OpenClaw has launched version 2.0, officially released as v2026.8.1, in what the project describes as its largest update so far. The release brings major changes to installation, browser interaction, memory, automation, collaboration, plugins and security.
The update arrives with a playful reference to Grand Theft Auto 6, with OpenClaw highlighting that its major release had arrived before the highly anticipated game. The comparison is promotional rather than a technical connection between the two products.
At the centre of OpenClaw 2.0 is an attempt to reduce the complexity of setting up an AI agent. The release introduces guided onboarding through its Custodian system, using structured options and native setup controls to help users configure the platform. Local model installation has also been improved, with the software able to detect available local models and provide model download options during onboarding.
The simplified setup is particularly relevant for users who may have previously found agent platforms difficult to configure. OpenClaw supports different model providers and local inference options, so reducing the number of manual steps can make the initial experience more accessible.
Another major area of the update is the browser interface. OpenClaw's browser system provides an isolated browser profile that an AI agent can control. The agent can open pages, read content, click elements, type information and work with browser sessions separately from the user's personal browser environment.
The broader 2.0 update expands the browser and web experience as part of its effort to make agents more practical for everyday tasks. Users can work through web and native interfaces, while the platform continues to separate agent controlled browser activity from personal browser sessions.
Shared sessions are another major addition. OpenClaw 2.0 allows authorised participants to share access to specific sessions and collaborate on ongoing agent work while retaining session context. The release includes controls for session visibility, membership, ownership and permissions.
The collaboration model is designed around a shared Gateway trust boundary. In practice, that means the feature is intended for trusted teams, households or groups rather than acting as a full multitenant security system for unrelated customers.
OpenClaw has also added features that allow sessions to operate beyond the main Gateway. The release supports running work on paired devices or cloud workers, moving session workspaces and reusing prepared machines and project resources for later sessions.
This approach could make OpenClaw more useful for people who want an agent to continue working even when the original device is unavailable. The session can retain its context while execution is moved to another supported environment.
Persistent progress tracking is another part of the update. OpenClaw can keep a durable session progress card across reloads and allow users to follow subagent activity and accumulating changes through web and native interfaces.
The platform has also expanded its interactive interface. Structured agent questions can now be answered through web cards, native cards, messaging buttons or plain text. This gives users several ways to respond when an agent requires clarification or approval before continuing a task.
Interactive widgets have received additional support as well. Users can place interactive results into session dashboards, grant specific actions or network permissions and export rendered views as images. The change moves OpenClaw beyond simple text based conversations towards richer agent workspaces.
Memory is another area that has been substantially revised. With Active Memory enabled, the platform can retrieve bounded private conversation context for the same agent in supported personal installations. The release also introduces background memory consolidation designed to promote useful, provenance qualified information into longer term memory.
OpenClaw 2.0 also changes how the system learns from completed work. It can capture reusable lessons and automatically apply scanner approved skills or Workshop owned skills, while user authored changes remain subject to explicit controls.
The session reset behaviour has also changed. When no specific reset policy is configured, conversations can now continue across idle periods and day boundaries instead of being automatically reset under a default schedule. Users can still use explicit reset commands or configure daily or idle reset policies.
Automation has been expanded as well. New automations can be tied to the conversation from which they were created, and owner controlled loops can run according to fixed intervals or self paced checks. The system also supports conversation specific heartbeat alerts.
For teams, OpenClaw 2.0 introduces verified operator roles that can limit which agents, sessions and operational scopes different users can access. Optional sandboxing can also be required for newly created sessions. These controls are intended to improve collaboration governance while remaining within OpenClaw's existing trust model.
Security has received attention in the release as well. External plugin installation now includes clearer trust information, including capabilities, source and version details. The system can also require explicit confirmation for arbitrary executable plugin sources.
Credential handling has similarly been improved. Agents can request credentials through masked prompts without placing the secret directly into the conversation or model context. OpenClaw has also added support for shared credential management in SQLite, with protected egress controls and separate handling for readable environment values and secrets.
The release also contains several important breaking changes. OpenClaw has removed its bundled OpenProse plugin and the /prose command, while a migration process is provided through openclaw doctor fix. Model references using older OpenAI route names also require migration to the updated configuration structure.
