GPTDiscord - A robust GPT + DALL-E3 Discord bot, multi-modality (GPT-4V), chat with documents, more!

[winged pollen]

maybe on next major version release

[neon quail]

Everytime I get confused xo, even if I can code myself xo I was like whuuut why does it not work xo

[gleaming summit]

Awesome project! I've had this running since yesterday and so far so good. 🤘
I only ran into an SSL OPENAI API error a few times but I think that may be an issue on my part.
I'm looking forward to future releases! 🥂

[winged pollen]

Thank you!

Errors like the SSL errors should automatically be retried up to 4 times :))

[gleaming summit]

Got it 🫡, thanks! 😎

[polar drum]

can you let us know when you finish it!!? thanks!

[winged pollen]

The README is already detailed enough for how to add it to the server actually, you should be able to do it with the current instructions

[polar drum]

oh cool thanks!

[regal rock]

Discord servers used to be called guilds and internally in the api are still called that but it's understandable to not know about that

[regal rock]

Try installing without the .9

[tired halo]

Ty

[worldly kettle]

hello how do i start your api on discord? I can't find any links @winged pollen

[winged pollen]

What do you mean?

The full instructions are on the README in the repo at https://github.com/Kav-K/GPT3Discord/blob/main/README.md

[tired halo]

this is not an issue with yourbot im sure but if i was making a different help command on a different cog using /menu.. and it worked if i launced using the python3.9 gpt3discord.py but not using the screen gpt3discord. any idea what would cause that? lol

[winged pollen]

They both should work. Try pip install . and then gpt3discord without the screen to see if it gives an error.

Update (v8.5)
Added some new context menu commands (Paraphrase), stability improvements and fixes!

[vale ivy]

sweet :D

I'll borrow some of your ideas this time? :D

[winged pollen]

Absolutely!!!! 🙂

[tired halo]

got er running 🙂 ty 🙂

[young flume]

Anyone know how to skip to the top of this channel? Can only see skip to bottom. There are a lot of comments...

[vale ivy]

#1050326404841082930 message

[winged pollen]

Update (v9.0)

• Added "Summarize" app menu action
• Removed "Paraphrase"
• Lot of bug fixes and stability improvements. conversation redos work much better now
• Settings from /system settings are properly persistent across restarts and etc now

[young tundra]

Hey @winged pollen. Your project is an inspiration. I love your implementation of chat memory. I was trying to figure out how I wanted to do long term memory for my project and had no idea about pinecone. I was just going to use a sqllite db! Keep up the good work.

[winged pollen]

Thank you! 🙂

[spiral venture]

Hello @winged pollen how do i fix this? im pretty new at programming

[young tundra]

Did you install all the requirements with python3.9 -m pip install -r requirements.txt?

[spiral venture]

i just did, its working. thank you so much!

[winged pollen]

Custom indexes coming soon where you can upload a file like a PDF, CSV, text file, or use a web page as a source of context for your GPT3 questions!

[tired halo]

Badass awesome

[urban rock]

Sweet. Just wish i had just a little free tokens with gpt3

[golden void]

So where can I get this

[winged pollen]

https://github.com/Kav-K/GPT3Discord

[smoky maple]

Hi

[brittle bison]

Having an issue with installation. Here is a screenshot of the error from my digitalocean console. Whenever I run "screen gpt3discord" it clears the console for a second then i get the message "[screen is terminating]". when I run python3.9 gpt3discord.py this is the error

[winged pollen]

What OS are you on?

Can you try installing libsqlite3-dev using pip?

[brittle bison]

got

ERROR: No matching distribution found for libsqlite3-dev```
Im using the remote console for the ubuntu server on digital ocean

[winged pollen]

Can you try installing sqlite-devel?

Actually

@brittle bison Run this command

apt-get install libsqlite3-dev

[brittle bison]

oh yea that command worked

should i run screen or python3.9 gpt3discord after this?

ran both and ran into my initial error.

[winged pollen]

Running python3.9 gpt3discord gives you the same error with _sqlite3?

[winged pollen]

After running this command, you need to build python3.9 from source

[brittle bison]

Im having an issue installing python on the droplet. Getting this error after running sudo apt install python3.9

[winged pollen]

Did you add the repo for python3.9 with the add repository command that shouldve been run before?

[brittle bison]

yes I ran sudo add-apt-repository ppa:deadsnakes/ppa
but when I do that this is what I see in console

Hit:2 http://mirrors.digitalocean.com/ubuntu kinetic-updates InRelease                                          
Hit:3 http://mirrors.digitalocean.com/ubuntu kinetic-backports InRelease                                        
Hit:4 https://repos-droplet.digitalocean.com/apt/droplet-agent main InRelease                                   
Hit:5 http://security.ubuntu.com/ubuntu kinetic-security InRelease                                              
Ign:6 https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu kinetic InRelease   
Err:7 https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu kinetic Release
  404  Not Found [IP: 185.125.190.52 443]
Reading package lists... Done                              
E: The repository 'https://ppa.launchpadcontent.net/deadsnakes/ppa/ubuntu kinetic Release' does not have a Release file.
N: Updating from such a repository can't be done securely, and is therefore disabled by default.
N: See apt-secure(8) manpage for repository creation and user configuration details.```

I guess im just tracing back the errors

I get this same problem when running sudo apt-get update

[winged pollen]

I think this is because you’re on Ubuntu 22

Those installation instructions are for Ubuntu 20, you can try googling the u Ubuntu 22 python 3.9 installation instructions, installing it that way, and then continuing with the rest of the installation

[brittle bison]

I see, I'll just create a fresh droplet with 20

[brittle bison]

Awesome, got it working. Thanks so much for the help.

[brittle bison]

Kaveen, when you mentioned you are coming out with custom indexes soon, will this have the ability to upload multiple files? Or will the bot only be able to reference one? For example, could you periodically upload pdfs to add to the source of context?

