Your coding agent can now talk to you. Hermes just shipped proper text-to-speech support through its managed audio gateway — and it’s surprisingly good.
What It Does
The TTS feature coerces a direct OpenAI model on the managed audio gateway, meaning Hermes can generate natural speech from any text output. Ask it to explain a concept, read back code, or narrate a debugging session — and it responds with spoken audio.
The commit (fix(tts): coerce direct-only OpenAI model on the managed audio gateway) handles the routing complexity: TTS models have different API shapes than chat models, and the gateway needs to recognize when a TTS request is coming in and route it to the right endpoint.
Why Voice Matters
Voice is the overlooked modality in coding agents. Everyone focuses on vision, tool use, and code generation. But voice is what makes an agent ambient — something you can talk to while doing other things.
With TTS, Hermes can:
- Read error messages aloud while you keep your eyes on the code
- Narrate a code review while you drive
- Explain architectural decisions while you sketch
- Read back documentation while you keep your hands on the keyboard
The Audio Gateway Architecture
The TTS route goes through Hermes’s managed audio gateway, which handles:
- Model selection (OpenAI TTS models are different from chat models)
- Audio format conversion
- Streaming vs. full-response modes
- Provider fallback (if OpenAI TTS fails, fall back to a different provider)
This is the same gateway that handles image generation and vision — a unified multimedia pipeline.
What’s Next
TTS is phase one. The audio gateway’s architecture also supports speech-to-text for voice input, which would let you talk to your agent instead of typing. Voice-activated coding agents aren’t far off.