Repository navigation
A whole new look
Android 183, iOS 43.
1.4.0 is the first store build with the redesign, on top of the RNE 0.11 engine that shipped in 1.3.6. Chat, the sidebar, models, settings and onboarding are redrawn, and long conversations remember more on phones with plenty of RAM.
- Redesigned chat, sidebar, Models, Settings and Benchmarks screens, with a lighter Liquid Glass style and a new chat screen with an animated gradient (#451, #454, #459, #461, #462, #468, #519, #527, #529)
- Reworked drawer: opening, search and footer (#511)
- Unified model rows, bottom sheets and chat options menu; model selection is easier (#488, #506)
- A model that does not fit says how much RAM it needs instead of "Incompatible", and names the phone when it is blocked (#510, #542)
- A simpler start: the optional features are folded into one onboarding section, the first model download is clearer, new onboarding images, and "Choose a model" opens the Models screen (#509, #524, #533, #535)
- Long conversations keep more history: the window is 4,096 tokens instead of 2,048 on phones with 8 GB of RAM or more when Think is off. Phones under 8 GB, and chats with Think on, stay at 2,048 (#543)
- An outdated model can be updated from its row in Models with an Update button; a cancelled update keeps the old files, and the new files load without restarting the app (#539)
- Phones on Android 12 and older are told that updates need Android 13 (#515)
- Chats open faster: the first chat opens without waiting, the model catalog never blocks startup, and only the latest 40 messages are mounted, with older ones loading as you scroll (#492, #493, #495)
- Chat sources: the cost of the sources is shown next to the cost of the answer, the citation points to the source the model named, and a web search says why it is slow while it plans (#414, #358, #355)
- Dictation ends with a stop button instead of a trash can (#514)
- Model families credit Software Mansion and show the Bielik mark (#520, #521)
- Fixes: the model's prompt scaffolding no longer shows up in answers or thinking, the next message is no longer held up by the conversation digest, a greeting after a question no longer restates the previous answer, downloads report a full disk and survive the background and a dropped connection, and model family icons no longer sit on a grey tile (#541, #540, #491, #523, #522, #536)
Known issues: dictation occasionally fails to start with "Could not start live transcript".
Release branch: chore/release-1.4.0, #545.
What's Changed
- Rework the image attachment flow to match ChatGPT's in-app picker by @kfaracik in #310
- fix(chat): cut a loop at the repeat, not at the copy it repeats (#255) by @kfaracik in #374
- Onboarding: one surface from intro to last slide, illustrations drawn from geometry by @kfaracik in #337
- fix(chat): give the chat title its width, and the rename dialog the menu's look by @kfaracik in #385
- fix(chat): let the Android menu sheet use the width it has by @kfaracik in #343
- fix(web): give a trace step the same key before and after the hand-over by @kfaracik in #345
- perf(drawer): stop the closed drawer re-rendering on every streamed token by @kfaracik in #346
- perf(chat): stop the chat screen redoing per-frame work while an answer streams by @kfaracik in #347
- perf(models): re-render only the card whose download is progressing by @kfaracik in #348
- fix(chat): scope the chat bar's generating state to its own chat by @kfaracik in #340
- chore(deps): react-native-executorch 0.9.2 → 0.10.4 via the legacy API by @kfaracik in #389
- perf(chat): parse a message's thinking blocks only when its text changes by @kfaracik in #350
- fix(android): tint text selection with the brand colour by @kfaracik in #420
- fix(toast): let a toast clear the header instead of cutting through it by @kfaracik in #401
- fix(chat): cut a generated title between words, not inside one by @kfaracik in #400
- fix(chat): drop the keyboard before the embedding download sheet by @kfaracik in #394
- fix(models): say at the list level when no variant of a family can run here by @kfaracik in #396
- fix(models): offer only starting models the device can actually run by @kfaracik in #421
