OpenClaw 2026.5.18 ships Grok OAuth, Android Talk Mode, and dialog-aware browser actions

• openclaw's release post framed OpenClaw 2026.5.18 as a polish-heavy update, but the concrete changes were four agent unblockers: Grok OAuth and sidecar auth fixes, realtime Android Talk Mode, Telegram forum-topic delivery fixes, and browser actions that can finally see pending dialogs.
• According to openclaw's xAI thread, the xAI work covered SuperGrok OAuth, sidecar auth cleanup, token exchange fixes, and video defaults, all tied to the xAI provider docs.
• openclaw's Talk Mode note says Android Talk Mode now runs through realtime Gateway relay voice sessions, with streaming mic input, realtime playback, tool results, and transcripts in-app, matching the Talk node docs.
• In openclaw's Telegram thread, the team said generated media completions now stay inside forum topics, replies preserve thread IDs, and failed topic sends stop instead of leaking into the base chat, which lines up with the Telegram channel docs.
• openclaw's browser thread adds a small but useful browser change: actions can report blockedByDialog, snapshots show pending and recent dialogs, and the browser tool can answer the exact pending modal described in the browser docs.

The GitHub release is mostly plumbing work, but it hits the places where long runs usually get weird: auth flows, voice relays, thread routing, and modal popups. You can also trace the xAI side to OpenClaw's provider docs, the new voice path to the Talk docs, and a parallel Grok OAuth setup in Hermes Agent's guide.

OpenClaw put the most specific changelog detail into its xAI thread. According to that post, the release adds Grok OAuth for SuperGrok and cleans up sidecar auth, token exchange, and video defaults.

That matters mostly as dead-end removal. The release post in openclaw's main thread describes the week as fewer papercuts, and the xAI provider docs are now the canonical place for the flow.

Android Talk Mode now uses realtime Gateway relay voice sessions. In openclaw's description, that means four pieces travel through the same path: streaming mic input, realtime playback, tool results, and transcripts.

The Talk docs suggest this is more than a UI toggle. It is a full relay-backed voice session inside the app.

The Telegram changes are all routing fixes, and they read best as a list:

• Generated media completions stay inside forum topics, per openclaw's Telegram thread.
• Replies preserve thread IDs, according to the same post.
• Topic sends fail closed instead of leaking into the base chat, again per openclaw's note.

Those are narrow fixes, but they target the exact class of bug that makes multi-threaded bot deployments hard to trust. The Telegram docs are the reference point for how channels and topics are expected to behave.

Browser automation gets a visibility upgrade here. openclaw's browser note says actions can return blockedByDialog, snapshots expose pending and recent dialogs, and the browser dialog tool can answer the exact pending modal.

That is a concrete improvement over agents silently stalling on a popup. The browser docs now cover the tool surface behind that behavior.

One useful bit of context came from outside OpenClaw. In dingyi's post about Hermes Agent, Nous Research's Hermes Agent was already showing a login-based way to bind X Premium+ access to Grok usage, with a linked Hermes OAuth guide.

That does not make the OpenClaw change less notable. It does show Grok OAuth becoming a real integration surface across agent products, not just a one-off checkbox in this release.