[winged pollen]

Multiple files! 🙂

And will still allow for querying any individual file/url too

It’ll be like /index query query:(prompt) index:(‘ALL’ or the name of a specific file or url that was indexed earlier

[brittle bison]

thats great, looking forward to that feature

will be very helpful to my goals with your bot 🙂

[tired halo]

me 2 ive been checking up constantly waiting for it to drop lol. love this bot

[winged pollen]

a single file/url version is actually available now, just git checkout the gpt-index branch on git

[brittle bison]

what is the hurdle to make it for multiple files? i've been interested in the embedding end point and the way i understand it it just translates the text into vectors that you configured to store into pinecone. So if you have functionality for individual uploads can't you just repeat that with multiple files and they will all be stored?

[winged pollen]

the hurdle is the context limit

it's really easy to index and embed multiple files

but when searching across multiple files, and searching across tons of embeddings, you need a good response synthesis system

And also there are other restrictions because of the context limit, for example, if you upload a huge document you can't just be like "summarize" this document

cuz it could only read in 4096 tokens at a time

so again there needs to be a good response synthesis system for cases like that, for example something like recursively summarizing max context chunks of the text, extracting relevant embeddings, repeat, etc

[winged pollen]

also currently it's not configured to be stored with pinecone, i'm still figuring out an effective way to do that with the library we're using called gpt-index for context management

[brittle bison]

That makes sense. It seems like no matter what, if you have many embeddings you are going to get watered down answers. Like if you had all of shakespears works embedded and wanted to reference every time there was a murder it probably would be too much to sift through and you would get an incomplete answer.

[winged pollen]

Exactly, there are ways to counteract this but it becomes prohibitively expensive

[brittle bison]

yea can only run all the information through so many different davinci calls to summarize before you end up paying a dollar per question

im about to enable pinecone on the bot and test its ability

how do you restart the bot from the screen gpt3discord console? i added the pinecone token and want the changes to take effect

or do i even need to restart it

[brittle bison]

figured that out. I am testing it and not seeing the results, I explained the geography of a made up place and then ended the conversation, created a new conversation and asked it to tell me about the geography of the made up place. It made up something else. I see vectors are being stored in my pinecone so not sure whats happening

ah I see, it does remember. But its hallucinating more than it is referencing the information it has.

it will drop a small detail here and there, maybe it needs more information or?

[brittle bison]

Kaveen can you explain how the bot utilizes embeddings to try and pull up relevant info and feed it in the prompt?

[winged pollen]

Can you DM me examples of this issue?

[winged pollen]

Yeah! So we embed each user message and bot response, when a user types a new message, that message is embedded too and it's compared against the previous embeddings to find the most relevant other pieces of conversation to what the user just typed. Then, we save a certain amount of those embedding-retrieved prompts along with a set-length sliding window of messages for every prompt

[brittle bison]

how does it judge the relevancy? I was reading a bit and saw that it can be by keyword or semantic search. What does your bot utilize?

[winged pollen]

semantic search 🙂

[jovial kraken]

is anyone getting this error

[urban rock]

AI Dungeon did a lot of the Middleman model stuff, searching , summarising, querying etc by using an intermediate GPT2 finetuned

Which can usually even be run on local computer.. so if you figure out a way to use GPT2 to do the lifting when it comes to saving tokens, it might work out. AI dungeon was doing amazing stuff with gpt3 even the betas for the playground became accessible over a year go

I mean GPT2 + GPT3

gpt2 for handling inputs and parsing them, summarizing, context, into a neat package, handing it over to GPT3 and then returning the response as is

[brittle bison]

Kaveen is there an easy way to take user fine tuned models and have them replace the davinci 003 model used in conversation

[winged pollen]

Yeah! You can replace instances of 'davinci-003' in the code

[brittle bison]

I did in openai_model.py

DAVINICI = "my model"

didnt seem to work after i restarted

[winged pollen]

and other locations as well, ctrl+f the whole project for text-davinci-003

[brittle bison]

i see

ok thanks

[winged pollen]

Major Update
We've added custom indexing! You can use websites, PDF documents, text documents, CSVs, powerpoints, images, and even youtube videos as sources of information and context for the bot! Everything is done magically behind the scenes. You can also save multiple indexes and compose multiple indexes together into one index to work with. There is also a "Deep Compose" feature which takes more time and tokens and does a better job at this. Beware, these features are expensive and take longer to run, but are very cool!

It will be available on release v10.0.0 very soon. Please read the README for more information on what changed, look at the "Custom Indexes" section. https://github.com/Kav-K/GPT3Discord

[brittle bison]

Amazing!!

how expensive was that video summary

[winged pollen]

Video summaries are pretty cheap! That video was like <6000 tokens, so less than $0.1 roughly

I tried a 168 page lab manual and it worked excellently with Deep Composition, but it cost a lot, over a dollar for the creation of the index, the queries were relatively cheap though

We also have composition of indexes:

[brittle bison]

that is crazy. are there pretty condensed queries? can the bot answer specific information after the summary? like do a search in the document

condensed summaries*

[winged pollen]

Yes, it can search in the full, entire document

The Deep Composition feature is complex, it will embed and save chunks, and then summarize those chunks, and then use multiple levels of refining and fine tuning a response before it outputs

It'll do something like, search through summary -> go deeper a level in the tree structure and search in more detailed chunks -> etc

The basic, non-deep composition feature is excellent as well, it's just a large vector store that is indexed using gpt-index

[brittle bison]

i need that in my life for sure

btw could only fine these two things for text-davinci-003

[winged pollen]

yeah that should be all

[brittle bison]

sweet

im a novice and dont understand the server hosting, is there a better way to restart for these changes to take effect other than turning it off and on?

[brittle bison]

I want to see this if you could demo it sometime. To get an idea of the capabilities with such a large text.

[winged pollen]

Like turning the bot on and off? No, you always have to turn the bot on and off 🙂

[winged pollen]

Just linking back to the major update message real quick: #1050326404841082930 message

[winged pollen]

Added AI-assisted google search functionality on the main branch for those of you that know how to get a google custom search api key and engine id, I'll polish it up and add detailed instructions soon but it works now as an experimental release

[winged pollen]

Major Update
AI-Assisted Google search is here! /search

[tired halo]

stoked on this. just when i think im already excited on what i know to be coming in the next update.. you guyz throw even more good stuff i dig. thanks guyz you rock 🙂

got this atempting a serach tho. doi need to wait for next update to add the api tokens in the .env file:?

