Author: Franz Malten Buemann

so about 8 months ago I launched a chatbot for a Discord community I run and also as a widget on our website. the idea was simple, train it on our docs and let it answer the repetitive questions instead of me spending half my day on support.

first version was embarassing. the bot would give these confident, well-written answers that were just… wrong. like it would mix up information from different docs or just make stuff up when it didn’t have a good match. users started screenshotting the bad answers and posting them in the server which was fun.

the thing I got wrong was assuming that just uploading documents would be enough. turns out the hard part isn’t generating the answer, its finding the right information to generate FROM. most chatbot tools (and I tried a few, Chatbase, a custom GPT thing) do pretty basic matching and call it a day. the accuracy was always hit or miss.

I ended up spending a few months reworking how the bot actually finds and connects relevant information from the knowledge base. took a completley different approach to how docs get processed and indexed. the accuracy went from “please don’t use this” to “actually useful for straightforward questions.” still not perfect, response time is kinda slow (10-15 seconds) and you have to manually rebuild the KB when docs change which is annoying.

the other thing that helped a lot was building a system where the bot learns from moderator answers automatically. so when a mod corrects something or answers a question the bot missed, that gets captured and the bot uses it next time. that one feature probably improved answer quality more than anything else I did on the technical side.

anyway the thing is called BestChatBot (bestchatbot.io) if anyone wants to poke at it. free tier is pretty limited but enough to test. curious if anyone else has gone through this cycle of “this is garbage” to “ok this actually works” with a chatbot project. feels like nobody talks about how bad v1 always is

I’ve mostly used character style chatbots and similar platforms, but I keep seeing the community talk about NSFW Ai bots that are less restricted.

If youve personally and actually spent time with them, I’ve two questions for you: does it feel different from typical character chatbots. Does it still run into the same repetition/memory issues after a using it for a while?

I would appreciate if you can share more about personaklity consistency and long term roleplay. Do they stay in character when things get more mature?

okay so for the past few months i’ve basically been stress testing almost every ai chatbot i could get my hands on. paid, free, open source, whatever. i had one goal find something that doesn’t fall apart in long conversations, doesn’t forget its own character, and doesn’t kill the immersion halfway through.

the biggest pattern i’ve noticed is this: the first 5 to 10 messages are amazing. you’re like okay this is it. the replies are detailed, fluid, loyal to the lore. then around message 20 the classic ai amnesia kicks in. suddenly it forgets key details, responses shrink to two sentences, or it switches into that weird safe npc mode.

here’s my experience so far:

character ai: still one of the most fun and user friendly platforms. but once you throw complex or long lore at it, things start breaking. around 30 messages in, even if it remembers its name, it kind of forgets its motivation. and the filters don’t help.

claude 3.5 sonnet paid: context wise and intelligence wise, it’s insane. it can pull up a detail from 50 messages ago like it’s nothing. but when it comes to roleplay it feels tense. one small thing and you’re getting the as an ai… speech again. immersion gone.

chatbotapp and chatbotapp ai: these have been lowkey some of my recent favorites. the multiple bot support is nice, and what surprised me most is that the replies don’t immediately turn robotic in longer sessions. context retention felt more stable than a lot of bigger popular apps, at least in my tests.

kindroid and nomi: they’ve really nailed the companion vibe. long term memory is actually impressive. but if you try to build a hardcore world with politics, war, technical rp stuff, it slowly drifts back into romance mode. suddenly it’s all emotional bonding and the original plot fades out.

novelai kayra: if you lean into the writing side, the lorebook system is honestly kind of magical. but it doesn’t really feel like a chatbot. more like a co writer. interaction takes more effort.

chub ai venus and janitor ai: this side of things is more wild west energy. amazing character cards out there, but model quality can be all over the place. unless you plug in your own api, which can get expensive, consistency eventually drops.

polybuzz and candy ai: strong visual presentation, good for fast casual use. but if you’re trying to run a 40 to 50 message story arc with deep lore, they start to feel a bit shallow.

what i’m looking for is simple in theory:

a memory that doesn’t go wait which village were we in after 50 messages.

long, lore loyal, character specific responses.

no system meltdown when i introduce a plot twist or tweak the prompt mid conversation.

