Summary
- OpenClaw has released version 2026.8.1, branded as OpenClaw 2.0, with contributions from 933 developers, including 569 new contributors.
- The latest update revamps the browser application, introduces shared multiplayer cloud sessions, and enhances session storage, catering to enterprise users.
- This release comes as Nous Research's Hermes Agent continues to attract users who are dissatisfied with OpenClaw's previous setup challenges.
This past weekend, OpenClaw unveiled its most significant update to date, dubbed OpenClaw 2.0. The team described the update as "accidental," a term that will be explained further.
The release notes cover various aspects including installation, messaging, memory, skills, models, automations, the browser, native applications, plugins, and security—essentially everything users need to know.
Myriad: Predict when OpenAI will release GPT-6. Share your prediction.announcement post, the project was supported by 933 contributors, with 569 of them making their first contribution, totaling over 16,000 pull requests. This represents approximately half of all pull requests that have ever been merged in OpenClaw's history, all released in a single update.Understanding OpenClaw
For those unfamiliar, OpenClaw is an open-source, self-hosted framework designed to transform a large language model (LLM) into an agent that remains active even after closing the browser tab. It operates on your device, activates on its own schedule, and communicates via WhatsApp, Telegram, Discord, or Signal, rather than being confined to a chat window. With the right permissions, it can interact with your email, calendar, files, and shell.
Peter Steinberger, an Austrian developer, created OpenClaw after stepping back from his company PSPDFKit. The project quickly gained immense popularity, earning 147,000 stars on GitHub in just a few weeks, and spawning a range of spin-offs, including a unique AI-only social network called Moltbook.
In February, Steinberger joined OpenAI to lead its personal-agent initiative. However, OpenClaw itself did not become part of OpenAI; instead, it transitioned into its own nonprofit entity, the OpenClaw Foundation, partnering with organizations such as Microsoft, GitHub, Nvidia, Atlassian, and Tencent. It remains independent, or so it claims.
The "Accidental" Update
Initially, this OpenClaw update aimed to simplify installation and rebuild the browser application, but as the team delved deeper, they expanded the scope to include improvements in memory, models, plugins, messaging, and security. Prior to this release, OpenClaw had issued 106 updates over 230 days—essentially on a daily basis. Therefore, going seven weeks without a release was uncharacteristic, as they later acknowledged.
The team dedicated this additional time to ensure that existing installations would not break during the upgrade, which is a common concern for users familiar with OpenClaw. Consequently, the release notes included a warning to back up sessions before transitioning to the new SQLite storage system.
The term "accidental" reflects the team's candid admission of scope creep. What began as two minor updates—onboarding and browser enhancements—expanded to encompass a wide range of improvements due to interdependencies in the code being modified.
Hannes Rudolph, OpenClaw’s Community Manager, explains that the team began utilizing their own workflow through multiplayer sessions, which revealed significant limitations in the previous single-user design. This realization prompted them to address those issues, resulting in 16,000 pull requests rather than just two.
New Features in OpenClaw 2.0
The new setup process now verifies what users already have—such as an existing Claude or ChatGPT subscription, an API key, or a local Ollama model—before prompting for any new configurations. It checks that the model functions correctly before saving it, leading users directly into a live conversation instead of a settings screen that many found frustrating.
The Control UI has been redesigned to prioritize chat, moving away from a separate overview dashboard, aligning it visually with ChatGPT, Claude, and Gemini. Now, files, Git diffs, pull-request statuses, a browser panel, and a terminal are all integrated around the conversation, eliminating the need to switch between multiple tools.
The team reports that a simulated latency test reduced initial JavaScript request times from 140 to 45 milliseconds and startup time from 1.6 seconds to 575 milliseconds. While faster performance is always welcome, users should interpret these numbers cautiously.
Multiplayer Functionality
The standout feature of this update is the introduction of shared cloud sessions, allowing colleagues to join an agent's task mid-process without losing context. Steinberger mentioned on X that his team has transitioned from individual local coding environments to a shared setup at team.openclaw.ai, labeling local setups as "relics of the past."
Security enhancements have also been implemented. OpenClaw 2.0 includes request-specific approvals, command permissions limited to specific arguments, and advanced features such as Docker and Podman sandboxing, role-enforced execution, and a team-scoped Secret Store that can insert credentials into requests without exposing them to the model.
However, it's important to note that sandboxing and approvals are not activated by default. OpenClaw's documentation states that one Gateway corresponds to one trust domain, which does not guarantee isolation between untrusting parties. Organizations requiring true tenant isolation will need to set up separate Gateways, referred to as "cells," rather than simply enabling a setting.
Competitive Landscape
OpenClaw is not the only player in this space. Nous Research's Hermes Agent, which emerged in March, positions itself as an alternative to OpenClaw, emphasizing procedural memory that transforms successful workflows into reusable skills automatically and security defaults that do not presume single-user access.
OpenClaw separates the Gateway daemon from the agents it directs, while Hermes combines conversation loops, tool dispatch, and memory into a single agent class, which complicates scaling but simplifies reasoning. The prevailing sentiment this year seems to suggest that OpenClaw excels in ecosystem and channel diversity, while Hermes is preferred for its self-improvement capabilities and cautious default settings.
The new sandboxing feature in OpenClaw 2.0 narrows the competitive gap, but it does not eliminate it, as the enhanced security features still require manual activation.
How to Access OpenClaw 2.0
Current users can update through the standard command-line interface. It is advisable to read the fine print first, as sessions created after migrating to SQLite will not be visible if a downgrade occurs without a backup.
New users can refer to OpenClaw's setup guide, which now offers guided onboarding for Mac, Windows, Linux, iPhone, iPad, and Android, along with staged imports for those transitioning from Claude, Codex, or even Hermes.
OpenClaw remains free, maintains an MIT license, and can be found at github.com/openclaw/openclaw.