[winged pollen]

You need to add API keys, there's more info in the README

and also make sure you re-install the project, there were a lot of new dependencies added

[tired halo]

aight il head over into it

that i did 🙂

tried an update and seen no indexes so i wiped and re installed 🙂

thanks again eh bro.. this update.. above and beyond.. i didnt even see .search coming with them lol

[winged pollen]

No worries!! Everything works fine for you?

[tired halo]

yep.. just got /search working aswell. i still need to go thru and test how well. eg indexing questions and what not.. and making a few indexes but over all everything seems to be functioning

🙂

adding discord channel index btw... great idea for sure.. i dig that

[winged pollen]

yeah, and you can index your ENTIRE discord server too

(expensive tho!!)

[tired halo]

about all i can see missing now is voice chat with tts stt 😛 and perhaps a wolfram 😛

hellz yea.. tho to be fair.. its not massivly expencive tho. just in relation to other prompting

[winged pollen]

yeah

[tired halo]

but a buck or 2 for indexing a large quantity of data for referance.. really not to bad

[winged pollen]

yeah I agree

I indexed a 200 page lab manual, it cost $2-3 for indexing, but it was able to do it and was pretty effective!

Queries on large indexes take a long time though, its really dependent on your computer, but I think the time cost is fair, especially for docs with hundreds of pages that you can have AI-assisted answering from

[tired halo]

nice 🙂 lol indeed you mean on the host system for the bot?

[winged pollen]

Yeah on the host system of the bot

[tired halo]

technically its not my computer lol

😛

[winged pollen]

We do OCR now too, if you upload an image, it will index the text in the image

so stuff like that is dependent on the host system

[tired halo]

but thats good to know. if it gets slow feeling il just upgrade the host system on do

[winged pollen]

yeah

[tired halo]

man.... i love this bot

lol

im prety interested in playing with the youtube video indexing into text.. thats a new alley of usage for me and im stoked on its implamentation

[winged pollen]

Yeah! It'll work for any video that has transcription enabled or auto-subtitles, or has audio that is reasonably transcribable into english text

If the video is in another language, it'll even auto-translate it into English

[tired halo]

ok nice

does it use whisper at all for voice analasys

or another service?

[winged pollen]

No, it uses a youtube api to do it

no transcription processing on-host

[tired halo]

ahh ok so its pulling pre transcirbed text

[winged pollen]

only OCR processing is on-host

yeah

[tired halo]

ok thats still awesome

🙂

[winged pollen]

Whisper can be added but I feel like it'll be problematic for some peeople with less powerful hosts

[tired halo]

i really want to see something like this doing voice analasys and chatting in the discord voice chat one day lol

[winged pollen]

but tbh, OCR Is high cost too

[winged pollen]

Soon!!!

[tired halo]

super awesome 🙂

[tired halo]

i have indexed the pdf for voicemeter potato.. and querrying the index is returning me proper information relivant to my querry.. SUPER AWESOME GUYZ mad love

super digging it.. its extra awesome because with a /search it does not return ideal results relitave to what im looking for in a reply on this task. and google never was ideal at doing that without sending me to a page to dig thru that info... but indexing... this seems to work super well.

[brittle bison]

How do we know when the deep compose index is created? I composed it from four different indexes and the message saying it was composing disappeared, not sure how to access it

[tired halo]

general question aswell .. kind of an idea is it possible to have an index conversation so we dont need to /querry every time

[brittle bison]

yea i would love the normal conversation functionality but it draws from the indexes for context rather than embeddings

[tired halo]

i had an issue with the index being created from web link

i downloaded pdf that i was trying to work with and it flew thru it and indexed properly

/index converse 😛 cant wait

[brittle bison]

it basically could use the existing functionality, running the index query, then returning it to the GPT prompt saying "this is the information that could relate to the users question, answer using this new info" or whatever

Also, what exactly is the difference between a normal index and a deep composed one?

[winged pollen]

You can check /index load to see if a file named composed_deep something is there

if so, it's that one and u can load that one

[winged pollen]

Deep composed indexes structure chunks of embeddings into tree structures and summarize each subsection of the tree and has multiple refinining phases

[brittle bison]

I've tried it a few times and no file is popping up

I've indexed a bunch of files though this is working really well its so cool!

[brittle bison]

is there an index folder where everything is stored including composed?

[winged pollen]

Yeah, its in the /indexes folder

[brittle bison]

How much do we think a 7300 page document is gonna cost to index, and per query on that?

go big or go home is the motto

[regal rock]

Per query the same as any other document unless you increase the nodes in the query command. To index it'll be a lot though lmao

[brittle bison]

based on what Kaveen said about it costing a little over a dollar for a 168 page manual, im guessing its around $1 per 100 pages, so $73?

[digital kernel]

hello

@winged pollen hello

[winged pollen]

i wouldn't reccomend trying 7300 pages unless you have a really strong server and lots of RAM

it would suck for it to run out of memory or crash while indexing then all that token usage is to waste

[brittle bison]

Good point. I have one around 2000 pages, think thatll be alright with the droplet you recommend in your readme?

[celest sedge]

Pages don't really matter, Tokens do.

So if those Pages are like typical Scientific Papers, ~750 Tokens per Page

More like $110.

still kinda cheap vs letting an employee read and understand the document 😉

[compact sentinel]

do you have the bot that I can try, I really like the permanent memory chats idea because official chatgpt now forgets everything

[urban rock]

Now for yandex search 😛

[hollow sky]

Wait does it have Dall e2 image generation?

[winged pollen]

You can install it and run it yourself 🙂

[winged pollen]

Yes

[hollow sky]

Can I run the bot on replit?

[winged pollen]

No

[bold flume]

I have a similar project like this when GPT-3 came out. But it does not retain old chats like yours 🙂
https://github.com/KeirLoire/discord-chatbot

[winged pollen]

Nice!

[slate dove]

How do I use this?

[tired halo]

Usage and install directions are in the Readme on github. Kaveen has shared the link a few times. Scroll up till ya see it. Hit the link and scroll down for directions

I'm not a developer but I've set it up a few times. It's not to hard tho it's not basic like clicking an invite link:p still if I can do it I'm sure you can to

[winged pollen]

https://github.com/Kav-K/GPT3Discord

[slate dove]

Thx

[sleek pendant]

PS C:\Users\Ryan\Desktop\GPT3Discord-main> python3 gpt3discord.py
Traceback (most recent call last):
File "C:\Users\Ryan\Desktop\GPT3Discord-main\gpt3discord.py", line 10, in <module>
import discord
ModuleNotFoundError: No module named 'discord'
PS C:\Users\Ryan\Desktop\GPT3Discord-main>

anyone know why i get this error? just exported the files from github, put my keys in then tried running the python script.