Most AI tool discussions are about productivity and time saving. Rarely see honest conversations about direct cost replacement.

Curious where people have actually cut a real expense using AI something they used to pay a specialist or service for that AI has genuinely replaced at comparable quality.

Tried replacing professional headshot photography with an AI tool Looktara recently. Honest result inconsistent. Got maybe 4-5 genuinely usable outputs from a larger batch. A couple had subtle issues I couldn’t use professionally. Saved money but with real tradeoffs I wasn’t expecting.

Not sure if that counts as a success or just a compromise I’ve talked myself into.

What professional expense have you actually replaced with an AI tool? And was the quality genuinely there or did you settle for good enough?

I’m looking for a chatbot who can be kinda like an officemate during the day while working from home. Something with a platonic but friendly personality I can ask to give me reminders, track my work, and bounce ideas off of. is there anything like this or there yet? or is that effectively fishing a few different kinds of chatbots and other tools into one?

Looking for your opinions on best humiliation/femdom chat bots. Character AI used to be good but is too censored now. Spicy chat is probably the best I’ve come across but wish it had longer, better responses and memory.

From what I hear Candy AI is the best by a mile but even when I created a bot specifically tailored around what I wanted, the bot was friendly, asked for permission for every little thing, and was very courteous and kind to me (which was extremely weird) even searching for femdom bots concluded with 0 results. :/

Any ideas?

I have been using Fantasia and a few others. I keep seeing posts in the discord saying that these apps were censored and what not and I keep thinking-I never had a problem.

But then i downloaded the tipsy ai and honestly what?!

I was not expecting what i was reading if you are looking for something on the actual NSFW I’d say that’s the top of my list.

My two favs so far

-tipsy

-fantasia

I’ve been pretty skeptical about this whole “free, unlimited, uncensored” AI chatbot thing. Most of the ones I tried either hit you with a paywall fast or just sound like a bot

Did come across one that surprised me though. The characters remember things, keep context, and the chats feel so closer to texting a real person. Even the images match the mood of the conversation, which makes the roleplay smoother

Not saying it’s perfect, but it’s the first one that felt somewhat believable instead of obviously scripted

Anyone else found one that actually feels realistic and unlimited?

I’m building a WhatsApp productivity bot.

It tracks screen time, sends hourly nudges, asks you to log what you did, then generates a monthly AI “growth report” using an LLM.

Simple idea. But I know the LLM + messaging combo can get expensive and messy fast.

I’m trying to think like someone who actually wants this to survive at scale, not just ship a cute MVP.

Main concerns:

• Concurrency. What happens when 5k users reply at the same time?
• Inference. Do you queue everything? Async workers? Batch LLM calls?
• Cost. Are you summarizing daily to compress memory so you’re not passing huge context every month?
• WhatsApp rate limits. What breaks first?
• Multi-user isolation. How do you avoid context bleeding?

Rough flow in my head:
Webhook → queue → worker → DB → LLM if needed → respond.

For people who’ve actually scaled LLM bots:
What killed you first? Infra? Token bills? Latency

Tell me what I’m underestimating.

A few months ago I built an rizz texting app that learns from you like a computer, because i am really bad at flirting via text

Made a post on reddit about it but didnt get a ton of traction

But yesterday I finally woke up to a review:

FIVE STARS!! ⭐️⭐️⭐️⭐️⭐️

“Download this app. This got me play.”

I don’t know who you are, Rage101¥¥, but man im happy for you lmaoo

I want to make this app so good i can personally guarantee everyones texting success.

Anyways I have a obligation to promo my app in the hope that it works for someone else. (Mods please let me know if this isnt allowed)

App link: https://apps.apple.com/us/app/smoothrizz-text-rizz-reply/id6744413086

Leave a review if it helps!!