Agentem-ai Launches Izwi for Private Voice AI
Agentem-ai has released Izwi, a local-first audio inference engine designed to run entirely on a user's machine, ensuring data remains private. This new open-source platform, licensed under Apache 2.0, supports text-to-speech, speech recognition, and voice AI workflows across macOS, Linux, and Windows without requiring cloud services or API keys.
The software requires a minimum of 8 GB RAM and 10 GB of storage, though the developers recommend 16 GB RAM and 50 GB of storage for optimal performance. It is optimized for Apple Silicon Macs with Metal acceleration and also supports NVIDIA CUDA GPUs.
Core processing features
- Voice Mode for real-time AI conversations.
- Text-to-Speech and Voice Cloning capabilities.
- Voice Design to create custom voices from descriptions.
- Transcription and Diarization to identify multiple speakers.
- Privacy-first local data processing.
System requirements & model support
The platform offers flexibility through a Web UI, desktop app, and command line interface to suit different user preferences. In practice, the system supports various model families including Qwen3-TTS for speech generation and Parakeet for recognition tasks.
Cross-platform support spans Windows, macOS, and Linux, allowing a wide range of developers to utilize the tools. Users can download specific models based on their needs, such as the Qwen3-TTS-12Hz-0.6B-Base for general purpose text-to-speech or specialized models for voice design.
Developer Insights on privacy
The development team emphasizes that privacy and local control are the central advantages of this release.
'Izwi is a powerful, privacy-focused audio AI platform that runs entirely on your machine,'
states the official documentation. The source material reinforces this by noting there are
'No cloud services, no API keys, no data leaving your device.'
This approach allows for secure workflows where sensitive audio data never leaves the local hardware environment.
Learn more about izwi?
- Learn more about the project on izwai.com.
- Access Izwi on Github.
