flutter_ai_elements 0.1.4

Composable, themeable Flutter UI for AI chat — conversation, message bubbles, a streaming-aware composer, and a loader — styled entirely through a mobile-first AiThemeExtension. No shadcn/forui dependency.

Changelog

0.1.4

• Internationalization: AiLocalizations (+ AiLocalizationsDelegate) holds the widgets' user-facing strings (defaults English). Every previously-hardcoded tooltip/label/action now reads from it, so apps can translate the UI by providing a delegate. AiConversationList.newChatLabel now defaults to the localized value.

0.1.3

• AiConversationList: a ChatGPT-style conversation sidebar (New chat + a list of ChatThreads with select/delete) to pair with a ChatThreadStore.

0.1.2

• Generative UI: AiWidgetRegistry (a dataType→widget allowlist) and AiDataView render DataParts the model emits as your own widgets — no reflection, unknown types fall back. The demo wires its chain-of-thought / task / confirmation cards through it.

0.1.1

• AiChat now anchors the message you just sent to the top of the viewport (ChatGPT-style) and holds it there while the answer streams in below, reserving just enough trailing space and releasing it as the answer grows. A drag releases the pin; a floating "scroll to latest" button appears whenever the conversation is scrolled above the bottom.
• New AiAnimatedResponse: a blur fade-in reveal (the Apple-Intelligence / Siri look) so streamed answers appear smoothly — each newly revealed word arrives blurred and faded, then sharpens into place over fadeDuration (blurSigma controls the starting blur). Text is paced at a readable charsPerSecond (default 120) and accelerates to drain a backlog within catchUpWindow so it never trails far behind a fast stream. Only the few words at the leading edge animate at once, so the cost stays bounded. The in-flight text renders as plain prose and settles into full Markdown once complete. MarkdownTextRenderer uses it automatically while streaming.
• AiMessageActions is restyled with compact, evenly spaced icon buttons (ChatGPT-style) and gains optional onSpeak/onGood/onBad/onShare actions.
• AiSources collapses past maxVisible chips (default 6) behind a "+N more" toggle, so grounded answers that return dozens of sources no longer flood the bubble.
• Pluggable syntax highlighting: AiCodeBlock, AiResponse, and MarkdownTextRenderer accept an optional CodeHighlighter that turns code + language into styled spans. The package ships no grammar engine (stays dependency-free); supply one from the app. Defaults to plain monospace.
• Fixed the Markdown block parser hanging (and exhausting memory) when handed a partial stream that ended mid-construct, e.g. a lone # before its heading text arrived — the parser now always makes forward progress.

0.1.0

Initial release.

• AiThemeExtension — a ThemeExtension of design tokens (bubble colors, shapes, ambient shadow, spacing, typography, motion, haptics) with copyWith, lerp, of(context), and a mobile-first fallback().
• Presentational widgets: AiMessageBubble (renders every AiPart type; streaming-safe semantics), AiConversationView, AiComposer (Send↔Stop swap, haptics), AiLoader.
• Controller-bound widgets: AiChat (live transcript with auto-scroll and a thinking loader) and AiPromptInput.
• AiResponse — a dependency-free Markdown renderer (headings, bold/italic, inline + fenced code, lists, blockquotes, links); MarkdownTextRenderer wraps it and is now the default AiTextRenderer. PlainTextRenderer remains.
• AiChainOfThought (stepwise timeline), AiTask (agent checklist), AiInlineCitation (numbered badge), AiBranch (version navigation), AiImage (loading/error/tap-to-zoom).
• Input upgrades: AiComposer gains an attach button, a model-selector slot, a voice button, and removable attachment previews; AiPromptInput stages attachments and switches models via the controller. AiModelSelector, AiConfirmation (approve/deny), AiContextMeter (token usage), and AiShimmer (loading skeleton).
• AiLiveSession — a full-screen, engine-agnostic Live voice surface (animated orb reacting to amplitude + status, live transcript, mute/keyboard/end). UI only; drive it from a realtime audio engine.
• Performance & a11y hardening: AiResponse parses Markdown and builds gesture recognizers once per text change (not every frame) — important on the streaming hot path; AiLiveSession animates only the orb (60fps no longer rebuilds the conversation); message bubbles no longer subscribe to window size in the common bounded case; list items carry stable keys; the composer measures with the ambient text scale + direction; disclosure widgets expose button/expanded semantics; modal sheets scroll; high-traffic layout uses directional insets/alignment for RTL.
• Re-exports flutter_ai_client (and flutter_ai_core).