OpenClaw shipped version 2.0 with more than 16,000 pull requests merged, contributed by 933 people, 569 of them first-time contributors. The company describes it as the largest update in the project’s history, and by its own count the release folds in roughly half of every pull request ever merged into OpenClaw.
The project did not set out to rebuild everything at once. Work began as an attempt to simplify installation and rebuild the browser app into what the team calls a first-class experience. Getting that right meant carrying the same cleanup through memory, skills, models, automations, native apps, plugins, and security, until the scope had absorbed the rest of the product.
That expansion shows in the release cadence. This is release 107 in 230 days, and the previous 106 mostly landed a day or two apart. Getting 2.0 out took nearly seven weeks, a gap the company attributes to a growing engineering team outpacing both the codebase’s foundation and the process used to ship it, forcing a rework of both at once.
A release built from half a project’s cumulative pull requests carries real regression risk for anyone running an existing install, and OpenClaw says it held the release specifically to protect people upgrading rather than starting fresh. The blog post does not disclose a beta cohort size, a rollback rate, or any regression-testing methodology beyond the extended timeline, so the claim that upgraders were protected rests on the company’s account alone.
Setup now begins by reading whatever the machine already has. A Claude subscription, a ChatGPT one, stored keys, models running locally: the installer adopts them, and most of the remaining configuration moves out of onboarding and into conversation with the assistant. The rebuilt browser app is now the primary entry point, designed to let a user set up, monitor, or resume an in-progress task without leaving that surface.
The more structural addition is shared cloud sessions, which let a second person join an assistant’s live working context rather than starting a fresh one. OpenClaw says its own team adopted this internally while building 2.0, using it to hand off or co-work tasks without losing what a given assistant instance already knew. That is a genuine capability gap closed: prior versions had no way to bring a collaborator into an active session without discarding its accumulated context.
A rewrite this large, arriving after seven weeks of silence following a near-daily release cadence, is a familiar shape in fast-moving open source: the codebase and the process both outrun what a small team can maintain incrementally, and the fix becomes a single large cutover rather than a series of smaller ones. That pattern is a signal about how much scope an assistant platform accumulates in its first year of shipping almost daily, not just about this particular release.
OpenClaw frames the product as independent of any single model provider, noting it is open source and does not require trusting one company or one AI vendor. Teams currently running OpenClaw in production should treat this as a major-version upgrade, not a patch, and test shared cloud sessions and the new install flow in a staging environment before rolling either out to a full team.
Reported by OpenClaw in a company blog post published by writer Hannes Rudolph at openclaw.ai.