Additional provider support has been expanded. OpenClaw's onboarding system can install official provider packages for services including BytePlus, Mistral, ComfyUI, OpenCode, Synthetic, Volcengine, Vydra and Xiaomi when required. Other providers and services have also been split into separate official plugins.
The platform is also becoming more modular in its model support. OpenClaw 2.0 adds model discovery for additional providers and models, while allowing administrators to use explicit model policies and per agent restrictions.
One notable feature is better conversation management. Users can search previous conversations by exact words or phrases and reopen the surrounding messages. Session organisation has also been improved through project based grouping, searchable sessions and persistent session colours across supported interfaces.
OpenClaw has additionally introduced conversation branching. Users can rewind or fork from a previous message and switch between retained branches. The feature changes the conversation context without automatically undoing files or other tool effects that were already produced.
Terminal and file handling have also received upgrades. Shared conversation terminals can be viewed by authorised browser participants, and agents can inspect or operate an operator opened terminal within the relevant session and its permissions. Files can also be dropped into terminals with upload progress, cancellation and retry support.
The update includes experimental features as well. Experimental Fleet allows separate OpenClaw cells to be provisioned on local Docker or Podman systems, while Experimental Swarm can distribute bounded tasks among parallel subagents and track their progress.
The scale of the release is also notable. OpenClaw's documentation describes more than 16,000 pull requests associated with the release, with contributions from hundreds of developers. This indicates that the 2.0 release is not a minor feature update but a broad platform overhaul.
For users upgrading from an earlier version, the project recommends backing up configuration and state before installation. The documentation also provides repair and migration commands for configuration changes introduced in the new release.
The update reflects a broader shift in the AI agent market. Instead of focusing only on conversation and question answering, agent platforms are increasingly attempting to interact with browsers, files, terminals, communication channels and external services.
OpenClaw's architecture follows this model. Its Gateway acts as the central control layer, while agents use models and tools to perform tasks. The new release expands the range of environments in which those tasks can run and the number of people who can participate in the same work.
For individual users, the main benefit of version 2.0 may be easier onboarding and stronger continuity between sessions. For trusted teams, shared sessions and operator roles could make collaboration more practical. For developers, the expanded provider support, browser control and plugin architecture offer additional ways to customise the platform.
However, the release also increases complexity in other areas. Features such as shared credentials, cloud workers, plugins and agent permissions require careful configuration because they give AI agents access to more tools and environments.
The project's own documentation repeatedly emphasises permission controls, plugin provenance and session boundaries. Those measures are important because an AI agent capable of browsing websites, operating terminals or interacting with external services can potentially perform actions with real consequences.
OpenClaw 2.0 therefore represents both a usability upgrade and a significant expansion of capability. The platform is moving towards a model where an AI agent can maintain context, interact with software environments, collaborate with authorised users and continue tasks across different machines.
The timing of the release has also helped generate attention because of the project's reference to GTA 6. The comparison has become part of OpenClaw's launch messaging, but the more important development is the scope of the software update itself.
For people who have been following OpenClaw, version 2.0 provides a much broader feature set than earlier releases. The new browser, onboarding, memory, collaboration, automation and security capabilities collectively change how the platform can be used.
For newcomers, the guided setup should make the initial configuration easier. For experienced users, the larger update may require migration work, particularly where older plugins, OpenAI routes or OpenProse functionality are involved.
The release also demonstrates the rapidly changing nature of AI agent software. New capabilities are being added not only to AI models but also to the surrounding systems that allow those models to interact with computers, applications and users.
OpenClaw 2.0 is therefore more than a routine version upgrade. It represents a substantial rebuild of the platform's user experience and infrastructure, with a clear focus on easier setup, persistent agent work, browser interaction and controlled collaboration.
As users begin adopting the new release, attention will likely focus on how well these features perform in real world workflows, how stable the migration process is and how effectively OpenClaw balances powerful automation with security and user control.
For now, the official v2026.8.1 release confirms the arrival of OpenClaw 2.0, while the project's own documentation provides the detailed picture of what has changed. The result is a significantly broader AI agent platform aimed at making complex computer based tasks easier to configure, continue and share.