OpenClaw v2026.9.3 🦞 Cleaner update recovery, faster reconnects, and live browser viewing

This release contains 1,844 pull requests, 40 direct commits, and contributions from 190 people. Read the complete OpenClaw v2026.9.3 release notes. For self-hosted installs, check the updated Node requirements. This release requires Node 24.16.0 or newer on 24.x, or Node 26.1.0 or newer; Node 26 is recommended. OpenClaw v2026.9.3 is out, with cleaner update recovery and faster session reconnects to help you get back to work after an interruption. Supported updates are checked before activation and verified afterward, while the web UI keeps loaded panels and saved drafts through reconnects. You can also watch your agent's browser activity live, share conversations through revocable read-only links, search saved meeting transcripts, and start repository work directly on a cloud runner or paired computer. Learned skills stay with their agent across workspaces, and the Mac app gains native browser tabs that remain open when you switch conversations. Update recovery and faster reconnects Supported core and plugin updates are rehearsed in an isolated copy before the running installation changes. After activation, OpenClaw checks the running version, service health, plugins, and channel readiness. A failed check follows the supported repair or rollback path, and restoring the previous working installation is reported as recovery rather than a successful update. In the web UI, loaded panels remain available through connection interruptions and refresh when the connection returns. Suspended tabs can recover missed update notices without losing their current page or stored draft. Eligible interrupted work, follow-ups, and attachments can also resume after an update, with completed child results held until their parent conversation recovers. Watch browser automation live The Browser panel can now show the agent's page loading, scrolling, and responding as it happens, instead of making you wait for the next screenshot. You can follow along with the existing click, type, scroll, Annotate, and Inspect controls still available. Streaming depends on the browser connection. Paired-computer browsers, Chrome MCP existing-session profiles, and other routes without streaming continue to use screenshots. Share a conversation with a revocable link Conversation creators and Gateway admins can publish a read-only link to a conversation's existing and future text. The shared page leaves out tool activity, reasoning, files, images, and interactive widgets, so it is a way to share the written conversation rather than access to the session itself. Anyone with the public link can read it until you revoke access. Review the conversation before sharing, and remember that revoking the link cannot recall copies someone has already saved. Search saved meeting transcripts The meeting library brings saved notes and full transcript search together, so you can look up what was said without reading through the whole meeting. Capture sources are managed from Communications settings, and exporting a selected meeting includes the complete transcript even when you have a search filter active. Start repository work in the cloud Repository sessions can start directly on a managed cloud runner or paired computer without first cloning the repository on the computer running OpenClaw. The remote runner handles checkout and setup, and checkpoints retain accepted changes when you stop the session. Moving execution back to the Gateway is an explicit choice when you want to continue locally. This needs the current worker bundle on paired hosts or newly provisioned cloud nodes. Checkpoints preserve accepted work, not a full Git history or a backup of unaccepted changes. Skills, Mac tabs, and model accounts A skill your agent learns now belongs to that agent's Skill Workshop, so changing workspaces no longer means moving the skill collection along with it. Installed skills and suggestions awaiting review have separate views. Links opened in the Mac app now appear in native Browser-panel tabs that stay with the window when you switch conversations. These are tabs you browse yourself, separate from the Agent browser tabs your agent controls. Models settings also makes individual provider accounts easier to manage. Admins can inspect each account's health, change supported account priority for an agent, and sign out of one account without disconnecting the others.