[winged pollen]

Did you install dependencies?

[tired halo]

go thru the readme under the files untill you see the install section read from here down. install is just below where this is. omit the git command and just use terminal inside of your install folder and follow the rest of the directions <if im not mistaken anycase> kaveen and haru are the top experts ive spoken to in here on this 😛 but its what i would do if i was trying a local install

[winged pollen]

Conversations and buttons will persist across bot restarts soon!

[brittle bison]

about to attempt a 2000 page index

wish me luck

worked no issue

kaveen how do you create a server index? so other people can use it?

and is there a way to transfer an existing user index to a server index

[regal rock]

Server indexes are just saved in a folder in /indexes that's mapped to the server id instead of the user id. When doing the load command you can either get an index that you own, or an index that the server you're running the command from has

So you can manually move a .json file from a user to a server which would make it available to everyone

@brittle bison

I originally added the server option so that when doing a full discord backup it'd be available for everyone since it's kinda lengthy and expensive if everyone were to do it

[brittle bison]

inside the indexes folder I only see a folder for my user ID. not one for the server

[regal rock]

It's only made when doing discord backups, so just make a folder yourself named after the server id

[cunning phoenix]

🫵🫵🫵

[brittle bison]

Ok sweet, thanks

[patent wren]

Im not nearly as smart as the people in this chat but Ill try adding it to my server because it looks amazing

It looks so well made

[winged pollen]

Thank you! We can help with set-up!

[midnight summit]

Hi Kav 👋🏻
I'm not that much professional in Python, I am JS dev but from your GItHub I see you are JS dev too can you give me little help with #1074064913540071504 👀

[winged pollen]

Hey unfortunately I don't really have any experience with discord bots in JS

[midnight summit]

I have that experience my question is about stuff which not related to Discord XD

[winged pollen]

Sure, let's move the conversation into DMs 🙂

[midnight summit]

yeah I was going to say that cause we are going off topic here XD

[winged pollen]

Google Search functionality!
/search is available to use to search google and have GPT3 synthesize an answer for you

[winged pollen]

Redo functionality for search added, alongside a bunch of other bugfixes for stability and document compositions

[patent wren]

This bot looks amazing but I'm absolutely stumped.... Do you have to download things to get it in your server? I was in the support server and I scrolled through the messages and could not make sense of anything.... I'm an artist, all this technical stuff goes right over my head

[winged pollen]

Yeah, you need to install it yourself, unfortunately it's not a bot you can just invite to the server

[tired halo]

It can feel a bit advanced but the Readme they have on the git hub under the files is quite thorough. And once you get the hang of working with it it gets alot easier. If I can get it.. I believe in your ability to also get er done

[patent wren]

Ill look over the readme one more time

Thanks

[tired halo]

Not a worry bud. Any questions don't hesitate to ask. Everybody in here is helpfull. And this bot is worth the learning curve. It's one of the most feature rich and least restricted gpt3 bots I've used

[regal relic]

Hi, is there a way to run this on arm? ERROR: no matching manifest for linux/arm64/v8 in the manifest list entries

[small flame]

hey @winged pollen thanks for your work. ChatGpt has a limitation on the size of data (copyandpaste csv, text, ) for analysis that it can intake. does your bot have that limitation?

[winged pollen]

No limitation

[vale ivy]

Does it split the data into parts I guess?

[small flame]

how do i let it analyze a csv or a text file?

[winged pollen]

Yeah it chunks and embeds into a tree structure and then summarizes tree nodes with children

then uses both the summary and the nodes to synthesize an answer

[winged pollen]

use /index add

[winged pollen]

Major Updates

• Ability for every user to start multiple conversation threads! This was highly requested
• You can just tag your bot like @winged pollen and the bot will kick off a GPT request for the text contained in the message that you tagged the bot in, this is good for quick prompting
• /search now allows you to conversationally send Follow-ups! After searching for something, you can keep the session active and keep asking follow up questions! This is like BingGPT, but better and more accurate lol!
• Indexes created by /search are now reusable and loadable with /index load
• Performance improvements and bugfixes
• made README look much, much better.
  https://github.com/Kav-K/GPT3Discord

[vale ivy]

You're on fire!!

[patent wren]

Frrrr

Y'all are doing amazing

Read the Read Me and even though it's easy to follow thankfully I got lost at the python and .env stuff 😂 oh well, I guess I'll stick to seeing the updates you all make with this

You guys are making things easier for people

[winged pollen]

I'd be happy to help you set it up!

[patent wren]

@winged pollen how

[honest echo]

Wondering if anyone did something similar with slack

[winged pollen]

DM me the issues you're having with the setup

[winged pollen]

Updates

• Optional pre-moderation for things sent to /gpt ask, converse, /search, /index, etc
• Optional language detection to force people to use English (very rough currently)

[tired halo]

Any chance I could request an update to the discord channel index function in the sense of a setting to add a list of channels for it to watch to real-time update the discord index as chats happen potentially with a conversational channel to talk about the conversations had or posts scraped into various channels without having to use the /index querry

Would be handy to index tutorial channels and the like and have it able to converse about how to use server functions and talk about chats had

Or does it already watch channel after it's indexed? To continue updating the index

[vale ivy]

I think that would be a cool feature, I will definitely include it in my own bot

[winged pollen]

Mark it as a github issue on the repo and we'll definitely roadmap it! 🙂

Major Update V10.7.0 (will be released within the hour)

• Lots of bug fixes and performance improvements, indexing and search is now up to 3x faster in a lot of cases.
• Deep compositions are faster and more detailed now
• Added more parameters to /index to control how much analysis gpt does before answering
• Added a command to rename indexes

[winged pollen]

Update V10.8.0

• Lots of beautifying, bot responses look much better now
• Token safeguards, before a deep search and before a deep compose, it will stop itself if the predicted usage price is too much
• Added ShareGPT integration! Share your conversations super easily

[winged pollen]

Bugfixes

• V10.8.2 fixes some conversation bugs, making conversations more cohesive
• Added typing indicator 🙂

https://github.com/Kav-K/GPT3Discord

[tired halo]

i indexed 2 pdfs but still ahve no /index folder :/ should i do a clean install? ive been updating for about the last 3 update versions i think

outside of that. im loving the updates and all the good work on this 🙂

[winged pollen]

Do the indexes show up though? when you do /index load

[tired halo]

they are there when i querry

they load fine

[winged pollen]

Hmm, try refreshing your file browser

if they show up on a load they're definitely saved somewhere

[winged pollen]

Upcoming Feature

• Google connected chatting like Bing Chat, coming soon!