- fix(chat): carry the background gradient across a chat screen remount by @kfaracik in #390
- fix(chat): start the scroll indicator where the messages start by @kfaracik in #402
- fix(chat): show the send button busy while the model loads or switches by @kfaracik in #395
- fix(chat): refuse the exits that strand a running turn (#370) by @kfaracik in #376
- feat(chat): let a new chat arrive in one movement instead of two by @kfaracik in #404
- perf(chat): stop the composer re-rendering on every streamed token by @kfaracik in #353
- Stop is never silent: say when a response was stopped, and make the stop actually land by @kfaracik in #373
- feat(chat): give the scroll-to-latest button an entrance, an exit and a press by @kfaracik in #403
- feat(chat): morph the composer button between mic, send and stop by @kfaracik in #408
- perf(benchmark): make the run reproducible and the memory reading real by @kfaracik in #349
- feat(benchmark): give Android a peak-memory reading of its own by @kfaracik in #410
- ci: get main green on the ubuntu runner again by @kfaracik in #422
- fix(toast): let the cross close the toast when the finger drifts by @kfaracik in #424
- fix(web): keep web search off only for the message a document comes with by @kfaracik in #426
- fix(chat): stop reporting a token rate the run never measured by @kfaracik in #398
- fix(chat): let a suggestion card end where its prompt ends by @kfaracik in #399
- fix(models): let only the current download own the download state by @kfaracik in #409
- fix(chat): let the model in the header be the one that answers by @kfaracik in #415
- fix(chat): greetings, special tokens and the background, in one change by @kfaracik in #427
- fix(chat): stop the measured causes of iOS generation crawling (#384) by @kfaracik in #425
- ci(android-release): stop asking the SDK manager for the removed tools package by @kfaracik in #430
- fix(speech): say what dictation is waiting for, and recover from a dropped transcript by @kfaracik in #405
- perf(llm): warm the model up right after it loads by @kfaracik in #431
- chore(release): 1.3.1 by @kfaracik in #428
- chore(deps): upgrade to expo 57 and RN 0.86 by @KatPo in #445
- refactor(ui): move every touchable to Pressable by @kfaracik in #416
- chore(release): 1.3.1 as shipped to the stores by @kfaracik in #435
- fix(ios): ship a dSYM for ExecutorchLib so its crash frames symbolicate by @kfaracik in #440
- fix(android): disable OpenMP thread affinity before ExecuTorch loads by @kfaracik in #441
- fix(android): tell ARMv8.0 phones they are unsupported instead of crashing (#437) by @kfaracik in #442
- fix(chat): keep the first answer when a refinement retry cannot run (#437) by @kfaracik in #443
- build(android): ship native symbol tables in the release bundle (#437) by @kfaracik in #444
- fix(chat): offer and send photos only for a model that can see them by @kfaracik in #446
- fix(models): load downloaded models from disk instead of the network by @kfaracik in #447
- fix(prompt): drop the default prompt's document sentence on turns without context by @kfaracik in #448
- fix(drawer): show the keyboard when chat search opens on Android by @kfaracik in #450
- feat(ui): starting model selection redesign by @KatPo in #449
- fix(chat): show an answer left in an unclosed think block as the answer by @kfaracik in #455
- release: 1.3.3 (Android 167, iOS 23) by @kfaracik in #456
- fix(chat): keep Gemma from hanging and crashing when the iPhone runs short of memory (UW-65) by @kfaracik in #457
- fix(speech): release the dictation model once its transcript stream ends (UW-65) by @kfaracik in #464
- fix(chat): give a photo sent alone a question the model can answer by @kfaracik in #465
- fix(onboarding): stop the model list clipping the selected card's border on Android by @kfaracik in #466
- fix(chat): ask for a store review only after a good answer, never on first launch by @kfaracik in #458
- release: 1.3.4 (Android 172, iOS 24) by @kfaracik in #467
- feat(ui): chat screen restyling by @KatPo in #451
- feat: settings redesign by @KatPo in #454
- feat: models screen redesign by @KatPo in #459
- feat: benchmarks screen redesign by @KatPo in #461
