Hobold

🔞 18+ only.

Proxy allowed on my bots. Definitions are open. I post my bots on Janitor and on a few other platforms. In case of repost elsewhere a mention would be appreciated.

My platforms and Bluesky can be found here: https://linktr.ee/hobold_kobold
Suggested Custom prompt: https://pastebin.com/TJifXWPc

Shoutout to fellow creators:

@Mawnipulator

@Master501

@DeVoutNumelran

Your average kobold (or dragonborn on occasion).

I make primarily M|4|Any furry characters or simulator ones. My bots put the user inside a given scenario, but do not force a specific gender, species, etc. My bots use around 3000 permanent tokens on average.

All of my bots are limitless for freedom of roleplay. Most of my bots have some 'worldbuilding' behind them, allowing for potentially much longer roleplays, while others are simpler smut bots (and tagged as much).

My SFW bots are fantasy / modern fantasy, or assistant bots meant to help others write their own characters

My NSFW bots explore different fetishes like feet, vore, master/slave, etc. If I use the Dead Dove tag, there's usually a good reason.

Constructive feedback and comments are always appreciated.

With the exception of my Vorecon bots and my more SFW fantasy ones, most of my bots are inspired by images in my Twitter / Bluesky or E621 feed, although I occasionally take requests.

I make bots for my own enjoyment, but I like to share them online to see who else has their fancy tickled. This means that every bot is done with a certain amount of care and that I don't have regular release schedules.

I do make a test run on each bot with the JLLM to make sure they function correctly, but I personally suggest chatting using a Proxy, with an LLM trained on furry datasets.

Current favorites:
1) GLM 4.6: Corpo model with >200B parameters, but naturally uncensored and suspiciously proficient in furry anatomy.
2) Kimi K2 Instruct: 1T parameters corpo model. Needs a good system prompt to work (mine works well enough) or you will get refusals at every answer.
3) Mag-Mell-R1 12B. Part of its dataset is a yiff-extreme collection and one can easily tell that it knows what it is talking about, especially regarding vore.
4) Anthrobomination 70B
3) Deekseek-V3