17.06.2025

- Basic AI integration
- Client based
- No System prompt
- No Error handling
- No Memory

All this was was just a simple AI Discord Chat bot

17.06.2025

- System Prompt created

17.06.2025

- Changed from Client to Bot
- Added /command support
- Added Memory Block

18.06.2025

- Added Temporary memory for up to 6 messages per user per channel
- Added /clear, /roll and /insult commands
- Improved System prompt

18.06.2025

- Performance and memory management improvements
- Improvements to profanity handling

V0.23 and V0.24 were smaller patches, no changelogs were made

- No recorded changes

19.06.2025

- Added /panic and /resume
- Added message cooldown spam protection
- Massively improved response times
- Cleaned up the spaghetti in the code

19.06.2025

- Improved System prompt and spam protection (Thanks Axe)

20.06.2025

- Improvements to the system prompt (thanks to every pen tester)
- Added more info to the /status command
- Added proper error handling, error codes
- Disabled forced restarts

V0.28 and V0.29 were attempted fixes that have been reverted due to stability

- Attempted Memory block improvements

20.06.2025

- Cleaned up the code
- Added /hangman game
- Made system prompt more strict
- Added profanity filters (thanks to the Goog Server)
- General improvements to the /status command
- Smaller fixes to the Memory block

20.06.2025

- General bugfixes (loads of bugfixes, again thanks to all the testers)
- Cleaned up the code further
- About a 20% improvement to response times
- Expanded memory to 10 messages per User per Channel

20.06.2025

- Adjusted parameters based on user feedback

20.06.2025

- Reworked query_gemma as an async function
> this prevents bot freezing and improves heartbeat stability
- Replies of more than 2000 characters are now auto split
> this prevents the bot crashing, restarting and loosing all memory cause the API refused to handle the request
- Removed /insult command as it stopped working as intended

20.06.2025

- Multiple rewrites to the code base and System prompts without any tangible improvement,

21.06.2025

- Switched AI Models from Gemma3:12B to Qwen3:14B
actual reasoning and thinking, basis for massive future improvements
- Fixed crashes due to character limit
messages are now split if over 2000 characters long even if in Debug
- Reverted temp memory to only 6 messages per channel
- Thinking process is now used for debugging in a separate channel

21.06.2025

- Rewritten system prompt
- Emoji usage is now a lot more context based
- Massively enhanced profanity handling, including Leetspeek filters
- fixed Ghost channel and double prompt issues
- Reimplemented full error handling specific to the new model

22.06.2025

- Added /slap command,
- Removed some code that was specific to the old AI model used
- Improved response times by around 15%

22.06.2025

- Added /giveaway command (only triggerable by Administrator)
- Added /hug command (works the same way as /slap)
- Improved embed formatting for slash commands.
- Cleaned up some code so this doesn't affect the rest of the code

23.06.2025

- Refactored core structure into cogs!
> commands and anything that isn't the central AI are now neatly organized into cogs
- Cleaned and centralized shared state handling (ai_enabled, conversation_history)
- Improved reliability of command syncing and error handling
- Squashed numerous command bugs caused by legacy logic
- All prior features preserved and functional
> This change allows me to speed up development by a considerable amount of time and makes the structure more modular, stable and expandable

25.06.2025

- /Panic And /Resume are now changing the server state and not the global ai_enable state
this means the AI wont be disabled in all current servers anymore but only in the server that wishes so
- Bolt now supports /name set (newname) and /name reset commands
> Names are stored in names.json and persist even if Bolt restarts.
> Custom names are respected in AI replies instead of your Discord display name.
- Main code is now lean and clean, just the main AI, command syncing and cog loading.
- Extended context window to 12 messages per channel due to improved memory handling

25.06.2025

- /Name command now actually works....
- Fixed /Status command as well cause it still used the global ai_enabled state

26.06.2025

- Added full suite of moderation features
- Bans-Unbans, Mutes-Unmutes, Kicks, (no Warns yet, will be added later)
- Moderation features can be turned on or off depending if the server wants them or not
- Full message logging for changes and deletions (channels for this can be selected when toggling the feature on
- Crapton of bugfixes and stability improvements
- Several fail safes for the mod features have been added
- Improved error logging and handling
- Global kill switch implemented for the AI
- Added a /blacklist command so admins can restrict users from using the bot entirely
> currently after a user is blacklisted the bot will just respond with HEHEHEHA! and a clash royale gif.... I thought that was funny okay?

