Repository navigation
v1.4.2
Pre-release
HeetKit v1.4.2
This pre-release adds DeepSeek alongside Gemini, saved AI connections and model
controls, and a clearer operator activity view on Desktop and Server. These changes
build on the previous stable release, v1.4.1. This is a pre-release for testing.
What's New
- DeepSeek support: Use the direct DeepSeek API for AI chat and the Desktop
Playground, with its own securely stored key, model discovery and saved settings.
Gemini remains available, and ordinary Twitch commands still work without AI. - Saved AI setup: Manage connected providers and choose primary and optional
fallback models from one AI setup section on Desktop or Server. Each selection
keeps its own provider, model, thinking and search settings. Cross-provider fallback
requires an explicit acknowledgement before questions and permitted context can
be sent to another provider. - Thinking and Internet search: Configure supported model capabilities, including
bounded DeepSeek Search/Fetch requests. Provider support and charges apply; unknown
model IDs do not gain invented capabilities. Gemini grounded output is withheld
where its required Search Suggestions cannot be displayed in plain Twitch chat. - Editable shared instructions: Save profile-wide AI instructions separately from
model choices, personality prompts and profile instructions. Configure question and
answer limits together in AI setup, with options to restore the shared defaults. - Operator activity: Follow command attempts, AI processing and outgoing replies
through linked events in Logs, with clearer distinctions between confirmed,
failed and uncertain sends. The authenticated Server panel uses the same activity
information as Desktop. - Larger random response pools: Custom commands can store up to 30 responses.
Improvements & Fixes
- One shared AI pipeline applies language, filtering, privacy, cancellation and output
rules across Gemini and DeepSeek. Fallback remains limited to a model-unavailable
response, rather than retrying authentication, billing or network failures elsewhere. - Reworked AI setup and Twitch Settings with aligned fields, clearer credential
actions, separate save boundaries and compact help on both hosts. - Fixed Dashboard setup summaries that looked incomplete after configuration, aligned
the resumed setup guide and corrected narrow-screen Server operation controls. - Fixed missing AI and Settings stylesheets in the Docker build context.
- Added a private Linux release-update helper with stopped-profile backups and
explicit rollback compatibility checks. Deployment remains an operator action. - Separated quick checks, full validation, candidate builds and publication, with
immutable candidate evidence and checksums. Updated setup and provider guides.
Updating
Back up the existing profile first. Quit Windows Desktop fully through tray Exit
before installing or replacing portable files. Existing profiles and credentials
remain supported. Add DeepSeek keys separately; configuration transfer excludes
credentials and clears imported cross-provider fallback consent until acknowledged
again on the destination. The AI Playground remains Desktop-only.
For Server, follow the Docker update and backup procedure.
Keep the same Compose project, persistent volume, external secrets and architecture;
retain the previous image for compatible rollback. Run only one instance per Twitch
bot account. This release does not automatically update a running server.
Windows downloads remain unsigned and retain the existing .NET Framework and
WebView2 requirements. Publication remains subject to the recorded
redistribution requirements.