the platform's largest release to date.
- Setup detects existing ChatGPT or Claude subscriptions, API keys and locally installed models by itself.
- A Session Rail displays live scoring, plan progress and the open pull requests of a run.
- Sessions run on the local gateway, on paired hardware, or on rented machines via Crabbox (AWS, Hetzner).
- The OpenClaw Foundation maintains the project; the code is publicly available on GitHub.
OpenClaw 2.0 is the largest release the open-source platform has shipped: most of the configuration work is gone from the first run, the browser app has been rebuilt, and several people can now share one running cloud session. The project puts the scope of the update at more than 16,000 pull requests.
Getting started without the configuration step
On first launch the software looks for what the machine already has: an existing ChatGPT or Claude subscription, stored API keys, locally installed models. Whatever it finds, it wires up — you no longer enter those details yourself. That removes a stack of steps that used to sit at the very front of the process.
The browser app becomes the front door
The rebuilt browser application is no longer a side channel but the normal way in. It opens straight into a conversation, and from there you can finish setting things up or follow a task that is already running.
Two elements are new. The Session Rail is a narrow status strip carrying live scoring, plan progress and the open pull requests belonging to a run. Next to it sits a companion thread for asking questions mid-run without stopping the agent that is working. According to The Decoder, that thread only sees a limited snapshot of the session and is served by a separate utility model.
Several people, one session
Shared cloud sessions are the third addition. Team members can work the same task together and hand partly finished work to the next person along with its context, instead of restating where things stand.
Where a session actually executes stays a choice. The local gateway is the default. Beyond that, paired devices can be attached with the openclaw connect command, or work can move to rented machines supplied by the Crabbox provisioning tool — the reporting names AWS and Hetzner as supported backends.
What the reporting does not cover
The project is run by the OpenClaw Foundation, and the source code is public on GitHub. The sources reviewed here say nothing about the license model, system requirements, what cloud sessions cost, or a timeline for further versions — none of that is verified in this article. Nor is there independent measurement yet of how stable the release proves in daily use.
FAQ
What's new in OpenClaw 2.0?
Three things: a setup that detects existing subscriptions, API keys and local models on its own; a rebuilt browser app with a Session Rail and a companion thread; and shared cloud sessions for multiple people. The release covers more than 16,000 pull requests.
Who is behind OpenClaw?
The OpenClaw Foundation develops the platform as an open-source project, with the code available on GitHub. The reporting reviewed here does not state the license model.
Where do OpenClaw sessions run?
On the local gateway by default. Alternatively on paired devices attached with the openclaw connect command, or on rented machines provisioned through the Crabbox tool — AWS and Hetzner are named as backends.