[lethal prism]

Hey

Can i know how to make the bot online 24/7

@winged pollen please help me

[tired halo]

Using the screen discordgpt command it should Stay running when the window closes. Screen runs it in background after window closes

Then if you go back to terminal there's another screen command to re attach to session

[lethal prism]

I use vs code and windows 10 so in that does it work?

[tired halo]

I'm using vs code in that end it could come down to permissions

When it comes to this command in particular I usually use the console from digital ocean

It works every time there and auto logs you into root terminal so I usually cs code for editing and file structure work

Updates and installs I use the digital Ocean droplet console

[lethal prism]

Broo can you please make a short video on this plss it would be very helpful

[tired halo]

I'm at work atm but I'm sure it could get done

Open digital Ocean click on your droplet

Then click the ip adress area and there should.be a console button top right

Click that open

There may be a command it will tell you about to update the droplet console put the command in and execute it but then just cd GPT3Discord

It's case sensitive. Cd GPT3Discord

Rm bot.pid <-- clears out the last session basically so you can re launch

Screen gpt3discord all lower case to launch

[lethal prism]

Ahh wait

[winged pollen]

Hey you can use DigitalOcean to do this, there are very detailed instructions with a link for $100 in free credit in the README

[tired halo]

Yea oooh wait your trying local eh?

[lethal prism]

Local in the sense?

[tired halo]

Your home PC vs hosting it on an external host like digital ocean

[lethal prism]

Btw dm you discord server link

[lethal prism]

🙃i just want that bot for my server thats it but i need for free so i am using my pc from many days as i am unable to keeo the bot online 24/7

[lethal prism]

Okay tq

[tired halo]

Digital Ocean actually works pretty wrll

And there's a free 100 dollar credit

For how often home net can drop and such. And the bandeith potential in use ide hit that free usage up

[lethal prism]

Pc in the sense keeping vs code open so that the bot stays online

[tired halo]

Yea bro trust me you should try the hosting with digital ocean

[rare lion]

@tired halo

RAMBONER's message blocked by AutoMod

We are committed to fostering a positive atmosphere and have implemented additional filters to block any inappropriate or disruptive language and topics.

Reminder: Refrain from bypassing the AutoMod filter as per server Rule #10.

[lethal prism]

Ohh okay but till what extent is that free

@tired halo dm the link of your discord server

[winged pollen]

$100 for 2 months 🙂

[lethal prism]

🙃and rest i have to pay?

[winged pollen]

Yup, but there are other free services you can switch to after the free credits expire on digitalocean

[lethal prism]

Ahh okay

But how can i connect my vs code to digital ocean

Or can i code in digital ocean only ?

[winged pollen]

You have to upload the files of the bot to digitalocean and then start it through the server's terminal

VSCode has some extensions that allow you to live-edit files in DigitalOcean

but you can just upload the files after you change them with a program like FileZilla

[tired halo]

You can also log into the terminal on vs code using an ssh plug in root@(serverip)

Then itl prompt for password and give you ssh file and terminal. But I only really ise that method if I'm editing files so I'm not on a console doing it

But yea. Digital oceans terminal is ideal use there. And it's been ez to work with

[lethal prism]

Guys guys i am confused if you guys know some YouTube video which explains this please send me the link it would be helpful

And thank you very much @winged pollen and @tired halo for you precious time 😃

[winged pollen]

https://github.com/Kav-K/GPT3Discord/blob/main/detailed_guides/DROPLET-GUIDE.md There is a detailed droplet guide here, and then after that, there are detailed installation instructions here: https://github.com/Kav-K/GPT3Discord/blob/main/detailed_guides/INSTALLATION.md

[tired halo]

Lol sorry the read me has good info if you haven't had luck by the time I finish work and get home hop on a videochat with ya and walk ya thru anything your having issues with

The Readme kav shared tho is pretty top notch. I was confused the first time but went thru step by step and it worked well for me in the end

[tired halo]

Kav does screen command work on a local install aswell?

[winged pollen]

If your local install is a Ubuntu or a Mac, then screen works locally

There are alternatives to screen for windows too tho

like u can use PM2 I think even tho it's for nodejs to run python apps

[tired halo]

OK cool cool 🙂 that's where his problem is then he on windows 10

[lethal prism]

Thank you

[tired halo]

You got this bro. I believe in you

:p

[lethal prism]

Thanks man i never met such a helpful ppl online thank you

[lethal prism]

🙃its already 1 am here so i gotta go and sleep rn

Sorry for that

[tired halo]

Tty soon let me know how it goes this bot rocks

[winged pollen]

Thanks ❤️

[velvet owl]

I Like to have a schematic of a Person in front of an Airbag in a car

[brittle bison]

Wouldnt we all

[winged pollen]

Upgrading to ChatGPT APIs!

