Text-to-Speech Packages

Built-in text-to-speech packages generate audio from assistant replies. Each provider installs an output slot and can optionally install an authored text_to_speech tool for on-demand generation.

Install one TTS output package per core unless you intentionally want multiple audio providers to run after every assistant response.

Packages

Package	Provider	Capabilities
tts_minimax	MiniMax	network.fetch, agents.run:tts_summarizer
tts_openai	OpenAI	network.fetch, agents.run:tts_summarizer
tts_gemini	Gemini	network.fetch, agents.run:tts_summarizer
tts_xai	xAI	network.fetch, agents.run:tts_summarizer

Install

Use the interactive manager:

uv run demiurge package

Or preview and install with subcommands:

uv run demiurge package install tts_minimax --core assistant --preview
uv run demiurge package install tts_minimax --core assistant

Install summary mode:

uv run demiurge package install tts_minimax \
  --core assistant \
  --option mode=summary

Install with the on-demand tool:

uv run demiurge package install tts_minimax \
  --core assistant \
  --option enable_tool=true

Shared Options

All built-in TTS packages support:

Option	Default	Description
mode	direct	direct speaks the assistant reply as-is. summary installs the shared tts_summarizer child core and speaks its summary. Choices: direct, summary.
enable_tool	false	Installs the authored text_to_speech tool and a provider-specific TTS skill.
api_key	unset	Optional direct provider API key. Leave empty to use environment variables.

In summary mode, the package installs or reuses a package-owned child core named tts_summarizer.

Credential Environment Variables

Package	Environment variables
tts_minimax	DEMIURGE_MINIMAX_API_KEY; optional DEMIURGE_MINIMAX_GROUP_ID
tts_openai	DEMIURGE_OPENAI_API_KEY or OPENAI_API_KEY
tts_gemini	DEMIURGE_GEMINI_API_KEY, GEMINI_API_KEY, or GOOGLE_API_KEY
tts_xai	DEMIURGE_XAI_API_KEY or XAI_API_KEY

Runtime Behavior

The installed output slot runs in the parallel output pipeline after the model response is available. In direct mode it strips common Markdown from the assistant reply, truncates to the configured provider limit, sends the text to the provider, and emits an audio artifact.

In summary mode it first calls the tts_summarizer child core, then synthesizes the returned summary.

Generated audio is written under the workspace:

.demiurge-tts/

Channels that support audio can forward the audio artifact to the user.

Tool

When enable_tool=true, the package installs:

agent/tools/text_to_speech/

The model can use that tool to generate a specific audio clip on demand instead of speaking every assistant response through the output slot.

Because all provider packages target agent/tools/text_to_speech, install only one tool-enabled TTS provider in a core at a time.

Verify

List installed packages:

uv run demiurge package list --core assistant

Run a turn and inspect the workspace:

.demiurge-tts/

If enable_tool=true, inspect the tool registry in the TUI:

/tools

Uninstall

uv run demiurge package uninstall tts_minimax --core assistant --preview
uv run demiurge package uninstall tts_minimax --core assistant

Uninstall removes package-owned output slots, provider libs, optional tool and skill files, and package-owned pipeline entries. It does not remove generated audio under .demiurge-tts/.