OpenClaw 4.24: New AI Voice + Browser Updates
OpenClaw 4.24 introduces Google Meet integration, allowing AI agents to join meetings with full tool access, alongside upgraded voice call capabilities via an 'agent console' feature. DeepSeek V4 Flash and Pro are now built-in, with Flash set as the default model for cost-effective use. Browser automation, channel integrations (Telegram, Slack, WhatsApp), and memory systems also received significant improvements.
Summary
OpenClaw 4.24 is a major release featuring over 139 contributors — the largest contributor count on a single release according to the presenter. The update spans meeting integration, voice upgrades, new AI models, browser automation improvements, channel fixes, and memory system enhancements.
The headline feature is Google Meet integration. AI agents can now join Google Meet calls via Chrome or Twilio, bringing their full toolset and context into live meetings. Agents can export recordings, transcripts, smart notes, and attendance lists as clean files, and can scan meeting history. A recovery system handles disconnects and permission blockers gracefully, reporting issues rather than failing silently.
Voice call functionality received a major upgrade through a new 'agent console' feature. Previously, agents on phone calls ran a stripped-down version of themselves without access to full tools or knowledge. Now, mid-call, the agent can consult the full OpenClaw system to retrieve data, check files, or run tools before responding. A new Gemini Live voice provider joins OpenAI Realtime and Twilio as options, and a 'smoke test' command lets users verify their Twilio setup before going live with real customers.
DeepSeek V4 Flash and V4 Pro are now natively integrated. V4 Flash is set as the default DeepSeek model, prioritizing speed and low cost for new users. A bug where switching to a DeepSeek V4 model mid-conversation broke history replay due to missing expected data has been fixed. The presenter suggested a model selection framework: DeepSeek V4 Flash for speed and cost, GPT-5.5 for power, and Claude for personality.
Browser automation saw three notable upgrades: coordinate-based clicking (pixel-level precision for unreliable websites), an increased default timeout of 60 seconds for slow-loading pages with customizable per-profile timeouts, and improved tab recovery that automatically clears stale browser locks after crashes. A new 'browser doctor' diagnostic command identifies specific issues like missing Chrome installs or bad profile configurations.
Channel-specific fixes addressed Telegram's phantom error messages and markdown formatting bugs, Slack's out-of-order message delivery and thread leakage issues, and WhatsApp's new voice note transcription capability (converting audio messages to text before passing them to the agent) and improved media delivery from tools.
On the memory side, search results now show the breakdown between text-based and meaning-based (semantic) matches, helping users tune agent memory accuracy. The local embedding system no longer requires a heavy AI package by default, speeding up startup. The memory dreaming system was decoupled from the heartbeat system, and a bug where heartbeat prompts leaked into normal conversations — causing agents to respond with silent acknowledgements instead of actual replies — was resolved. MCP connections now properly shut down after use, idle sessions have timeouts, and the conversation compaction system was fixed to rebuild summaries from original conversation content rather than stacking summaries on top of each other.
Key Insights
- The presenter argues that the new 'agent console' fundamentally changes voice call capability — previously agents on phone calls ran a lighter, tool-limited version of themselves, but now they can pause mid-call to query the full OpenClaw agent with all tools and context before responding.
- The presenter notes that DeepSeek V4 Flash is now the default model for new DeepSeek setups, framing this as a deliberate accessibility decision so users learning the platform aren't defaulted into expensive API calls.
- WhatsApp voice note transcription is described by the presenter as 'huge for customer-facing WhatsApp agents' because people commonly send voice messages, and the agent previously had no way to understand raw audio — now it receives a text version automatically.
- The presenter identifies a compaction flaw that was silently degrading agent memory: the summary system was building summaries on top of old summaries like a 'game of telephone,' causing increasing distortion — OpenClaw 4.24 fixes this by regenerating summaries from the actual original conversation.
- The presenter highlights that OpenClaw 4.24 had over 139 contributors, calling it the largest contributor count he has seen on a single release, and uses this to frame the project as a rapidly accelerating open community effort.
Topics
Transcript
[0:00] OpenClaw 4.24 just dropped and your AI agents can now join Google Meet calls. So you can make and handle phone calls using your full agent, not some stripped down version. Deepseek V4 Flash and V4 Pro are now builtin. And the browser automation got a serious upgrade. This is a packed release. Let me show you what changed and why it matters for you. Let's start with the one that blew my mind. Google Meet is now a built-in plug-in inside OpenClaw 4.24. So your AI [0:31] agent can now join a Google Meet call. It connects with your Google account. It can join meetings through Chrome or through Twilio or for audio, right? And whilst it's in…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Julian Goldie SEO
NEW Nvidia Autonomous AI is WILD!! 🤯
Nvidia announced Nemo Clo, a new autonomous AI agent system that operates independently without continuous prompting. Powered by Nemotron 3 Ultra (a 550 billion parameter model), the system is five times faster and cheaper than previous versions, with OpenShell providing secure sandboxed execution.
Laguna XS 2.1: New FREE + Opensource Local AI!
Julian reviews Laguna XS 2.1, a new free open-source local AI coding model from Poolside that performs comparably to Qwen 3.6 and outperforms Claude Haiku on benchmarks. He demonstrates its practical capabilities by building landing pages and functional apps, highlighting its speed, offline functionality, and multiple deployment options through local setup, Claude Code, or OpenRouter's free API.
How to Run Hermes FREE Forever!
The video demonstrates how to run the Hermes AI agent for free using Gemma 4, a local open-source model from Google, with significant speed improvements through MLX optimization. The setup works on Apple Silicon Macs or via free APIs on Open Router, enabling autonomous agents to work offline and privately without subscription costs.
This NEW Chinese AI is INSANE! (FREE + Open Source!)
Long Cap 2.0 is a new open-source Chinese AI model from a food delivery app company that offers 1 million tokens of free context memory, beats GPT-4.5 on SWE bench pro benchmarks, and uses efficient parameter activation to reduce computational overhead while maintaining high performance.
Claude Code is now FREE: Here’s how…
Google's new Gemma 4 model running on Ollama is 90% faster on Apple Silicon, enabling free Claude Code usage locally without token costs. The setup requires three simple steps: downloading Ollama, Gemma 4, and installing into Claude Code, with alternatives available via OpenRouter API for non-Mac users.