Skip to main content

Demiurge 0.4.1

Demiurge is alpha / developer-preview software. APIs, runtime layout, and authoring contracts may still change.

Summary

This patch release adds domestic speech-to-text packages, shares common STT input handling across providers, and completes the current public manual and release surfaces without publishing wheel or sdist artifacts.

User-Visible Changes

  • Added stt_baidu, stt_dashscope, and stt_tencent package recipes for provider-specific speech-to-text input.
  • Added Baidu Cloud Speech Recognition support, including access-token lookup from DEMIURGE_BAIDU_ACCESS_TOKEN or OAuth token retrieval from Baidu API and secret keys.
  • Added Alibaba Cloud Model Studio / DashScope STT support through the compatible chat-completions audio API.
  • Added Tencent Cloud ASR support with TC3 request signing.
  • Shared STT attachment filtering, network capability checks, transcript rendering, and metadata handling through the common STT library.
  • Added an Agent Slots explanation page and updated slot-module contract docs.
  • Completed the zh-CN website manual pages for the current documentation set.
  • Added README and website demo media for speech-to-text input and text-to-speech output packages.

Fixes

  • Kept GitHub Release publication aligned with the current alpha policy: release notes are published, but wheel and sdist artifacts are not built or attached by default.
  • Removed duplicated STT input-slot implementation from individual provider packages.
  • Updated package documentation, CLI references, and release links to reflect the current package and slot surfaces.

Breaking Changes

  • package-repository/skill/stt_transcription/SKILL.md was removed. The STT packages now carry their reusable behavior through package components and the shared STT library.

Known Limitations

  • Code slots run in the host-shared Python environment by default.
  • Candidate Agent Core evolution cannot add dependencies automatically.
  • Telegram is still long-polling only.
  • Package repositories can be local, built-in, or explicitly trusted Git/path sources; install-time trust still belongs to the host.
  • Provider profile tests and real STT provider calls require explicit provider credentials and may make live network calls.
  • GitHub Releases do not attach wheel or sdist artifacts by default yet; use the source checkout or managed checkout install path for this alpha release.

Verification

  • uv run python -m compileall demiurge tests
  • uv run pytest -vv --durations=20 -o faulthandler_timeout=60
  • uv run demiurge --help
  • uv run demiurge init --help
  • (cd ui-tui && npm ci && npm run typecheck && npm test -- --run && npm run build)
  • cmp ui-tui/dist/entry.js demiurge/ui/tui_dist/entry.js
  • (cd website && npm ci && npm run build)
  • git diff --check -- ':!demiurge/ui/tui_dist/entry.js'
  • uv run python -c 'import tomllib, demiurge; print(tomllib.load(open("pyproject.toml", "rb"))["project"]["version"]); print(demiurge.__version__)'
  • rg -n "0\\.4\\.1|docs/releases/0\\.4\\.1" pyproject.toml demiurge/__init__.py uv.lock README.md README.zh-CN.md docs website