- feat(ui): update remaining components by @KatPo in #462
- feat: redesign follow-ups by @KatPo in #468
- feat(onboarding): update images in onboarding by @KatPo in #469
- feat(onboarding): improve onboarding on tablets by @KatPo in #471
- fix(onboarding): list only starting models the device can run by @kfaracik in #487
- test(i18n): fail when a new per-language word list appears by @kfaracik in #356
- docs(readme): launch-ready README with a new banner and feature grid by @kfaracik in #485
- build(ios): declare that the app uses no non-exempt encryption by @kfaracik in #470
- fix(web): let the time-scope rule cover a span of days, not just one by @kfaracik in #351
- docs(readme): give the images rounded corners on a transparent background by @kfaracik in #497
- build: react-native-executorch 0.11.0 with chunked multimodal prefill and chat-template flags by @kfaracik in #498
- feat(models): add LFM 3 1.7B and LFM 3 VL 2B with thinking driven by the chat template by @kfaracik in #499
- fix(chat): LFM 3 reasoning fixes from the 1.3.5 regression by @kfaracik in #500
- fix(chat): language anchor for Latin-script questions, English request words, Whisper silence annotation by @kfaracik in #501
- feat(catalog): gate manifest entries by minimum app version by @kfaracik in #503
- feat(catalog): add LFM 3 1.7B and LFM 3 VL 2B for apps from 1.3.5 by @kfaracik in #504
- release: 1.3.5 (Android 173, iOS 26) by @kfaracik in #502
- fix(stt): pause model downloads while Whisper is fetched on first dictation by @kfaracik in #490
- perf(startup): create the web-search WebView when a search needs it by @kfaracik in #494
- perf(startup): open the first chat without the 50 ms navigation wait on a cold start by @kfaracik in #495
- fix(chat): cite the source the model named, not the one it reads like by @kfaracik in #358
- perf(chat): keep message rows memoized while the list re-renders by @kfaracik in #489
- fix(chat): greeting after a question no longer restates the previous answer (#417) by @kfaracik in #491
- fix(sheets): fix content disappearing and sheet jumping when closing by @KatPo in #496
- feat(web): say why a search is slow while the planner is still running by @kfaracik in #355
- perf(startup): never wait for the model catalog, apply a fresh one live by @kfaracik in #493
- perf(chat): mount only the latest 40 messages and load older ones on scroll by @kfaracik in #492
- feat(chat): say what the sources cost, next to what the answer cost by @kfaracik in #414
- refactor(ui): unify model rows, bottom sheets and chat options menu by @KatPo in #488
- feat(ui): model selection improvements by @KatPo in #506
- feat(onboarding): initial model download improvements by @KatPo in #509
- feat(models): say how much RAM a model needs instead of "Incompatible" by @kfaracik in #510
- feat(drawer): rework drawer opening, search and footer by @KatPo in #511
- chore: prepare the repository for open source by @kfaracik in #512
- feat(chat): redesign new chat screen by @KatPo in #519
- fix(dictation): end a dictation with a stop button instead of a trash can by @kfaracik in #514
- feat(models): credit Software Mansion on the model family screen by @kfaracik in #520
- feat(models): show the Bielik mark on the Bielik family tile by @kfaracik in #521
- fix(downloads): report a full disk, survive the background and a dropped connection by @kfaracik in #523
- fix(models): show family icons without a grey tile, sized to the text by @kfaracik in #522
- release: 1.3.6 (Android 182, iOS 42) by @kfaracik in #531
- feat(chat): new chat animation and gradient update by @KatPo in #527
- feat(chat): liquid Glass and layout polishing by @KatPo in #529
- feat(onboarding): images in onboarding update by @KatPo in #533
- feat(notice): improve positioning and style by @KatPo in #532
- feat(welcome): open the Models screen from Choose a model by @kfaracik in #535
- fix(models): show the family icon without a grey tile on details and the family header by @kfaracik in #536
- feat(android): tell phones on Android 12 and older that updates need Android 13 by @kfaracik in #515
- feat(onboarding): fold the optional features into one section by @kfaracik in #524
New Contributors
Full Changelog: 1.3.0...1.4.0