Auto-GPT

[indigo tide]

https://github.com/Torantulino/Auto-GPT

Disclaimer: not my project. Just a community member.

If you haven't heard of this yet, you're living under a rock, and You're welcome.

YouTube search "AutO-GPT"
It's a lot easier to comprehend than asking on chat.
It can do a lot of things as an A.I. Pet that interacts with your PC. A Jarvis basically.

---

What it needs..

• OpenAi API key

• Pincone's API and Location
  (The vector Memory brain GPT can use for Long and Short term memory)

• Google API (not really necessary but it makes it easier to browser the web)

-Eleven Labs API (If you want a voice

---

If you haven't heard of Langchain
Langchain let's GPT interact with its environment.

---

It asks for permission everytime it's about to do something, unless you give it complete freedom.

---

If you have questions join the Auto-GPT community server.

[buoyant rune]

Please explain how this works.

[indigo tide]

connect it to the internet,
Give it a task, even tell it to create a folder with a file if necessary, or a pdf

it's an autonomous assistant

It asks permission everytime it's about to do something on it's own,
Unless you give it complete freedom

Which might get a little dangerous

It doesn't have to be GPT4
Gpt3.5 work pretty good with a Vector Memory from Pinecone too

[dry drift]

When I do this I get a whole bunch of traceback errors and my knowledge with this type of stuff is almost 0. Is there a way to work out these kinds of errors

Ignore that, apparently I needed the pinecone API and Env, I thought it was optional oops

[indigo tide]

Pinecone's the vector memory.
That's pretty necessary for gpt to have a brain to keep recalling from forever

long and short term memories

[dry drift]

It seems it wasn't what I thought it was anyway, oh well. Soon

[indigo tide]

It's your very own Jarvis, when you've got it fully setup

[blazing matrix]

Is each request to GPT 3.5 or 4 a paid one? Do you have to pay based on API calls?

Specifically talking about how Autogpt uses API and price associated with that?

[indigo tide]

You can make it so gpt3.5 is the only one being used

The command is in github

API call is free

You just need a pinecone account

[versed flower]

How much is it to run on Pinecone? @indigo tide

[indigo tide]

Free

[indigo tide]

Auto-GPT has a community server. It's in github discord server link

[modern lynx]

getting stuck on pinecone

Found the solution, should have been using the git shell to execute pip. Thanks again, it just OK.

[indigo tide]

API key and region location..

[indigo tide]

Thought of another use case for auto

I'll put my story summary to a .text file

Have it review my story plot.

Create a folder for each chapter and create 1 text file per page.

Name each folder Chapter 1 to Chapter 10

And fill each Chapter folder 20page text files.

[jagged garnet]

I've been playing with it a bit using gpt-3 only, but it seems like GPT-3 isn't really up to this task.

it frequently drifts away from what it's told in /scripts/data/prompt.txt causing it to have unknown command errors constantly.

[sage nexus]

I'm also stuck with gpt-3 and I'm underwhelmed by the results so far... it's trying to make something for an hour, burning through tokens but never really finishes something because it's running into errors. The output sounds extremely rudimentary. Yeah I know the technology itself is absolutely incredible and I'm stunned by the possibilities. I can't wait to get approved for the gpt4 API access 😄

[hazy moon]

@indigo tide are you the author of AutoGPT?

or one of the primary contributors?

[indigo tide]

A contributor

Err community member

Lemme put that as a disclaimer on top

[indigo tide]

I kept redirecting them everytime someone DMs me

[hallow yacht]

This is a highly experimental and unpolished and I would personally recommend people to wait on it to mature more before you invest time in the tedious setup. It's not very smart with gpt-3.5, and very few of us have access to anything else. If you have to ability to contribute to a project like this; it seems like a worthwhile place to invest time for future payoff.

[jagged garnet]

Yeah, it would probably be decent with GPT-4. At the very least you can review the source to see what's going on under the hood with GPT-3, but if you are not interested in developing software it would be best to wait for at least a beta release.

[indigo tide]

my friend's about to beat me on our Langchain race

😂

The man used over 1million tokens in 1 go

[jagged garnet]

I noticed yesterday Torantulino started a new repository on GitHub at Torantulino/Auto-GPT-Plugins
Sounds interesting.

[pastel temple]

lets talk! add me

[indigo tide]

Have they already started refactoring?
Nice

[pastel temple]

Are you planning on contributing to the repo more?

