OpenClaw 2.0 was contributed by 933 developers, including 569 first‑timers. The update rebuilds the browser app, introduces shared multiplayer cloud sessions, moves session storage, and adds enterprise‑focused features. The release arrives amid growing competition from Nous Research’s Hermes Agent, which has frustrated users with OpenClaw’s setup challenges.
OpenClaw shipped its biggest update ever over the weekend, branding it OpenClaw 2.0 and describing the effort as “accidental” due to scope creep.
The release notes cover installation, messaging, memory, skills, models, automations, the browser, native apps, plugins, and security—essentially everything a user should know.
Myriad: When will OpenAI release GPT-6? Click to make your prediction.
A total of 933 contributors, 569 of them first‑timers, submitted more than 16,000 pull requests for this single release—roughly half of all PRs the project has merged to date.
What OpenClaw actually is, for the uninitiated
OpenClaw is a self‑hosted, open‑source framework that turns an LLM into a persistent agent. It runs on your machine, wakes on a schedule, and communicates via WhatsApp, Telegram, Discord, or Signal. With permission it can access email, calendar, files, and a shell.
The project was created by Austrian developer Peter Steinberger after he left PSPDFKit. It quickly amassed 147,000 GitHub stars and spawned an ecosystem of spin‑offs, including the AI‑only social network Moltbook.
Steinberger joined OpenAI in February to lead its personal‑agent initiatives, but OpenClaw became an independent nonprofit—OpenClaw Foundation— partnered with OpenAI, Microsoft, GitHub, Nvidia, Atlassian, and Tencent.
Why “accidental” is the right word
The original plan was modest: simplify installation and rebuild the browser app. However, the team expanded the scope to include memory, models, plugins, messaging, and security. After 106 releases in 230 days—roughly a daily cadence—seven weeks of silence was unusual, prompting the “accidental” label. The extra time ensured existing installations would not break, and users were warned to back up sessions before the SQLite migration.
Community manager Hannes Rudolph described the effort as “eating our own dog food.” The team adopted the new multiplayer sessions internally, which exposed gaps in the previous single‑user design, leading to the surge of contributions.
What’s actually new
Setup now checks for existing Claude/ChatGPT subscriptions, API keys, or local Ollama models before prompting for new configuration. It validates the model and jumps straight into a conversation rather than displaying an unwanted settings screen.
The Control UI was rebuilt around chat, aligning it with ChatGPT, Claude, and Gemini. Files, Git diffs, pull‑request status, a browser panel, and a terminal are embedded around the conversation, eliminating the need to switch between multiple tools.
The team reports a simulated latency test that reduced initial JavaScript requests from 140 to 45 and cut startup time from 1.6 seconds to 575 milliseconds.
The multiplayer pitch
The flagship feature is shared cloud sessions, allowing team members to join an agent’s work mid‑task without losing context. Steinberger said his own team replaced local coding harnesses with a shared environment at team.openclaw.ai, calling local harnesses “relics of the past.”
Security enhancements include request‑specific approvals, command permissions scoped to arguments, Docker and Podman sandboxing, role‑based execution, and a team‑scoped Secret Store that injects credentials without exposing them to the model.
However, sandboxing and approval features are disabled by default. OpenClaw’s documentation clarifies that a single Gateway defines a trust domain; true tenant isolation requires deploying separate Gateways (called “cells”).
The competition is heating up
OpenClaw faces competition from Nous Research’s Hermes Agent, launched in March. Hermes differentiates itself with automatic procedural memory that turns successful workflows into reusable skills and with security defaults that assume multiple users may access the system.
OpenClaw separates a Gateway daemon from the agents it routes, while Hermes integrates conversation, tool dispatch, and memory into a single agent class—making it simpler to reason about but harder to scale.
The consensus this year is that OpenClaw leads in ecosystem breadth and channel support, whereas Hermes excels in self‑improvement and safer default security. OpenClaw 2.0’s new sandboxing narrows the gap but does not eliminate it.
How to get it
Existing users can update via the normal CLI. Note that sessions created after the SQLite migration will not appear if you downgrade without a backup.
New users can follow OpenClaw’s setup guide, which now offers guided onboarding for Mac, Windows, Linux, iPhone, iPad, and Android, plus staged imports from Claude, Codex, or Hermes.
The project remains free, MIT‑licensed, and hosted at github.com/openclaw/openclaw.
