Update 1.1.8 OBT

Changes

• Reworked and optimized settings and other parts
• Updated prompt manager
• Prompt exports in .md
• Prompt imports in .md

Fixes

• Fixed issue with chats not being saved
• Fixed issue with resource names not being translated
• Fixed issue with prompts not being downloaded from GH

Update 1.1.7 OBT

Changes

• Auto-continue on long responses
• Per-chat temperature
• Reworked and optimized settings

Known Issue

• Chats may not save
• Prompts are not loaded from GitHub
• Chat from "Testing prompt" counts towards total chat counter
• Autocoplete is in testing and may not behave correcly:
  • It may prompt model to continue when it's answer is finished (especially if the answer took exactly 25 seconds to complete)
  • It may NOT prompt model to continue even if it was cut-off

Update 1.1.6 OBT

Changes

• Prompt manager
• Prompt editor
• Per-chat prompts
• System and user prompts are replacing master and user prompts, now you are in control!
• DB storage instead of shared preferences
• Early implementation of remote configs from Firebase

This is a prototype of the new prompt engine for evaluation, features are not final, translations are not done and it may break compatibility both with older and newer versions

Update 1.1.5 OBT

Changes

• Fixed Google Analytics misreporting
• Added Google Crashalitics
• Fixed downloading logs
• Added logs to the intro screen for debugging (Sorry, Basim)
• Fixed app getting stuck in offline mode
• Added Turkish - thanks, mikropsoft!
• Added German - thanks, NewNiklas!

Update 1.1.2 OBT

Changes

• Added optional analytics for debugging and crash reporting
• Added internal logs for the app

Update 1.1.1 OBT

Changes

• Chat list now shows status of generation in the correct chat
• You now can jump only to the chat that is being generated now
• New chat card is at the top of the list
• "AI Title" function now describes the whole chat, not the first message only
• Some code refactoring

Fixes

• Fixed that "Pinned" category would appear even if there are none
• Fixed that "Other chats" category was not shown when there are pinned chats
• Fixed the issue with messages appearing in wrong chats
• Fixed the issue where messages were not in order

Miscellaneous

• Changed color variables throughout the app (you shouldn't notice any changes visually, it was internal)

Update 1.1.0 OBT

Changes

• Multichats!
  • You can now have more than one chat and switch between them.
  • You can now pin chats to the top of the list.
  • You can now rename chats yourself or get a title from Gemini Nano.
• Updated Settings
  • Adding per-chat settings, some options were moved from settings to per-chat ones.
• Refactoring and optimisation
  • Rewrited dart-side of the generator,
  • Removed obsolete files,
  • Optimized file structure.
• Bug fixes
  • (some of them were fixed just by changing the structure)
• Master Prompt changes
• Ability to go instruction-less
  • When toggled, model will only receive minimal instructions with chat history.
  • This WILL produce weird and incorrect responses, with lots of hallucinations.
  • Use at your own risk!
• Auto-retry now works better and in more situations

Fixes

• Fixed some unwanted prompting behavior in the master prompt

Known Issues

• Rapidly switching between chats (especially when the AI is still generating) may cause messages to not save.
• Exiting the chat into the chat menu when generating may cause some temporary lagging.
• Closing the app before response generates may cause loss of the last generated text.
• Closing the chat before the name of the chat changes will likely make chat history empty.
• Typing curses or slurs may soft-lock the app if the auto-retry is enabled
  • This happens because model hits an internal censorship and reports an error;
  • When encountering error while generating, with auto-retry, app sends the same input back, getting the same error;
  • This feedback loop will not end untill the app is restarted, and it also consumes huge amount of battery life.

Update 1.0.1 OBT

Changes

• New Intro page
• Model download status display
• Updated Settings
• Code optimisation
• Bug fixes
• Master Prompt changes

Fixes

• Fixed the issue with app skipping the initialization when model is unavailable
• Fixed issue with app not requesting model download
• Fixed issue with translations not being loaded from repository

Known Issues

• The AI is unstable so it can:
  • Ignore requests
  • Generate erroneous information
  • Mix historical events with current time
  • Hallucinate
  • Ignore instructions
  • Mix requests and responses in chat log

Initial Release