[indigo tide]

ye

Working with my friend

Modifying code base, and he's gonna go get a better pull request

[pastel temple]

@indigo tide know of or have any ideas for new features rolling out?

[indigo tide]

Gonna name it "Prometheus GPT"
A completely re-vamp of auto-GPT for Langchain.
and ones that will work with LLM and Alpaca models too

tons of stuff

You'll see when it's done

[feral patrol]

@indigo tide this looks good man. I got another project that you may like to start to pm me just sent ya an friend request

[feral patrol]

Yep. See they too

[jagged garnet]

Torantulino/Auto-GPT-Plugin-Template now exists, too

[sage nexus]

I've tested many things so far and I'm blown away by the speak function, that's so fantastic.

Unfortunately I'm always getting errors along the way which prevent the GPT to execute the stuff it thinks it should do.

Is this because atm I'm bound to GPT3?

[jagged garnet]

It seems it's partly that, and also some internal parsing errors (estimating based on their recent GitHub issues)

[sage nexus]

That‘s sich a bummer… I have endless ideas and I‘m eagerly waiting for my gpt4 api access which could come every minute or in 6 months…

[feral patrol]

Me 2 I think GPT 5 will be out by then.

[sage nexus]

Don't get me started about GPT5.. I'm blown away by GTP4 and I've read a lot about the possible improvements until GPT5 and I just can't wait for it to be released.

[lime pewter]

im also blown away by the demo on the YouTube , though i'm a complete laymen about coding, i want to try it out cause' this kind of automation is something i always dreamed about, it feels like i can't catch up with the progress of AI in the recent months, today i noticed the open source of deepspeed, it feels so excited and exhausted at the same time.

[jagged garnet]

I have been having some luck using GPT-3.5 in Auto-GPT by re-writing it's prompts. I will open a pull request on GitHub if I get anything worth sharing.

Last time I ran it using the default goals with my own prompting, it got as far as looking at how it could open a credit card to fund a new business, but then realized it couldn't because it is not a human. A far-cry better than it was doing before with 3.5 going into loops or not producing the correct JSON response format and erroring-out.

[sage nexus]

I had one situation where one of the agents where not sure about some api handling and then it searched the website or an email address to contact for more details. While the idea is pretty good, I stopped the agent 😉

[jagged garnet]

Yeah, I was just running a test on my prompt changes and Auto-GPT was looking up contact information for investors and composing email drafts to them. I also stopped it.

[gusty parrot]

My favorite project

[west tree]

i see that this project has voice for the AI as an option, but does it also have speech recognition?

cuz if it does, its already 100% better then some other projects here, that require you to join a waitlist

though i see nothing related to speech recognition in the github page

[west tree]

aah.. the elevenlabs doesnt even seem to be working

[sage nexus]

Elevenlabs works great, it's a blast to hear the AI speak 🙂

[jagged garnet]

I agree. I love that it lets you make custom voices, too

[unreal wren]

Traceback (most recent call last):
File "C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.10_3.10.3056.0_x64__qbz5n2kfra8p0\lib\runpy.py", line 196, in run_module_as_main
return run_code(code, main_globals, None,
File "C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.10_3.10.3056.0_x64__qbz5n2kfra8p0\lib\runpy.py", line 86, in run_code
exec(code, run_globals)
File "C:\Users\giaco\Desktop\Auto-GPT\autogpt_main.py", line 4, in <module>
from autogpt.agent.agent import Agent
File "C:\Users\giaco\Desktop\Auto-GPT\autogpt\agent_init.py", line 1, in <module>
from autogpt.agent.agent import Agent
File "C:\Users\giaco\Desktop\Auto-GPT\autogpt\agent\agent.py", line 2, in <module>
from autogpt.app import execute_command, get_command
File "C:\Users\giaco\Desktop\Auto-GPT\autogpt\app.py", line 25, in <module>
from autogpt.commands.git_operations import clone_repository
File "C:\Users\giaco\Desktop\Auto-GPT\autogpt\commands\git_operations.py", line 1, in <module>
import git
File "C:\Users\giaco\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\git_init.py", line 91, in <module>
raise ImportError("Failed to initialize: {0}".format(exc)) from exc
ImportError: Failed to initialize: Bad git executable.
The git executable must be specified in one of the following ways:
- be included in your $PATH
- be set via $GIT_PYTHON_GIT_EXECUTABLE
- explicitly set via git.refresh()

All git commands will error until this is rectified.