[tired halo]

Stoked to see these released and happy I can select between now lol question is.. is it neutered like the website

I've assumed it's the starter script on the site that has done this...I sooooo hope we get an Un neutered api from them lol plz god

[brittle bison]

Excited to see it

[winged pollen]

main branch has ChatGPT integrated

will do a full release soon after integrating it into gpt-index and etc as well

[tired halo]

Bro youguyz rock....like so hard... so happy i found this project and you fine people

[tired halo]

Does the test discord have an announcement channel? I would love to have a space to follow updates in my server announcement channel like I have with the other bots :p

Nvm lol I found the link in and I'm following thr announce channel now :p

[winged pollen]

Major Update 10.9.0

• ChatGPT model support has been added and it can now be selected as the default model for the bot. You can select it with /system settings and change the model value to one of the models starting with gpt3.5-.
• We have switched to using the ChatGPT model for /search and /index, with the exception of deep search and deep indexing, we are waiting on some changes from the gpt-index library in order to continue with this. Once those are completed, I will do a new release with those changes.
• Currently, price reporting for deep searches are disabled and it will not stop you if you try to deep search a ridiculous amount of usage, so be careful, but it should be okay for the most part. This will be fixed when the gpt-index changes are pushed up and we can switch deep indexing to using the chatgpt models.
• You can use the new model as a default for a conversation with the /gpt converse model parameter, a default for the whole bot, or a default for /gpt ask.
• The two new models are called gpt-3.5-turbo and gpt-3.5-turbo-0301
• Deep searches are fixed to be properly more informational than regular searches and are faster and cheaper
• Indexing, and searching perform much better with this new model, and will be faster and much cheaper
  https://github.com/Kav-K/GPT3Discord

[winged pollen]

Please let me know what you think of the new model's behaviour, do you think it's more effective at /search and /index features? Does it behave and remember better than text-davinci-003 for /gpt converse? Would love to hear thoughts!

[winged pollen]

Hooked into the new audio API that released as well! /transcribe (v10.9.1), this is a bit rough but will improve over time!

[winged pollen]

We hit a thousand github stars! Thanks so much everybody for your support of this project!

In other news, release 10.9.11 was sent out earlier today with a lot of bugfixes

[tired halo]

Excellent. You guy, deserve every star. Thank YOU GUYZ. We didn't do it we just followed the project and downloaded.. the thanks. Should be all for youguyz 🙂 a result to your dedication to this fine bot 🙂

[obtuse summit]

there's a problem when the bot_taggable is on.
anyone even those who doesn't have gpt roles can use and tag the bot and receive a response from it, just like doing /gpt ask command but without role restriction. is it a bug or part of the feature?

[winged pollen]

That would be a bug! Can you make a github issue? We'll address it asap

[obtuse summit]

done!

[winged pollen]

Bugfixes 10.9.12

• Fix prompt optimizer
• Fix too long conversation messages failing to send on discord
• Fixed bot taggable bypass
• Added back token safeguard for deep search
  cc: @obtuse summit

[analog socket]

Absolutely ❤️ it! 🥂🥂

[winged pollen]

Late reply, but thank you!!!

v10.9.15 has some minor bug fixes, have been inactive recently due to being busy

[analog socket]

Most welcome Kaveen 🥂🥂

[exotic rapids]

ok

[nova heron]

I would love to learn more from this project. Being; all in one

How is permanent conversation memory achieved? Any tutorial? Starting point? Good prompt?

[winged pollen]

We use pinecone to store embeddings for user messages and bot responses, and then we find the most similar embeddings every time a user chats, and we use that to dynamically construct a prompt

Major Update v11.0.0

• Conversations now persist through restarts. You will no longer lose conversation history if you restart the bot!
• Conversation data is stored in a local pickles folder, currently, a lot of data is stored, so if you find your bot getting too slow, you may have to delete this folder, a cleaner solution will be implemented in the future
• Conversations auto-save roughly every 20 seconds.

https://github.com/Kav-K/GPT3Discord

[winged pollen]

GPT-4 Support v11.0.3

• If you have been granted GPT-4 access from the API waitlist, you can use it with this bot

[brittle bison]

One day

[hollow patio]

Very nice project.

[winged pollen]

Updates & Bugfixes v11.1.4

• Channel-based conversations, designate an entire text channel as a GPT conversation
• GPT-4 conversations are more stable and reliable now (a temp workaround)
• Fix GPT-4 pricing
• Lots of bugfixes surrounding gpt-4, 3.5-turbo

[muted arch]

Does this bot currently has support for summarizing a pdf file? Can chatgpt receive the file as input?

[still stream]

This project has features you asked: https://discord.com/channels/974519864045756446/1086269524266979328

[winged pollen]

Hi @still stream please don't link to other projects in different project posts.

but yes, you can check out that project for summarizing a PDF file! However, this bot also indeeds support summarizing PDFs, and for free (you pay for your own API usage) and without any "subscription" stuff

[still stream]

Apologies for linking it, i wanted to help students and developers who can take benifits

Thanks i will keep it in mind

[muted arch]

Thanks for your help and these developments.

[eternal stratus]

Thanks for your help

[rustic estuary]

@winged pollen I finally have this up and running... a pretty huge accomplishment for someone with limited programming skills. Great work on the guide and thank you for linking the walkthrough video. Only thing I'm struggling with is the Pinecone link as persistent memory doesn't seem to work. Maybe I misunderstand the intent?

[tired halo]

They definatley did a great job on the Readme and the bot 🙂

[winged pollen]

Development has been slow this week, apologies for delays on updates and bug fixes. Progress and improvements will ramp up steadily starting from early next week, some planned features that should be completed by the end of next week are an index-based chat, a search-connected chat, switching to using GPT-4 for indexing and searching, will also clear out many of the bugs on github issues, fix inconsistencies with conversations, and improve our conversation context logic

[vale ivy]

Will the GPT4 for indexing/searching be optional?

Also you're doing awesome things, please don't burn out 🙏 take your time :)

[winged pollen]

It will be! Yeah! And thank you ❤️

[winged pollen]

Updates V11.2.0

• GPT4 Support for /search and /index - You can now use GPT4 to do your searching and indexing tasks! Be careful, GPT4 is expensive
• Multi-step query decompositions - When doing a query with /search or /index, the bot will break down complicated questions into smaller questions and progressively build up an answer, use the multistep option in /search and /index query to try this out, this will take longer!
• Optimizer was added to /search and /index to remove unneccessary sentences, this will reduce token usage and possibly also increase quality of answers
• Deep Search functionality improved, deep search no longer builds a knowledge graph but a QASummaryGraph: https://github.com/jerryjliu/llama_index/blob/main/examples/composable_indices/QASummaryGraph.ipynb. Deep searches are still slow though. I'd recommend using multistep queries with regular searches
• More graceful timeout errors
• OpenAI organization being sent in a request is toggleable with /system settings, organization coverage is more cohesive now

/search chat and /index chat are still coming soon! I've just been busy. As usual, if you liked the project, please give the project a star over at https://github.com/Kav-K/GPT3Discord

[winged pollen]

GPTDiscord - A robust GPT + DALL-E2 Discord bot, permanent memory chats, AI moderation, more!

