Table of Contents
OpenClaw has launched OpenClaw 2.0, the largest update yet to the open-source AI agent platform, following nearly seven weeks without a major release.
The update was built by 933 contributors, including 569 first-time contributors, and incorporates more than 16,000 pull requests. The final contribution record lists 16,962 unique pull requests in the release, which touches nearly every part of the platform, including installation, messaging, memory, skills, models, automations, browser and native apps, plugins and security.
But the project did not originally set out to build what it would ultimately call OpenClaw 2.0.
The OpenClaw 2.0 Origin Story
According to the team, development on OpenClaw 2.0 started with an effort to simplify installation and rebuild OpenClaw’s browser application as a first-class interface. The changes eventually spread throughout the rest of the system, turning what began as an onboarding and interface overhaul into a much broader release.
That process also broke with OpenClaw’s historically rapid release cycle.
Before OpenClaw 2.0, the project had shipped 106 releases in 230 days, with many arriving only one or two days apart. The team said its growing contributor base and development volume eventually outgrew both the project’s existing technical foundation and its release process, prompting it to rework both. Roughly half of all pull requests ever merged into OpenClaw were included in the new release.
One of the most visible changes is a simplified first-time setup process.
OpenClaw can now detect resources already available on a user’s computer, including existing ChatGPT or Claude subscriptions, API keys and local models. The team also moved or removed parts of the initial configuration process so users can reach their first agent conversation faster and continue configuring OpenClaw by talking directly to their Claw.
The browser application has also been rebuilt around that conversation-first model, opening directly into a conversation while allowing users to configure their agent, return to existing work and follow ongoing tasks from the same interface.
OpenClaw 2.0 Under The Surface
Many of the more significant changes in this release sit beneath the redesigned interface.
Conversation Search
OpenClaw 2.0 introduces conversation search, allowing users to search visible conversation history for exact words or phrases and reopen surrounding messages. Sessions can also now extend beyond the machine running a user’s OpenClaw Gateway, with work able to run on paired devices or cloud workers while carrying the session workspace with it.
Users can follow longer-running agent work through persistent progress cards that survive page reloads and expose activity from subagents and accumulating edits across web and native applications. OpenClaw has also added interactive widgets that can appear in conversations, be pinned to session dashboards and receive narrowly defined permissions to perform actions or access specific network origins.
Security & Automation
Security and automation received upgrades as well. Agents can request credentials through masked prompts without placing the credential itself into the visible conversation or model context.
Users can also approve an exact recurring automation operation once, inspect or revoke that permission later, while changes to the operation require fresh approval.
Active Memory
OpenClaw is also expanding how its agents remember and learn.
With Active Memory enabled, personal installations can retrieve bounded context from previous private conversations with the same agent. A new background memory-consolidation system can promote qualified information into long-term memory, while an automatic self-learning system can capture reusable lessons and apply approved or Workshop-owned skills. Users retain controls to disable or restrict both behaviors.
Collaboration
Another major part of the release is collaboration.
Shared cloud sessions allow OpenClaw users to bring another person into ongoing work or hand a task over without discarding the agent’s existing context. The release also adds team operator roles that can restrict which agents, sessions and operator capabilities individual users can access, alongside a shared credential store for team-scoped secrets and environment variables.
OpenClaw said the shared-session capability emerged from its own development process as team members increasingly used their Claws to build OpenClaw itself and needed a way to collaborate on agent-driven work without starting over each time.
Additional Changes
The release also introduces several more experimental agent-orchestration features. An experimental Fleet system can provision and manage separate OpenClaw environments across a local Docker or Podman host, while an experimental Swarm feature can fan work out across multiple parallel subagents and collect structured results while users follow their progress through the OpenClaw interfaces.
There are breaking changes as well. OpenClaw removed its bundled OpenProse plugin and /prose command and is migrating previously shipped codex/* and openai-codex/* model references to openai/*. The project recommends using its openclaw doctor --fix command to handle supported migrations, while external plugin developers face additional SDK migration requirements ahead of deprecations scheduled for September 1.
OpenClaw’s Bigger Bet on Personal Software
For OpenClaw, the significance of 2.0 goes beyond any individual feature. The broader vision is software that starts with a simple task, then gradually becomes more capable as users connect more of their work, data and communications to it.
"Your OpenClaw starts with one useful workflow and grows as far as you want it to, with the possibility of reaching across more of your life and work or becoming multiplayer with your family or team when you want and how you want. This has changed the relationship people have with software, turning it from something designed somewhere else that you have to accept into something you can tell what to do, shape around your life and work, and actually own."
OpenClaw 2.0 is the project’s attempt to turn that idea into a more complete product. Features such as persistent memory, cross-device sessions, recurring automations, and multiplayer workspaces give a Claw the ability to accumulate context and responsibilities over time rather than behaving like a disposable chatbot session.
That also explains the project’s emphasis on remaining open source and model-agnostic.
OpenClaw is betting that increasingly capable AI agents will become a persistent layer across people’s digital lives, and that users should be able to shape and own that layer rather than depend entirely on software designed and controlled by a single provider.