This initial warning can be silenced or aggravated in the future by setting the
$GIT_PYTHON_REFRESH environment variable. Use one of the following values:
- quiet|q|silence|s|none|n|0: for no warning or exception
- warn|w|warning|1: for a printed warning
- error|e|raise|r|2: for a raised exception

Example:
export GIT_PYTHON_REFRESH=quiet

hmmm

[jagged garnet]

They have an issues reporting area on the github page

[pearl trail]

Did you connect to Pinecone?
I am able to run Auto-GPT if the .env setting is MEMORY_BACKEND=local
But switching to MEMORY_BACKEND=pinecone just hangs. I have my pinecone api key and region in the .env file. But apart from signing up for a free pinecone account, haven't done anything else on their end.

[jagged garnet]

Pinecone has been having issues with the recent influx of users. People's indexes are stuck initializing for days on end. I am entering day 4 of initializing my pinecone index.

It is supposed to only take a few minutes at most, but they got overwhelmed. If you have a free account, you will get backburner treatment. Upgrading to a paid account does not help expedite it, you would need to have been a paid user before starting the index. You can't delete the index until it initializes, so that's not an option either.

At this point, I am moving away from Pinecone. Auto-GPT is starting to add alternatives like Milvus, Weaviate, Redis, etc.

[pearl trail]

How can I make the web browsing actually stop popping up windown on my screen? The google API info is in the .env file

[mossy bramble]

i don't think its worth it especially with all the traffic of users

[balmy gull]

how long is the waitlist? I applied

for pinecone

[pale hedge]

Can anyone help me with auto GPT? Am fairly new and have tried it and been disappointed so far but it's most likely me using it wrong.... Problem is when I tell it what to do, it just acts like it wants to do something and keeps repeating itself and just going around in circles.... nothing meaningful...

[balmy gull]

lol same do you have local memory

i think you need the pinecone api stufgf

[jagged garnet]

Looping is a common issue when running in gpt-3 only mode. GPT-4 is what it was actually designed for.

[balmy gull]

the pinecone api makes a big different right

because my autoGPT is just being a dumb

[jagged garnet]

It looks like they finally scaled up a couple days ago. It meant that everyone who was stuck waiting had to delete and re-create their indexes, but after that it took about 2 min to initialize a free index.

[balmy gull]

oh really?

[jagged garnet]

Pinecone is used to store vector information in regards to embeddings. When you embed data, some of the fidelity is lost because embeddings are just strings of token ids. The vector database provides context info to GPT and is much faster than the localcache memory

[jagged garnet]

I was meaning the wait for a pincone index, GPT-4 api waitlist is killing me too.

[balmy gull]

yeah i meant pinecone too

[jagged garnet]

ok

[balmy gull]

only been on it a day but

so with pinecone, autogpt can actually do things?

[jagged garnet]

after they split the northamerica server into uswest, uscentral, and useast.. the times are down to about 5 min or less to make an index.

It will be better at it, but unless you have GPT-4, you are going to have to put extra effort into the ai's role and goals entries

[balmy gull]

oh man im dumb

i forgot gpt-4 api has a waitlist

i thought i was using it

[jagged garnet]

basically apply GPT-3.5 prompt engineering to it. Unfortunately, they took out the prompt.txt so we can't modify their instructions easily.

[balmy gull]

that sucks huhh

maybe old versions work

[jagged garnet]

Yeah. They have the application creating the system prompt on the fly, so it's made up of various chunks and assembled by code.

On the earlier versions, strict prompting like SudoLang or similar was very effective

On one of my best runs of it, it tried to apply for a credit card and composed emails to potential investors. I was using their default goal of running businesses.

I can only imagine what GPT-4 could do, since it's actually good at using external tools.

[balmy gull]

whaaaat

haha thats amazing

yeah this is gonna be ridiculous technology

very soon

[jagged garnet]

Yeah, it gave up on the CC when it realized it's not a human and doesn't have a SSN

[balmy gull]

i wonder what would happen if you just jailbroke it with gpt-4 and infinite context memory

cause when i jailbroke my weak setup it kept trying to virus me lol

but was failing hard

[balmy gull]

haha

[jagged garnet]

Hard to say, the issue with jailbreaking is that it goes off-script so far that it doesn't really do tasks well. It just sasses you, mostly.

