The OpenClaw Foundation just shipped version 2.0 of its open-source AI platform, and the scale is hard to ignore: over 16,000 pull requests went into what the team calls its largest release to date. The headline features target friction points that have long made open-source AI assistants feel like projects for tinkerers rather than tools for teams.
The most significant change is onboarding. Instead of walking users through a maze of configuration screens, OpenClaw 2.0 scans the machine for resources you already have — ChatGPT or Claude subscriptions, API keys, local models — and wires them up automatically. Anything you want to tweak can be handled conversationally with the bot afterward. That’s a meaningful shift in philosophy: the software assumes you have existing AI infrastructure and adapts to it, rather than forcing you to configure everything from a blank slate.
The browser app was rebuilt from scratch, and the new design opens straight into a conversation rather than a setup wizard. A “Session Rail” gives you a compact, real-time view of what a running session is doing — ratings, plan progress, pull requests — without forcing you to dig through logs. There’s also a companion thread that lets you ask questions about an active session without interrupting the agent. The documentation notes this uses a limited snapshot of the session and a separate utility model, which suggests the team thought carefully about not polluting the main agent’s context.
Shared Cloud Sessions are the feature most likely to change how people actually use OpenClaw. Multiple users can now work on the same tasks, pulling teammates into ongoing work or assigning them tasks with full context attached. OpenClaw says its own team already uses this to build OpenClaw itself — dogfooding at a level that’s genuinely reassuring. Sessions can run locally on the gateway, on your own hardware via Paired Devices and openclaw connect, or on rented disposable machines through the Crabbox provisioning tool with AWS and Hetzner backends. Provider credentials stay on the gateway and never touch the remote machine, which is the right call for a tool aimed at teams that care about security.
Open-source AI platforms have struggled to match the polish of commercial alternatives like ChatGPT’s shared workspaces or Claude’s projects. OpenClaw 2.0 is a serious attempt to close that gap — not by copying the incumbents, but by leaning into what open source does best: flexibility, transparency, and control over where your sessions run and who can access them.