Updates V11.2.5

• Send text files as input to /gpt ask
• Set the pre-instruction per-user or per-channel (so, you can make /gpt ask and etc behave differently)
• Openers now persist indefinitely in pinecone conversations, previously they were lost after a dozen or so messages
• Fixed up some more package bumps

[brittle bison]

Sweet, was just thinking about a per channel instruction. Does that work with openers per channel for gpt converse as well?

[winged pollen]

It works on top of the openers yeah!

Updates V11.3.0

• We have released our internet-connected chat! You can now chat with GPT3.5/4 and have the conversation be connected to the internet. You need your google API tokens to use this, it also supports wolfram and using GPT-4! /internet chat (https://github.com/Kav-K/GPT3Discord/blob/main/detailed_guides/INTERNET-CONNECTED-CHAT.md)
• You can adjust how many top-level links GPT searches for/uses whenever it needs to use the internet
• We have renamed /search to /internet search
• The internet connected chat functionalities add some new dependencies, please reinstall the requirements.txt before trying out this new feature.

As always, please give the project a star and fork over at https://github.com/Kav-K/GPT3Discord if you've liked it!

[vale ivy]

🔥🔥🔥

[winged pollen]

V11.3.2
A bunch of bug fixes!

[winged pollen]

**Updates **V11.4.0

• Our /internet chat functionality now allows the bot to access links that you provide it in the chat! It will go to the link and look at the content to answer your questions. NOTE: This doesn't support links that are PDFs, google docs, etc, nothing fancy yet
• Bugfixes for conversations

[winged pollen]

V11.4.2

• Show used tools in a connected conversation
• Bugfixes
• Updated py-cord to the latest

[lyric plume]

Hey. First i wanted to say that your Bot is awesome. I tried a few GPT Bots but yours is by far the best.
I have a small question regarding Wolfram. I made an acc over at wolfram and generated the api key and added it to the .env. But im not sure if it works correctly. I took your example from the screenshot to test it. When the Bot uses Wolfram is it always like in your screenshot with Assumption/Answer? Cause thats not how it is displayed in my chat. And when i look in the api menue on wolfram it says 0 queries.

[winged pollen]

Thank you! So now with the latest update, if it uses wolfram it'll tell you in the response message that it uses wolfram

11,4,2

[lyric plume]

Thanks for the help. I figured it out. I have another Question im not sure about. So if i understood correctly when you create a sub Thread for internet or conversation, its basically like when you open a new chat tab in GPT. What i mean is every sub Thread has its own memory right? The Reason why i ask is because when i take the bot offline and restart it, the Bot doesnt answer in the open Thread anymore. I have to start a new internet sub thread and then it works. But now i lost all the memory from the previous conversation. Is this working as intended or do i have some kind of Problem i can look after? Btw im using pinecone and the log says it connected properly.

[lyric plume]

Just noticed that the normal conversations still get saved in the pickles, despite having pinecone. Log says this when i start the Bot: "Got the pinecone service". So pinecone should be configured correctly right?

[winged pollen]

The internet chat doesn't use pinecone at all

and it doesn't save during restarts

only regular conversations save during restarts

[brittle bison]

hey Kaveen, can you update your installation instructions to reflect the new repo name GPTDiscord not GPT3Discord. I accidently used the old one and only have a README file haha

[winged pollen]

Oops! Totally forgot about this!!!

I'll do that tonight very shortly, thanks 🙂

[winged pollen]

Updates V11.4.5

• Initial basic multi modality image understanding support!

[vale ivy]

e p i c

[winged pollen]

Updates v11.5.0

• Improved conversation image understanding, the bot will now also do OCR on images that you extract (this uses Google Cloud Vision API) https://github.com/Kav-K/GPTDiscord/blob/main/detailed_guides/MULTI-MODALITY.md
• Bugfixes

[brittle bison]

Wow very impressive with the translations

[winged pollen]

Thanks! A bunch of credits to @vale ivy for guiding me towards how to implement this!

[lyric plume]

Thats really cool. 😀 So with the newest Version i have to switch to Google Cloud Api to use the image recognition? Or can i still use Blip too?

[winged pollen]

It uses both! 🙂

If one isn't registered, the other one will be used

[proper bough]

hey I am having trouble the GPT4 discord bot integration I have everything setup but getting some errors with bot permissions

I gave it admin permissions when setting up the bot on discord and the discord bot pannel I also kicked and added the bot back
everything else seems to be running fine though

[proper bough]

anyone active?

[simple chasm]

Wow, I didn’t know the gpt4 api was out

[proper bough]

Fixed the issue

[proper bough]

If anyone runs into this issue watch this video: https://youtu.be/oXA1dIbNJmE

My issue is that I had two ids in my allowed guilds in the .env file I changed it to just my server Id and it fixed the issue

[lofty zenith]

This looks great. Is there no button to "Add Bot to Server" :/

[proper bough]

no u need to make the bot

[winged pollen]

No you have to install it yourself 🙂

[frigid halo]

here

[winged pollen]

A bunch of fixes have gone out! Sorry for no updates here for a while!
https://github.com/Kav-K/GPTDiscord

[winged pollen]

Will add gpt-4v to this when it's out on the API!

[brittle bison]

Kaveen hope you are doing well! Thanks for the continued work on this project

[winged pollen]

Likewise! And no problem! Main development is stopped currently but when major features like vision come out I will definitely keep it up to date

[winged pollen]

Hi everyone! I've fixed up a few bugs and released a V11.8.3 that has the project updated with the latest llama-index and langchain version. This project is still being cleaned and maintained so if you have any major breaking issues please don't hesitate to report them on github issues as always. As a reminder, this bot supports a lot of features! Internet chat with google and wolfram alpha, image understanding, and even chatting with your documents!

https://github.com/Kav-K/GPTDiscord

GPTDiscord - A robust GPT + DALL-E2 Discord bot, multi-modality, chat with documents, more!