[balmy gull]

its been doing some gpt 1 type things with me

scraping the web and finding a file on that website with some data it wanted

just assumed that file was in the root directory

and got stuck in a loop of trying to access it lol

[jagged garnet]

Yeah, GPT-3.5 feels like an idiot compared to GPT-4

Another behavior that is more prevalent in GPT-3.5 is that the AI will stop relying on itself and will web search everything. Even things it could have answered fine without the internet.

I had to stop one run where I asked it to make a python app because it spent a long time trying to read the python documentation and look for examples of it on GitHub when it could just write in python.

[balmy gull]

yeah this was making me mad lol

acting like bingAI

[jagged garnet]

I just had an idea to test for that. Use the goals like
Goal 1: "Step 1 - Without using your commands, do your best to complete the task and save your output to 'file1.txt'."
Goal 2: "Step 2 - Using your commands, research the topic further online and save summary to 'file2.txt'."
Goal 3: "Step 3 - Use the contents of 'file2.txt' to update and refine the contents of 'file1.txt'."

then use goals 4 and 5 for additional processing.

I will try it later when I have some time

[balmy gull]

nice

yeah i was also thinking of how to get around that

i thought multiple agents were like

simultaneous for a bit

so i was trying to get them to use its commands

[jagged garnet]

I will also suggest a feature to them of an instruction that should be performed by the AI before it is prompted about the commands available. This would ensure it relies on it's own logic first, then gets introduced the commands.

[balmy gull]

i cant wait for pinecone

hate how everythings on a waiting list lol

ai's too popular

[jagged garnet]

Pinecone is doing a waitlist? They were slow-walking index creation for new users, but I didn't realize they started a waitlist.

There are alternatives for Auto-GPT now:

Redis

Milvus, Weaviate

[balmy gull]

ok

are they not as good

[jagged garnet]

Redis will be slightly less-good, because it isn't a native vector DB

the others are

Milvus and Weaviate have self-hosting options, but you need to check the hardware requirements out

In your env file, they didn't update the ### MEMORY_BACKEND section, but if you are on the latest stable release, you can scroll down to the weaviate area and it tells you to set the memory backend to 'weaviate' to use it.

If you are on Linux, Weaviate has a cool feature (embedded mode) that allows applications to deploy docker instances of it on-demand. That means you don't need to configure it manually and it will auto-start with Auto-GPT.

[balmy gull]

ooh

sounds cool

combining technologies = op

@jagged garnet have you discovered any interesting prompts with auto gpt

[jagged garnet]

I haven't really come up with anything since the app changed to not use the prompt.txt. With that, though, I had mostly described the expected behavior in SudoLang, which really helped GPT-3.5. It was like adding training wheels for it so it could ride the bike.

[indigo tide]

I got banned from AutoGPT discord, no warning

what I do?

[jagged garnet]

They have a discord?

Oh, cool I see it in the GitHub

[jagged garnet]

Best idea is to reach out to one of these people

[indigo tide]

both me and my wife got banned

[jagged garnet]

Yeah, I didn't even realize they had a discord until now.

[indigo tide]

they had no reason to ban me, I was just on general chat, talking about current topics

[jagged garnet]

Significant Gravitas is Torantula#5659, try DM'ing them I guess

[indigo tide]

thanks

[jagged garnet]

yw

[indigo tide]

ouf, I gotta friend them

don't i?

[jagged garnet]

I googled it

If you want to DM someone who is not on your friends list on Discord, you can search for their name in the search bar at the top of the screen. Once you have found them, click on their name and then select "Send Message."

[indigo tide]

ah, thanks

[jagged garnet]

it means this search when viewing friends list

[normal rivet]

anyone here

can't setup autogpt

Why it is not executing tasks

• gives warning: the file "AutoGpt.json" does not exist .Local memory would not be saved
  Requirements.txt gives this after pip

[balmy gull]

i had this problem last night when i was trying to install it lol

shoot i don't remember what i did specifically

oh

do you have a AutoGPT.json file

AutoGpt

[normal rivet]

yes

[normal rivet]

i am missing this one

i am using it on ubuntu

[normal rivet]

did it resolved?

[balmy gull]

I believe it fixed the problem. add a AutoGpt.json in your primary directory

then add two empty brackets

"{}"

and save

i think that may have fixed it

idk if it was only that

[normal rivet]

where do i find json file it should be in autogpt repo but this is not the case

[normal rivet]

i wish i could find it

[balmy gull]

yeah idk

whats your specific error again

?

