local_inference_flutter 0.1.0
local_inference_flutter: ^0.1.0 copied to clipboard
Flutter layer of local_inference: a customizable chat UI, model setup, background model download and a device plugin for on-device LLM chat on iOS, Android and macOS.
Changelog #
0.1.0 #
Initial release.
- Chat UI in three tiers:
LocalInferenceChat(one widget that makes, starts and disposes the model manager, the assistant and the model setup),ChatPageandChatViewover anAssistantControllerand aModelSetupControllerkept above the Navigator (LocalInferenceScope), and headless Listenable adapters. - Model setup UI: consent with the download size, progress with pause, resume and cancel, the mobile-data prompt, the
model picker and
ModelLibrary. - Customization:
LocalInferenceTheme(aThemeExtension),LocalInferenceStrings(Italian and English, withLocalInferenceLocalizations.delegate),ChatFeatures,ChatBuildersandModelSetupBuilders(each builder can wrap the default widget),ChatCallbacksandLocalInferenceKeys. - Compliance and accessibility: an AI disclosure, reply reports, an AI-generated semantics hint, live regions, screen-reader announcements, 48-point targets, text scaled up to 2.0 and right-to-left layout.
- Background model download on background_downloader (
BackgroundModelDownloader): downloads that go on while the app is in the background, Wi-Fi only or on mobile data, resume after a failure or a relaunch (RetainingResumeStorage), the content checked before a download counts, and notifications in Italian or English. LocalInferenceFlutter:installDownloaderStorage,createModelManager,createAssistantController,platformDefaults,platformLoadOptionsandcheckDeviceSupport.- A device plugin for iOS, macOS and Android: memory, free disk, a native SHA-256, the backup exclusion, the no-backup folder, the network type and a diagnostics snapshot (thermal state, battery, memory).
LifecyclePolicy: what a running turn and the model do when the app leaves the foreground.- The licenses of the bundled native libraries on the app's license page, registered before
main()with no app code (NativeLicenses). package:local_inference_flutter/testing.dartre-exports local_inference's test doubles.dart run local_inference_flutter:doctorruns the host doctor;--install-ios-scriptadds the build-phase script that fixes ITMS-90208 for the llama.cpp framework.- Flutter 3.44 or newer. iOS 16.4, macOS 14.0, Android minSdk 24 with API 28, arm64-v8a and a 64-bit process checked at run time (not run on an Android device yet). Linux and Windows get the license notices only.