- There is only one slash command, and that is to bring up the character list!
- Because you only need to say a character's name or answer to their reply
- All characters uses Webhooks! So different avatar for different name~
- Adding character is as easy as making a json and putting them in the characters folder! (refer to character MD file please)
- Uses the all new Microsoft's Florence 2 AI for object recognition
- Uses Llava by default for vibe detection
- Combined, this puppy can almost beat most Open Source Image detection out there! Recognizing the object and getting the vibe and aesthetic at the same time!
- Yes! You can edit your character's response like it's SillyTavern
- Yes! You can delete your character's response!
- No! You cannot regenerate character's response, I should really add that feature
- Uses GBNF Scripting to let the AI understand simple commmand.
- Sample command included in lam.py.
- Use > at the beginning your message (while also mentioning a bot's name) to trigger the LAM
- There's not many feature since it was only implemented a few hours ago...
- Honestly not a feature, more like a neat little trick the AI can do..
- (This is partly another joke about Rabbit R1)
- ((All available commands are written in lam.py))
- (((Grammar Creation still required when adding new commmand)))
- ((((Character need to be told about the command thing in their definition to properly use it))))
- (((((Remind me to add proper intergration, this is still barebones)))))
- Uses some epic prompt enginneering to keep the AI stable (not that epic actually)
- If character is made in the correct format, the AI shouldn't hallucinate much outside of their character
- That's all~
If you want to be cultured, you'll also need to download one of these bad boys~
- Stheno - One of the best Llama 8B Model <- The Most Stable One Yet
- Anjir - Top Performer in Chaiverse Leaderboard <- Ranked High On Chaiverse
- Nyanade - One of the best Llama 7B Model <- 7B Model, Use At Your Own Risk
Requires Koboldcpp for back end, download it here: Koboldcpp
- Microsoft New Visual AI
- Decent, ish
- By default this AI will use both Florence and Llava embedding at the same time
- Florence for Object Recognition and OCR
- Llava for Image Vibe Detection
- Discord API key in .env in the global variable
- make new character json files in the characters folder
- Or just, yknow, make lots and lots of character json file!!! Refer to characters/default.json for example
To run this bot:
- Load the LLM model of your choice in Koboldcpp (Make sure it's marked as not safe for audience)
- Download this repository
OpenKlydeDiscordBot - Make a .env file as written in example.env
- Install the requirements. I suggest using an Anaconda or Miniconda instance.
pip install -r requirements.txt
- Run the bot with
python bot.py
Refer to README.md file in the characters folder.
- Make a better README.
- Add Webhook Support for more seamless Avatar Change
- Fixed the Webhook not detecting reply, (hopefully I don't get rate limited)
- Refactor this whole entire mess...
- Redo the character trigger and swapping system with webhook.
- Implement GBNF for literal fucking Large Action Model feature.
- Add a nicer way to add the Large Action Model Feature
- Cannot Reply In Thread
- Skill Issue on my part, will figure it out later
- Terrible Documentation
- As in like, I did not make any proper Github Commit Comment :v
- Honestly this whole thing started as a joke, if you see the Read Me file, it's all just irony
- I never expected it to work this well
Yeah that's all the current issue. Let me know if there's another one, have fun then~
- Big thank you to @badgids for the fork (even though this has no resemblance to the original)
- Bronya Rand for creating Kisna Kaalana (the default character)
- All the Beta Tester at Ambruk Academy Discord Server
- You <3