[normal rivet]

warning: the file "AutoGpt.json" does not exist .Local memory would not be saved

[balmy gull]

huh

thats just a warning though right

not an error

[normal rivet]

see this

this come when i set goals and run it

[balmy gull]

its saying theres an error in your azure environment

or

config file

azure variable

did you set that upright

waiiiit

you';re getting this error?

i didnt even see that

yeah you just have to add the file then i imagine

[normal rivet]

ok ig azure issue but without using it i don't even proceed it says thinking

[normal rivet]

yes

[balmy gull]

wait do you have an AutoGpt.json file

in your main directory

or nah

[normal rivet]

no its not i can't find anywhere

[balmy gull]

oh

want me to just tell you where to put it

just go to the file explorer

and add "AutoGpt.json" under your main directory

meaning it would be alongside requirements.txt

and other things

[normal rivet]

i am in ubuntu

ok

leme see

no json file here

[balmy gull]

just add it

click make new file whereever that is in ubuntu

name it "AutoGpt.json"

then edit the text so that the inside of it is just "{}"

and save it

then close

and try again

[normal rivet]

ok

only brackets

without quotation

it stucked on thinking again however json file error fixed

thxs

[balmy gull]

np

[grizzled geyser]

So I am wanting to setup a memory backend (first time playing around with Auto-GPT and since pinecone is now waitlisted, I have been looking into all of the alternatives. But I'm trying to find something that can install standalone on a Windows 11 machine self hosted. Are none of these alternatives an option?

[normal rivet]

i am using vmware cause of this

windows 10

[grizzled geyser]

I mean, I was just hoping to keep everything on the same OS. I don't have anything against spinning up a VM but isn't one of the vector databases installable standalone on Windows?

[normal rivet]

What about Faiss

[grizzled geyser]

What's that?

[normal rivet]

Faiss is an open-source vector database that can be installed on Windows, and can be used to store and search high-dimensional vectors.

You need standalone on windows

VMware is easy solution if you're ok

[normal rivet]

What to do when api reach limit Openai

[pale hedge]

So do you think it works better with the pinecone api?

[pale hedge]

Am in the GPT4 waitlist. How long does this take approximately?

[jagged garnet]

No idea, I've been on the waitlist since day 1. OpenAI said they will grant access to people who submit 'exceptional' evals to the eval project on github and have their pull request merged. Last I looked, they had a backlog of 400 pull requests.

https://github.com/openai/evals

[jagged garnet]

It works better with any vector database. Pinecone is a good choice, because it offers a free tier and you don't need to worry DIY'ing it. There are other options, such as Redis, Milvus, and Weaviate which offer both hosted and self-hosted options.

When you embed data, which is what A-GPT is doing, it converts it into a string of numeric tokens. The vector database stores information about how those tokens interrelate so the AI can quickly use them in it's generation.

Localcache and Redis are not vector dbs, but can be used in a similar way. This means they will work, but a true vector DB would be much more efficient.

As for behavior, it doesn't really matter that much. It's more a performance thing than anything.

[pale hedge]

Ok thx

[balmy gull]

i did something cool with autoGPT even though im using local memory

it made a list of apartments still leasing for fall 2023 near my college campus alongside their advertised starting prices if any

and appended it to a new txt file

i can imagine it would have been able to do much more if i had some memory manager

i need to set one up asap

[jagged garnet]

cool

real-world use cases ftw

[balmy gull]

for real

i bet i could have gotten it to automatically contact the ones that required an inquiry for leasing price / availibility? or at least create a txt file with the email address, subject line, and body

i'd be more comfortable with that anyway lol

[jagged garnet]

Yeah, let the project mature before you allow the AI to start emailing people on your behalf lol

[balmy gull]

yeah lol

jeez imagine a virtual environment entirely for an automous agent like autoGPT

once it gets more advanced

imagine having permenant memory stored accross different session

and creating an agaent with its own email, desktop, etc

file storage

i bet a project like that is something that is already plausible tbh

**__Question: __**In what way is docker related to autoGPT? I am trying to use milvus for my memory backend but my current autoGPT setup doesn't seem to be using docker

[jagged garnet]

Perm memory is already a thing with Auto-GPT. If you continue from last time, it should have the old convo in Auto-GPT.json to load up.

