OpenClaw 2.0 arrives with 16,000+ pull requests from 933 contributors

OpenClaw has shipped version 2.0, an update the project describes as the largest in its history. According to the announcement on the OpenClaw blog, which reached the Hacker News front page on August 30, the release was assembled from more than 16,000 pull requests contributed by 933 people, 569 of whom were first-time contributors.

What the release covers

Nearly every layer of the project changed: installation, messaging, memory, skills, model handling, automations, the browser and native apps, plugins and security, plus a long list of smaller fixes. The announcement puts that scale in context — this single release represents roughly half of all pull requests ever merged into OpenClaw.

The “accidentally” in the title describes how it happened. The team set out to simplify installation and rebuild the browser app, but doing that job properly, they write, meant carrying the cleanup through the rest of the codebase until the work grew into a full 2.0 release.

Why it took almost two months

Before this update the project shipped constantly: 106 releases over 230 days, most only a day or two apart. Going nearly seven weeks without shipping was therefore a break from habit. The team attributes the pause to growth, saying the increasing volume and pace of work outgrew both the foundations of the codebase and the process used to ship it, so both were reworked at the same time.

They also say the extra weeks went into verifying two distinct paths — fresh installs and upgrades of existing setups — rather than shipping quickly and handing users something that broke what they already had.

Onboarding starts with what you already have

For first-time installs, OpenClaw now begins from whatever already exists on a user's machine: an existing ChatGPT or Claude subscription, stored API keys, or local models. Much of the configuration was cut or simplified, and the remainder moved out of initial setup, so a new user can reach a first conversation faster and finish configuring the assistant by talking to it.

The browser app, where most people first try OpenClaw, was rebuilt as a complete front end rather than an afterthought. It opens directly into a conversation and supports finishing setup, returning to ongoing work, or following tasks live as they run.

From one workflow to multiplayer

The announcement sketches a growth path for users. Start with one small automation, such as watching an inbox for a child's school emails and forwarding the important ones to Telegram. From there, tasks can reach across more services without becoming harder to use — for instance, answering a relative's question by finding an old receipt and sending the reply, without the user doing the searching.

The team says it hit the same progression internally while building this release, wanting to share tasks, collaborate on them, or hand them off entirely. Until now there was no way to bring another person into the work without losing what the assistant already knew. Shared cloud sessions close that gap: a colleague can join live work or take a task over with its context intact. The team now uses this mode to build OpenClaw itself, and the announcement shows a user-built dashboard inside one of these shared workspaces as an example of what it enables.

Ownership rather than a vendor pitch

The project stresses that it is not selling anything or asking users to trust a single company, model, or AI provider. OpenClaw is open source, and the announcement frames it as belonging to the people who use and build it — closing with an invitation for users to criticise the team bluntly when it gets things wrong.

Why it matters

A release equal to half of a project's total merged history is a significant event for anyone running OpenClaw, and the fresh-install-versus-upgrade question clearly deserved the extra testing time the team describes. The contributor figures — 933 people, more than half of them new — suggest open-source agent frameworks are attracting sustained community effort, not just corporate sponsorship. The shared-session feature also signals where agent software appears to be heading: away from single-user assistants toward collaborative, multiplayer tools. And onboarding that reuses existing subscriptions and local models keeps the project positioned as a vendor-neutral alternative to closed, single-provider assistants.