OpenClaw Update 4.24 Overview

OpenClaw 4.24 ships Google Meet integration, DeepSeek V4, and a stack of browser/session hardening fixes.

Google Meet is the headline

OpenClaw 4.24 lets agents join Google Meet calls with personal Google auth, Chrome sessions, Twilio sessions, and paired-node Chrome setups. The agent can listen, take notes, and trigger follow-up workflows after the meeting, and it supports artifact exports plus attendance tracking. If a Meet tab is already open, the recovery tooling picks it up. The creator frames this as a signal that any team running an OpenClaw agent in calls is "pretty serious about AI workflow."

DeepSeek V4 and Gemini live voice

DeepSeek V4 Flash and V4 Pro are bundled, with V4 Flash set as the onboarding default. The release fixes DeepSeek thinking and replay behavior on follow-up tool-call turns so the model stays coherent across multi-step research, tool calling, coding help, and browser tasks. Gemini live voice bridges are also in, bringing faster model catalogs and a lighter plugin startup path.

Browser, session, and Telegram stability

Playwright now ignores benign route races during guarded navigation, and ARIA snapshot references bind to live DOM nodes so follow-up browser actions don't time out when the page re-renders. Browser tabs auto-close on idle, daily resets, /new, or /reset, stopping tab leaks. Telegram now blocks duplicate in-process long pollers for the same bot token and prevents polling restarts from self-conflicting. Tool result pruning is hardened against malformed text blocks from void or undefined handlers — one bad plugin can't corrupt the whole context window.

Compaction, usage, and provider fixes

Manual /compact now respects your keep-recent token settings, re-distills safeguard summaries instead of snowballing older ones, and enables safeguard quality checks by default. The creator recommends /compact <what to keep> over bare /compact so the agent doesn't guess what matters. /usage pulls from a fresh context snapshot and includes cache write tokens in the cache hit denominator. For OpenAI Codex, system prompts route through top-level instructions while keeping native payload controls intact, and GitHub Copilot preserves encrypted reasoning item IDs across replay.

Plugin compatibility and Windows fixes

A new compatibility registry tracks SDK versions, config schemas, setup requirements, runtime compatibility, and deprecation records with dated migration metadata. Package installs now fall back to copying shared chunks when bundle plugin errors hit, which fixes Windows npm updates that previously failed to load copied disk modules.