[jagged garnet]

Docker is used to containerize it, so it can't effect the host filesystem. Some of the database options can also auto-deploy as docker containers. Containerizing means to trap it in a tiny virtual machine so it can't do things outside of it's "sandbox"

It looks like people are working on API access for Google Drive. Also there is an effort to make Auto-GPT capable of reading it's own GitHub repo and submitting pull requests to enhance itself.

I spent some time on their GitHub reading pull requests and the discussions

[balmy gull]

how do i know if my pinecone memory storatge is working?

my autogpt is still getting stuck in the same lops

loops

[jagged garnet]

Pinecone lets it store and ingest data to increase performance, but it doesn't make GPT-3.5 smarter, unfortunately. For looping, try adjusting the wording of the goals / role.

[balmy gull]

is there any way to get it to communicate its findings with its ganets? its been stuck in this look a while

[jagged garnet]

Also, they released version 0.2.2 12 hours ago, not sure which version you have there.

[balmy gull]

the previous version

have you found any ways to get it to stop looping errors

[jagged garnet]

I need more time in my days, lol. My guess is there could be a way to modify how they communicate via prompting, but I haven't had time to understand the new way AGPT builds prompts.

[balmy gull]

like a goal of "every time a command doesnt work remember what happened and dont repeat the same behavior" or something

i told it to input all the data that it wants its agents to use into the message it sends it and that seems to be helping a bit

[jagged garnet]

That could be adjusted with the Frequency penalty variable

[balmy gull]

oh ok

nice

[jagged garnet]

idk what they have it set at

First glance, it looks like they don't set the frequency_penalty at all

In llm_utils.py, they define the temperature as 0.9, but I can't find any references to frequency_penalty

[indigo tide]

So does it see your computer to see what's it's doing?

@indigo tide

And can I train it myself

And can it train itself

So I need gpt4?

Also what does it use to scan the screen

Oh

Dang

Dang

[jagged garnet]

Where did you get the idea they are making a private version? Their GitHub seems to be in active development on it as an OSS project. They recently started adding plugin support so people could expand it's features easier.

https://github.com/Significant-Gravitas/Auto-GPT-Plugin-Template

[jagged garnet]

OK, I hadn't heard of that. I've only heard talk of BabyAGI and Auto-GPT.

Well, until then, these public projects are the most intriguing things in this realm so there's no reason to abandon them before there's an alternative. I also wonder why it matters to keep it private for safety, then release it public.. someone will just edit the code to make it dangerous again. It's the ever present 'some people are why we can't have nice things' problem... which seems to be why ChaosGPT exists.

[hazy moon]

Hey so there's a new rule for api project posts, it has to be the primary maintainer(s) posting

[pale hedge]

So you are on the waitlist since day 1 and you already have access or not?

[jagged garnet]

I do not

[pale hedge]

ah ok

[jagged garnet]

I got the confirmation email, but no invite yet.

[pale hedge]

are there even people who have access to it? I mean if you are since day 1, it is already a little longer now... and on github 400 pull requests as if they don't comply

[jagged garnet]

I know some people got it, but idk if there have been any recently

[pale hedge]

Hm interesting

[rapid hatch]

Does anyone know of others API's clients that give internet access to GPT4? How does autoGPT do it?

I want it to use it manually but using API, so I have infinite queries

And would be sweet with a good interface for different conversations etc

[ebon plinth]

yeah,
just chat interface web UI with webscrapers

[rapid hatch]

Which one is it?

[ebon plinth]

My personal favorite is "OpenCharacters"

but there's also turbogpt.ai , which has a better looking UI

[pale hedge]

@rapid hatch since when do you have access to gpt4?

[rapid hatch]

Since release in March

[jagged garnet]

At first, it just used the Google Search API and a basic scrape function. The prompt tells GPT about some internal commands it can use like {"browse_web": "<URL>} and {"google_search": "<keywords>"} If the software detected these commands, it would run corrisponding functions, then return the data back to GPT.

Now it has options for Chrome/Selenium libraries, alternate search engine options, etc.

[rapid hatch]

Thanks!

[red matrix]

Hi, can anyone help me to run autogpt?

Warning: The file 'auto-gpt.json' does not exist. Local memory would not be saved to a file.
from selenium.webdriver.safari.options import Options as SafariOptions
ModuleNotFoundError: No module named 'selenium.webdriver.safari.options'

[west hornet]

can i use anything else aside from pinecone for the api?

