mercury-tts is the speech-rendering service for HAL.
Its responsibility is narrow:
- accept text or sentence batches over HTTP
- render speech using the configured TTS backends
- return or queue playback work
- expose health and service diagnostics for the speech path
This repo no longer owns:
- runtime orchestration
- chat workflow
- memory/state management
- aware mode
- sensory processing
- review or story generation
- canon or prompt governance
Those concerns belong in other services:
hal-orchestratorhal-memory-fabrichalo-chathal-voice-recognition
Primary local service endpoints:
GET /healthPOST /speak
The current operator helper script is:
These scripts are thin TTS clients only. They are not runtime shells.
Run Go validation:
go test ./...Important files:
Older runtime/state assets were removed from this repo during the orchestrator split. If you need HAL workflow behavior, use hal-orchestrator, not mercury-tts.