Skip to content

Konjugate v0.5.12: A conversational AI assistant and shape library previews

Choose a tag to compare

@github-actions github-actions released this 18 Aug 04:55
· 71 commits to master since this release

AI assistant

The model assistant now holds a conversation instead of treating every request in isolation. It remembers the last few turns of back-and-forth, so a follow-up like "make it 20% higher" correctly builds on what you just asked for rather than starting from scratch.

When a request is ambiguous — an unclear target, a missing value, or an unspecified node to disable — the assistant now asks a clarifying question instead of failing outright, and where possible offers suggestion chips you can click to fill in an answer without retyping.

Fixed a bug where the prompt field wasn't cleared after a question was answered, and a CSS issue that could show the assistant's empty state and its clarification prompt stacked on top of each other. The "New conversation" button is now a clearer "+" icon and asks for confirmation before clearing your conversation history.

AI-generated proposals that use LaTeX operator notation (\operatorname{Max}, \operatorname{Min}, and similar) for named functions are now understood correctly instead of being rejected by validation.

Shape library

Browsing the shape library no longer means clicking through shapes one at a time to see what they look like. Every shape now shows a static thumbnail directly in the grid, generated once and cached, so you can scan the whole library at a glance.

Selecting a shape also opens a live, rotatable 3D preview in a detail pane — drag to inspect it from any angle before committing. Applying a shape is now an explicit step (an "Apply shape" button) rather than happening the instant you click a card, so a stray click can no longer swap a node's geometry by accident.

Also in this release

  • The Example Guide window's scrollbar no longer cuts through the title bar.
  • Fixed a source of intermittent interaction test failures around drag-based node selection.