[jagged garnet]

Pinecone is one of several options for databases. Review the readme.md on github for other options like Weaviate, Milvus, Redis, etc.

[jagged garnet]

Do you have the latest release version 0.2.2?

Ensure you are on the latest, then install all packages in requirements.txt

[pastel oar]

Auto-GPT is really making waves. It got mentioned by Prof. Gary Marcus on the BBC News yesterday.

[ripe otter]

does anyone know how to fix error? "cannot import name "main" from "autogpt"(path). thanks

[balmy gull]

can you provide more context

like the file stack and if you've already done "pip install -r requirements.txt" and what version you're using

[ripe otter]

r requirements r installed, pip 23.1.1, python 3.10

also this happened

thanks

[balmy gull]

@ripe otter add python3 to your environment variables

path*

just look up how to do it links r banned

[ripe otter]

didn't work, more suggestions? thnks

[balmy gull]

@ripe otter do you have AutoGpt.json in your root directory?

if not create it and add "{}" inside of it and save

[ripe otter]

run tests, auto-gpt.json added, openai key is set

is it a big problem?

[balmy gull]

It’s trying to test hugging interface

Which you don’t have

Just don’t test it

[jagged garnet]

I usually start by asking if they are using the stable branch, since A-GPT has master on their GitHub being used for development and stable is the release branch.

If you just clone it like normal, you may get a buggy development version

[iron quiver]

I did not give PineCode and Google API keys; but it kept running for hours? Is that the issue then?

[iron quiver]

Lol someone like me 🙂 been waiting for ever

[iron quiver]

Is there a document explaining the configuration? New to AutoGPT

[jagged garnet]

on the GitHub page, the Readme.MD

[outer cliff]

How to get the api keys for google custom search for keying in env file? I can create project but could not see any button to click get api keys. Anybody?

[jagged garnet]

It's in the readme

Can't TLDR the readme on experimental software 😉

[outer cliff]

ya I read the readme and gotten an api key but when running autogpt it keeps give me this _> returned "API key not valid. Please pass a valid API key.". Details: "[{'message': 'API key not valid. Please pass a valid API key.', 'domain': 'global', 'reason': 'badRequest'}]">

[ripe otter]

same here can't make it work:(

[jagged garnet]

are you using the release, or did you clone the master branch of the repo?

the master branch is their development branch, stable branch is.. stable.

[west hornet]

hey am i chatgpt has been like this since i set up autogpt could that be the cause?

oh and also am when i give autogpt a command it doesnt really do anything

[outer cliff]

stable branch

[jagged garnet]

What does this have to do with Auto-GPT? Take it to #chatgpt-discussions

[west hornet]

i was asking if auto-gpt might have caused this

[jagged garnet]

no

API is working

[west hornet]

oh okay

[jagged garnet]

its just ChatGPT

Playground site still works, too
https://platform.openai.com/playground?mode=chat

[west hornet]

to clearify when ever i give it a command it says thinking for a while then restarts

[jagged garnet]

Pop over to #chatgpt-discussions and you will see dozens of kids having a meltdown about homework they were using it for when it went down.

[west hornet]

this is about autogpt

[jagged garnet]

Oh, I misunderstood then.

Not sure, maybe kill and restart the task? I am not running auto-gpt at the moment, but my other API powered apps are still working.

[west hornet]

hm i will get to u later am about to head to school thank u for ur time

[ebon plinth]

did you set up the .env file properly?

[outer cliff]

anybody manage to setup google api keys and ID to make google search work properly?

does it work with 3.5 or you need to subscribe to 4

[west hornet]

yes i think so

its just stuck like this

[pastel oar]

You dont have a .json for memory. Go to the Main Directory of auto gpt and Create a file named "auto-gpt.json". Open it and put brackets in it. Dont know if this has something todo with this. But its definetly something you should do.

[ebon plinth]

you've got the wrong branch

[west hornet]

am how do i change it

[ebon plinth]

https://significant-gravitas.github.io/Auto-GPT/installation/

[west hornet]

so i get this?

the stable build

[ebon plinth]

just do it by the git

[west hornet]

it says i need billing set up but am still able to get the api for open ai or is it somethin else

[jagged garnet]

Yes

[jagged garnet]

It works somewhat on 3.5, but it's more likely to get stuck in a loop or go off on a tangent than GPT-4 would.

