Skip to content

v1.4.0

Latest
Latest

Choose a tag to compare

View all tags
@bigsk1 bigsk1 released this 01 Aug 09:54
v1.4.0
062cb7b

Voice Chat AI v1.4.0

Version 1.4.0 modernizes the supported AI providers, resolves vulnerable and outdated dependencies, expands voice and local-model configuration, and adds a live server-log panel to the dashboard.

Highlights

  • Added a collapsible, real-time server-log panel to the dashboard, including connection status, history, follow mode, and a dedicated /ws/logs stream.
  • Added Grok 4.3 and Grok 4.5 chat support with model-aware reasoning settings and timeouts.
  • Expanded the xAI TTS picker from the original five voices to the current 26 built-in voices.
  • Added current OpenAI low-latency model choices and request compatibility handling.
  • Refreshed Anthropic choices to Claude Sonnet 4.6, Sonnet 4.5, and Haiku 4.5.
  • Added Kokoro-FastAPI v0.4 voice-response compatibility, mixed voices, and restoration of saved/custom voices in the dashboard.
  • Added optional Ollama context-window configuration.
  • Updated vulnerable and outdated Python dependencies across standard, CPU Spark-TTS, CUDA Spark-TTS, and uv installations.

Provider and model updates

xAI

  • grok-4.3 is now the default chat model and runs with reasoning disabled for responsive voice conversations.
  • grok-4.5 is available as an optional low-reasoning model with a longer default request timeout.
  • Retired grok-4-1-fast-non-reasoning configurations are normalized automatically to grok-4.3.
  • Image understanding now uses configurable XAI_VISION_MODEL, defaulting to multimodal grok-4.3, instead of the retired grok-2-vision-1212 model.
  • The dashboard restores the saved xAI model selection correctly after reload.
  • The xAI TTS picker now includes 26 built-in voices with descriptions.

OpenAI

  • Updated the OpenAI SDK from 1.68.2 to 2.52.0.
  • Added GPT-5.6 Luna, GPT-5.4 mini, and GPT-5.4 nano choices.
  • Added shared request compatibility logic so supported low-latency models use reasoning_effort="none" and current token-limit parameters.

Anthropic

  • Updated the Anthropic SDK from 0.49.0 to 0.120.2.
  • The dashboard now offers Claude Sonnet 4.6, Claude Sonnet 4.5, and Claude Haiku 4.5.
  • Removed deprecated Claude 3.5 and Claude 3.7 choices from the dashboard.

Ollama

  • Added optional OLLAMA_NUM_CTX support for configuring the model context window.
  • Invalid or non-positive values are ignored safely.
  • Improved initial provider/model selection behavior in the dashboard.

Kokoro

  • Supports both legacy string voice lists and Kokoro-FastAPI v0.4+ voice objects.
  • Supports mixed saved voices such as af_bella+af_sky or weighted mixes such as af_bella(2)+af_sky(1).
  • Saved or custom voices remain selectable even when they are not returned by the server voice list.

Environment changes

Review and merge the following settings from the new .env.sample into existing .env files:

Variable Status Description
XAI_MODEL Default changed Now defaults to grok-4.3; grok-4.5 is also supported.
XAI_VISION_MODEL New Image-understanding model; defaults to grok-4.3.
XAI_CHAT_TIMEOUT New, optional Overrides xAI chat/image request timeouts. Defaults are 45 seconds normally and 120 seconds for Grok 4.5.
OLLAMA_NUM_CTX New, optional Sets Ollama num_ctx; larger values require additional RAM/VRAM.
KOKORO_VOICES_LEGACY New, optional Set to true only when a Kokoro server requires the legacy voice-list response.
UI_SERVER_LOGS New Defaults to true; set to false to disable the dashboard server-log panel and log WebSocket.
KOKORO_TTS_VOICE Enhanced Now accepts mixed and weighted voice expressions.
XAI_TTS_VOICE Expanded The existing variable now supports the expanded built-in voice catalog shown in the dashboard.

No environment variables were removed. Existing .env files are not overwritten during an upgrade.

Dependency and security updates

Important upgrades include:

  • transformers: 4.53.0 → 5.5.0
  • openai: 1.68.2 → 2.52.0
  • anthropic: 0.49.0 → 0.120.2
  • aiohttp: 3.13.4 → 3.14.1
  • cryptography: 46.0.7 → 48.0.1
  • pillow: 12.2.0 → 12.3.0
  • fastapi: 0.115.11 → 0.141.1
  • pydantic: 2.7.4 → 2.13.4
  • Added pinned nltk 3.10.0
  • Updated the build requirement to setuptools>=83.0.0

The dependency updates are synchronized across pyproject.toml, uv.lock, and the standard/CPU/CUDA requirements files.

Breaking changes and upgrade notes

  • Reinstall or synchronize Python dependencies when upgrading; this release includes major-version upgrades for Transformers and the OpenAI SDK.
  • Custom code that imports the OpenAI or Anthropic SDK directly may need adaptation for their newer APIs. The application’s built-in provider paths have already been updated.
  • Deprecated Claude 3.5/3.7 model IDs were removed from the dashboard. Update a manually configured ANTHROPIC_MODEL if it still references one of those models.
  • The old xAI chat and vision model IDs are retired. The application automatically maps its previously supported retired values to Grok 4.3, but updating .env is recommended for clarity.
  • CUDA users should continue following the CUDA/Spark-TTS installation instructions so the correct PyTorch CUDA build is preserved when refreshing dependencies.
  • Docker users must rebuild their local image or pull the v1.4.0 image when available for the updated dependencies and application code to take effect.
  • No database or conversation-history migration is required.

Validation

  • 36 automated tests pass in the standard CPU virtual environment.
  • 36 automated tests pass in the Python 3.11 CUDA virtual environment.
  • PyTorch 2.6.0+cu124 and the CUDA 12.4 build were preserved during validation.
  • Live xAI validation passed for Grok 4.3 streaming, Grok 4.5 streaming, and Grok 4.3 image understanding.
  • Updated OpenAI, Anthropic, and xAI voice paths were also tested live during development.

Full changelog: v1.3.0...v1.4.0

Assets 2
Loading