Several improvements are planned still but I have been working on this for 12 hours my brain is fried Known Problems:

- AI can be a bit to sassy and uses too many emojis in some cases
- Parts of the chatlog are sometimes included in the message
- Will repeat itself sometimes due to how the memory block is handled
- Profanity filter is still too sensitive
- /blacklist causes the HEHEHEHA to be sent after every message that's being sent by the blacklisted User

27.06.2025

- Added rich presence statuses
- Added suffix system with clean and dynamic switching between default, help, and insult modes
> These work as follows:
--nosys - Bolt will respond without using its default system prompt
--help - Bolt will respond with a system prompt specifically written to be helpful
--insult - Bolt will respond with mocking the fuck out of you (this response is not saved in the memory block for obvious reasons)

Please note that suffixes are lower case, if you mess that up suffix won't trigger

- Fixed the blacklist function
> Now it only responds with the HEHEHEHA if prompted by a blacklisted user and has a 10s cooldown to prevent spam
- Fixed a bunch of memory block issue
> Optimized it to avoid long response times and unintentional repeats
- Minimal revisions to the System prompt to deal with some of the common response problems
- Cleaned up some unused logic relating to AI behaviour
- Fixed and cleaned a bunch more shit for better performance, its too long to add here and I am lazy

Known Problems:

- AI can be a bit to sassy and uses too many emojis in some cases
- Profanity filter is still too sensitive

28.06.2025

- Finally fixed that one really persistent memory bleed bug, it should not repeat entire chatlogs anymore

29.06.2025

- Added warns, unwarns and a punishment History, now Bolt has a full moderation suite!
- fixed a dual prompt issue
- cleaned up some unused logic

30.06.2025

- fixed mod log embeds and general moderation related issues,

01.07.2025

- Added auto role assignment on join,
> You need admin to set it up with /setautorole, Bot needs to have manage roles permissions and needs to be above the role its meant to assign, can be disabled again with /disableautorole

01.07.2025

- Added toggleable and customizable Welcome DM,
- Added toggleable and customizable punishment DM,
> these can be set up with /commands
- Added a levelling system that tracks messages sent and voice activity
> Channel for notifications can be set up with /set_level_channel
> Also added /rank and /leaderboard commands....

- Didn't exist but I couldn't just skip the funny number

11.07.2025

- Added a few custom requests that are only available on certain servers
- Added new --web suffix, fetches web search results and summarizes the search as well
> only limited to 100 searches per month cause of the API that I am using
- Improved handling of the memory block
- Killed a crapton of bugs that have been present in v0.65
- rewrote system prompt, AI is now less sassy, uses way less emojis and has a better personality generally

16.07.2025

- swapped the original model for a larger version, responses are somewhat slower but way better,
- added user based spam protection, spam the Bot too much and it will get tired of you
- Added SDXL API and command for image generation
- Improved error handling and streamlined AI logic
This is all in preparation for Bolt to move to its own Server
Any feedback is greatly appreciated especially now

25.07.2025

- swapped the original model for a larger version, responses are somewhat slower but way better, added Ultra efficiency Mode
This uses a 4B version of Qwen for quicker responses that don't blow up the backend during high traffic
- Added custom System prompt and code for a Server
- Added a few Silly features that have been requested

29.07.2025

- Rewrote the entire Bot to use a central database instead of separate .json files
This improves efficiency by quite some margin and makes the base more expandable
- Got rid of the horrible cog loader
- cleaned up and optimized a lot of the base
- improved moderation features in terms of design, safety and reliability