New OpenClaw 4.25: AI Voice Features Are Wild!
OpenClaw 4.25 is a major update focused on voice capabilities, introducing six new text-to-speech providers, customizable voice personas per agent, and wake word support on Mac. The update also brings significant performance improvements including 90% faster plugin startup via a cold persisted registry, smarter browser automation with iframe awareness, and numerous bug fixes across Telegram, Slack, and WhatsApp integrations.
Summary
OpenClaw 4.25 is a significant release centered on voice functionality, performance, and reliability. The headline feature is a complete overhaul of the text-to-speech system, adding six new voice providers to the existing lineup: Azure Speech with full SSML support, Xiaomi's Mimo TTS (a local, offline option), InWorld (a personality-driven voice engine from ByteDance, the company behind TikTok), and ElevenLabs V3, their newest and most natural-sounding model. These providers give users a wide range of options depending on their use case and budget.
The most notable new concept introduced is 'voice personas.' A persona bundles together a specific provider, voice, and style settings into a single reusable package. Users can switch between personas using a slash command in chat, allowing agents to instantly change their voice and personality. This enables different voices for different platforms — for example, a warm, professional tone for a WhatsApp customer service agent and a different vibe for a Telegram channel. Voice settings can also be configured per agent and overridden per account, providing granular control. New commands like /ts latest, /ts chat on, and /ts chat off allow users to trigger read-aloud and auto-TTS without editing config files.
Voice note handling was also improved. Agents now properly send audio as voice notes (with a play button) rather than file attachments on WhatsApp, Facetime, and Telegram. Incoming voice notes from users are now framed as machine-generated transcripts in the agent's context, helping the agent respond more naturally. Mac users also gain wake word support, allowing a trigger phrase to activate a specific OpenClaw agent hands-free.
On the performance side, the plugin system was reworked to use a 'cold persisted registry,' meaning OpenClaw reads from a saved index on startup rather than scanning all plugins fresh each time. This results in dramatically faster startup times, especially beneficial for users on slower hardware like Raspberry Pi. Model listing was also optimized using static catalogs for built-in providers, eliminating unnecessary network calls.
Browser automation received several upgrades: tab URLs are cleaner and safer in agent responses, iframe awareness allows agents to see content inside embedded iframes on modern web apps, a new headless launch mode enables background browser operation without a visible window, and CDP readiness timeouts are now configurable for slow machines. The browser doctor diagnostic tool now performs live snapshot probing for more actionable troubleshooting.
Numerous platform-specific bug fixes were addressed. Telegram now auto-includes native quote excerpts in thread replies. Slack message ordering is corrected for rapid multi-message sends, and Slack thread broadcasts are now properly visible to agents. WhatsApp group voice notes that mention an agent now correctly trigger a response. A new log redaction feature allows sensitive data patterns to be masked before reaching log files. Session reset logic was fixed so background tasks no longer accidentally keep sessions alive. Installation and update processes were hardened across platforms, including a new low disk space warning and a version guard to prevent older OpenClaw versions from overwriting newer ones. The release was contributed to by 197 individuals, the largest contributor count the reviewer has seen for a single OpenClaw update.
Key Insights
- The presenter highlights that voice personas bundle a provider, voice, and style settings into one reusable package, allowing users to switch an agent's entire voice identity instantly via a slash command rather than reconfiguring settings across multiple places.
- The presenter explains that OpenClaw 4.25 introduces a 'cold persisted registry' for plugins, meaning the system reads from a saved index on startup instead of scanning all plugins fresh each time, resulting in dramatically faster startup — particularly noticeable on slower hardware like Raspberry Pi.
- The presenter notes that incoming voice notes on WhatsApp, Facetime, and Discord are now framed as 'machine-generated text' in the agent's context, so the agent understands it is reading a transcript rather than a typed message, enabling more natural responses.
- The presenter argues that what was once a custom development project requiring developers, API stitching, and weeks of work — getting an AI agent to speak naturally on WhatsApp — can now be accomplished in an afternoon using OpenClaw's new voice persona system.
- The presenter flags that OpenClaw 4.25 introduces log redaction, allowing users to define patterns for sensitive data so it gets masked before appearing in log files or the terminal — described as a meaningful privacy feature for anyone handling sensitive business data.
Topics
Transcript
[0:00] Open Claw 4.25 just dropped and this one is all about voice. Six new voice providers. Voice personas for your agents per agent voice settings. Auto read aloud for your latest messages. Wake words on Mac plus plugins start up 90% faster and the browser automation just got smarter. If you've been wanting your AI agents to sound like real people, this is the update that makes it happen. Let me show you what's inside. [0:31] And let's start with the headline text to speech. Inside open core just got a complete overhaul. Before this update, you had a few voice options. You had open eye 11 laps Microsoft Edge. You had Google. They worked, but the setup was…
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.