[winged pollen]

Recent issue with docker builds fixed

[winged pollen]

Added my own implementation of Code Interpreter / Advanced Data Analysis into the bot! Just like ChatGPT! It is able to fully execute code, install packages, and install system packages and manipulate the environment.

V11.9.1

[winged pollen]

V11.9.5

• Added the ability to upload files to code interpreter for it to directly work with them
• Added the ability for code interpreter to use google as a tool automatically when the corresponding keys are available
• Added the ability for code interpreter to run terminal commands to manipulate the environment

[winged pollen]

V11.9.7
Made conversations infinite length for /internet chat, /code chat, and /index chat.
Improved index chat functionality to be a bit more accurate when answering questions

[winged pollen]

V11.9.8

• Fixed conversation summarization, stability fixes, slightly improved image understanding

[winged pollen]

V12.0.0 - Big /index chat revamp, better chatting with your documents!

• /index chat was revamped to allow drag-and-drop and a much better user experience for chatting with documents. You can now either just simply drop files into your /index chat and GPT will understand the files and allow you to chat with them, or you can put a link in the chat and GPT will connect to the link and interpret the webpage/data present there. You can then proceed to chat with the data however you'd like. One limitation is that GPT will not be able to give you direct quotes from the data, only answer using it, but this is something to be improved in the short future.
• Indexing youtube links / chatting with youtube links is now fixed
• Some code clean up

[frigid halo]

Have you thought about somehow implementing regex for the responses? after a message from GPT is generated

so that the bot's messages would be like:

I do not have an opinion.

Hello, how may I assist you today?

Instead of:

As an AI, I do not have an opinion.

Botname: Hello, how may I assist you today?

[winged pollen]

I dont feel like this provides any benefit

[winged pollen]

V12.0.5 - GPT-4-Vision and GPT-4-Turbo support
GPT-4-Vision Support
GPT-4-Turbo support across all features
Ability to understand multiple images in /gpt converse
Ability to use multiple tools in code, internet, and index chats (e.g search multiple times for one request, just like browsing with bing)
This will work right out of the box if your OpenAI organization has access to these models.

GPTDiscord - A robust GPT + DALL-E3 Discord bot, multi-modality (GPT-4V), chat with documents, more!

[frigid halo]

Not much benefit but I think it would be great as I got the impression that their new GPTs like Character AI won't say they're an AI

[winged pollen]

Ah I see, personally I like having the raw LLM output come out of the bot without any behind the scenes magic or editing

[valid gust]

Hi @winged pollen - The functionality that would interest me from having something incorporated into a DISCORD environment would be if it were able to read and parse within the hierarchy of the DISCORD, and to be able to respond to details about the contents. Specifically the DISCORD I am interested in has a lot of diverse information regarding trading stocks, options and other quantitative material spread over multiple folders. What is really needed is to articulate a query and to have the helper search through the whole DISCORD to try to answer the query in an intelligent way that goes beyond simply indexing or searching on keywords. Can I ask whether your GPT4 version is able to do this please, or whether you are aware of any other similar approaches that have been made. I am aware that there is a DISCORD helper (TLDR) that sort of does this, but it is a rather dumb application that does not have integration with the wider training database of GPT4, or any other LLM? Many thanks

[winged pollen]

@valid gust We have a way for the bot to index discord messages and then you can ask questions about that index but that's the extent of the functionality

I like the idea though of being able to answer questions across discord

If you submit it as a new issue on the repo I can work on it sometime

[winged pollen]

V12.1.5

• Fix pagination for connected chats
• Fix gpt-vision and gpt-turbo early stoppage due to weird max tokens
• Upgrade langchain, openai, llama_index for breaking changes

[winged pollen]

v12.2.0

• Check if attachments are images in /gpt converse
• Allow for text files to be used as prompts when conversing (e.g when discord makes your message into a text file because its too long)
• Fixed some bugs surrounding invalid attachments
• Dalle optimize uses DALLE 3
• Stability and error-catching fixes

[true river]

Hi, I just installed this and it works great, ty! Question: It's not clear to me how to tell it to use gpt-4-turbo and dalle-3. Am I just not seeing it in the docs?

[winged pollen]

You can do /system settings

and then set the model param to gpt-4-1106-preview

dalle-3 is just /draw by default

[true river]

perhaps it's my misunderstanding, but when I create an image via chatgpt/dalle3, it's creating far better images than what /draw is doing (which is why I assumed it wasn't using 3). Is there a knob I need to tune?

[true river]

e.g.: I used the exact same prompt for these 2 images. The left one was done via chatgpt, the right via this bot:

[winged pollen]

yeah I agree with this, I'm not sure why that might be happening, I'll look into it

[winged pollen]

can I see the images that the bot generated?

oh nvm

@true river track this: https://github.com/Kav-K/GPTDiscord/issues/417, will try to fix it up soon

[winged pollen]

Fixed!

https://github.com/Kav-K/GPTDiscord/releases/tag/v12.2.9

[winged pollen]

v12.3.0

• You can now work with GPT in a conversation to create images. This will work when you are in a gpt-vision conversation, the model will be able to draw (up to 4) images by itself, and will also be able to see images that you upload. Keep in mind that the vision API is limited to 100 requests per day, so you don't want to be spamming this until the API is fully released. This feature will ONLY work when the conversation model is gpt-4-vision-preview.

[true river]

@winged pollen - I thought I read that the bot will chat via direct mention, e.g.: @gptbot hello would just invoke a gpt chat. Did I get that wrong?

BOT_TAGGABLE = "True"```

[winged pollen]

No it will!

[true river]

@winged pollen - seems I am having trouble with permissions. But I can't tell from the docker log which permission is not available/set properly. discord.errors.ApplicationCommandInvokeError: Application Command raised an exception: Forbidden: 403 Forbidden (error code: 50001): Missing Access when I do some commands like `/private-test. I checked the GH readme's but don't see anything about which settings to use for discord roles for admins and users...but I'm also not certain that is the problem.

Also, when I mention the bot directly, I get no response at all, not even docker logs

[true river]

ok. think I finally got it figured out. I had commented some of teh roles out in .env since the docs said that if I comment them out it would just default to allow. Once I re-added them, things started working

[true river]

happy to report all is working well (images included) ty for the hard work @winged pollen !'

[winged pollen]

Awesome I'm glad to hear!!