ElevenLabs Launches 11ai: A Voice Assistant That Actually Does Stuff

ElevenLabs launches 11ai voice assistant

ElevenLabs has just rolled out 11ai, an experimental voice-first AI assistant that doesn’t just answer—you can use it to execute real tasks by voice, thanks to deep integration with work tools via the Model Context Protocol (MCP).

Key Points:

1. Voice-Powered Task Execution - Instead of merely responding, 11ai connects to platforms like Perplexity, Linear, Slack, and Notion (and supports custom MCP servers) to carry out workflows—such as scheduling meetings, creating tickets, summarizing messages, or doing customer research—all by voice command.

2. Massive Voice & Multimodal Support - The alpha supports over 5,000 voices, language detection, low-latency voice/text interactions, and Retrieval-Augmented Generation—all powered by ElevenLabs’ own conversational AI infrastructure,

3. Free Alpha With Developer Flexibility - Available now in a free alpha for a few weeks, ElevenLabs invites feedback on integrations and capabilities. Developers can connect custom MCP servers, enabling private or internal tool support.

Conclusion

ElevenLabs is aiming to elevate voice assistants from helpers to doers: 11ai not only understands your commands—it acts on them. As the first hands-free agent built on robust speech models and full tool integration, it could make Siri feel outdated—and show us where voice-based productivity really begins.

Microsoft’s Mu Language Models

Microsoft just launched Mu, a compact language model powering on-device agents for Windows Copilot+ PCs, especially within the Settings app. It runs entirely on your PC’s NPU, mapping natural language queries directly to system actions—no cloud needed.

Key Points:

1. Tiny Yet Powerful - At just 330 million parameters, Mu is an encoder–decoder model designed specifically for efficient, real-time use on NPUs. It processes input in one pass, enabling fast, low-latency performance—delivering 100–200 tokens/sec, with first-token delays under 500 ms.

2. Privacy & Offline Execution - Since Mu runs entirely locally, all computations stay on your device. This improves speed and privacy—ideal for sensitive settings configuration without relying on cloud services.

3. Enhancing Settings Experience - Loaded as part of the Settings AI agent, Mu interprets commands like “make my mouse pointer bigger” and applies them directly, offering an intuitive, conversational interface for system control.

Conclusion

Mu demonstrates how compact models can deliver powerful, private, and responsive assistance. As more Copilot+ PCs roll out, expect more tasks to be handled seamlessly—without sending your data to the cloud.

🚀 Other AI updates to look out

Disney in Talks to License IP with OpenAI

Disney is reportedly considering licensing its characters and IP to OpenAI, announcing plans to collaborate with “companies like OpenAI” even as it pursues lawsuits against AI platforms like Midjourney. Disney’s legal battle is just the first in a wave of anticipated actions to protect its intellectual property in the generative AI era

U.S. Warns DeepSeek May Support Chinese Military Ops

A senior U.S. official has accused Chinese AI firm DeepSeek of aiding China’s military and intelligence agencies. The company allegedly used shell companies to bypass Nvidia chip export controls and may share user data with government entities—prompting a bipartisan bill to ban its tech from U.S. agencies

Google Launches Magenta RealTime for Live Music AI

Google released Magenta RealTime, an open-source, low-latency music model designed for live performance on consumer devices. Built on an 800M parameter transformer, it’s optimized for real-time audio generation and fine-tuning—bringing professional-grade AI music creation to creators everywhere

Thankyou for reading.