[west hornet]

so i got to pay to work it?

[jagged garnet]

Not sure which API you are asking about, I was talking about google and it's free for a limited number of projects, and they have paid tier if you want more.

The Google API is also optional for AutoGPT

https://significant-gravitas.github.io/Auto-GPT/configuration/search/

OpenAI's API is not free, but new users are given a credit of ~$18 to test it out. That $18 expires if not used in 3 months.

[ripe otter]

what gpt version r u using? v0.2.2 stable does not work. stuck with error.

please note API key setup in .env

thanks

[jagged garnet]

I use 0.2.2

For mine, I am doing the docker devcontainer method

try copying your .env file into the Auto-GPT/autogpt/ folder

[ripe otter]

OK. Made it work. had to set up api as environment variable. Thanks all

[outer cliff]

I can't even make it do a google search it keeps saying error with api keys.....

SYSTEM: Command google returned: Error: <HttpError 400 when requesting https://customsearch.googleapis.com/customsearc.... "API key not valid. Please pass a valid API key.". Details: "[{'message': 'API key not valid. Please pass a valid API key.', 'domain': 'global', 'reason': 'badRequest'}]">

[opaque jungle]

Is the only part of this project that costs money the OpenAI API?

[ripe otter]

does anyone know how to increase models maximum context length(tokens)? I get same error nomatter what I run through autogpt.

[ebon plinth]

do you have the memory set up?

Local / Weaviate or Pinecone

Tokens will remain the same.
it's just gonna API call the long term and permanent memory from the Vector DB

[severe fractal]

https://github.com/Significant-Gravitas/Auto-GPT#readme

This is the latest repo, just follow their documentation, its super easy.

All you need is to install powershell, python, git for windows and docker to run.

https://significant-gravitas.github.io/Auto-GPT/setup/

[covert zinc]

168К words man. 42k Tokens its... Bruh.

[silver moss]

how to purchase openAI api from africa...please help

[jagged garnet]

Sadly, if your country is not on this list then it's not supported: https://platform.openai.com/docs/supported-countries

[hollow marsh]

is the memory available since everytime I run it, it needs a name, and I named it different each time😅

[silver moss]

do you have info about Rapid APIs?

[jagged garnet]

Hadn't heard of it before, no.

[wind basin]

This wil be the future and using chat Gpt to improve chat Gpt will be epic

[visual bone]

SYSTEM: Command create_folder returned: Unknown command 'create_folder'
I have the file_operations.py from Gravitas. Is there an updated version that includes create_folder?

[west hornet]

Question what do i need to pay to use auto?

[lavish bolt]

is it free for gpt 3.5

do I need to pay after several prompt based on tokens

[native anvil]

I don't see a community server link in the github repo. Community server do you mean like a discord to join? I only see twitter to follow

[ebon plinth]

Ven's gone

[native anvil]

😩

[ebon plinth]

try using youtube

description discord links

[native anvil]

Alright thanks

[opaque jungle]

So I noticed that even though I never got a notification I now have access to the GPT4 API ..

I found this out only by starting AutoGPT with GPT4 only mode and then checking my usage at OpenAI and sure enough I was being billed for the GPT4 API even though OpenAI never sent me an email saying I had access

[covert zinc]

Maybe in "spam"?

[vast sphinx]

Autogpt have memory yet?

[balmy gull]

wdym

autogpt has whatever memory you configure for it

if you dont have pinecone memory setup or any other memory management then it will default to local memory (trash)

[pearl sequoia]

How much would you estimate that each multi-task goal you give it costs in terms of api requests. GPT 3.5 or GPT 4

[ebon plinth]

it's been so long since I touched this project

[torn vortex]

is possible to integrate with the project that allows infinite context?

[torn vortex]

i had some troubles to make it work, and somehow i manage it

but, context is an big problem, it can write small and simples apps but im missing something to meke it fully operational

[river crown]

i'm interested in doing something like this. will you be willing to share how you did it?

[kind hill]

how long did it take you?

[opaque jungle]

Idk .. I didn't get any notification... I just tried it and it worked

[balmy gull]

Was this just a trend

I feel like it was a proof of concept at this point

[jagged garnet]

check out the project boards, they are still working on new features
https://github.com/Significant-Gravitas/Auto-GPT/projects?query=is%3Aopen

[balmy gull]

Will do

I want it to take off

Cause sooner or later something like that will