Overview
Hush is an open-source noise suppression model by weya AI, designed to clean real-time call audio for voice AI agents. It removes background noise, competing voices, and interference, making speech clearer for ASR systems. Hush is lightweight (8 MB), runs on CPU, and is trained on 10,000+ hours of noisy audio. It's ideal for BFSI teams, lenders, and any voice AI deployment in noisy environments.
AgentMicro is a local-first macOS menu-bar companion for supervising parallel Codex Desktop and CLI tasks. It shows task state, project, elapsed time, and results at a glance, and opens the matching Codex Desktop task with one click. It reads local session metadata only, ensuring privacy—no prompts, code, or history are uploaded. It's free, open-source, and community-maintained.
Feature Comparison
| Feature | Hush | AgentMicro |
|---|---|---|
| Primary Function | Real-time noise suppression for voice AI calls | Task supervision for Codex |
| Target Users | Voice AI developers, BFSI teams | macOS developers using Codex |
| Deployment | Cloud/on-premise model | Local macOS app |
| Privacy | Self-hostable; audio processed in deployment | Local-only, no uploads |
| Performance | <1ms per 10ms frame on CPU | Instant status display |
| Integration | Voice AI stack, ASR systems | Codex Desktop/CLI |
| Pricing | Free open-source; pilot available | Free open-source |
| Platform | Cross-platform | macOS 14+ |
| Use Case | Noisy call environments | Parallel coding tasks |
Pricing
Hush is free and open-source, available on GitHub. Weya AI offers a 2-week pilot program where they implement, measure, and show lift before full rollout. Enterprise pricing is likely based on support and customization.
AgentMicro is completely free and open-source. There are no costs, and it's community-maintained. No premium tiers or paid features are mentioned.
Pros and Cons
Hush
Pros:
- Real-time noise suppression on CPU with low latency
- Open-source and lightweight (8 MB)
- Trained on 10,000+ hours of real-world noisy audio
- Improves ASR accuracy and reduces repeat requests
- Top-5 on Hugging Face audio-to-audio leaderboard
Cons:
- Requires integration into existing Voice AI stack
- May need pilot and tuning for specific use cases
- Audio data processing may raise privacy concerns if not self-hosted
AgentMicro
Pros:
- Local-first privacy: no data uploads
- Free and open-source
- Quick glance at task status and results
- One-click open in Codex Desktop
- No Accessibility permission needed in base mode
Cons:
- macOS only
- Limited to Codex tasks
- No advanced features like noise suppression or audio processing
Verdict
Choose Hush if you need to improve voice AI call quality in noisy environments, especially for ASR accuracy. Choose AgentMicro if you're a macOS developer looking to manage parallel Codex tasks privately and efficiently. Both are open-source and free, but serve entirely different purposes.