Case study
Hermes Live Voice
Hermes Live Voice adds continuous voice control to Hermes Agent. Background tasks continue while users speak, receive progress, or interrupt the conversation.
Founder-led open-source projectProject factsPublic release
Published npm package
Protocol v6 clients
Documented security model
MIT license
Problem
Why this project exists
Long agent tasks can block a voice conversation. A complete system must preserve task identity, progress, cancellation, reconnection, and private local operation.
Approach
System design
The system separates each live voice turn from supervised Hermes tasks. A protocol gateway controls routing, task lineage, progress, recovery states, and client connections.
Result
Current release
The public package supports local speech, Gemini Live, and OpenAI Realtime. It includes security boundaries, diagnostics, upgrade paths, and a dependency-free browser client.
Technical details
Stack and source
- TypeScript
- WebSocket
- Hermes Agent
- Realtime voice
- macOS