Brave Enough For AutoGPT

[Bluedevil678]

Hi all

I was wondering if there was anyone brave enough to put together a AutoGPT container for the "Apps" section of Unraid? 

 

With the great success of Stable Diffusion that was added by pyrater, I was hoping we could encourage someone to help put together this bleeding edge tech container for us. 

 

Let me know your thoughts. 

Edited April 17, 2023 by Bluedevil678
Spelling correction

[freeridekid]

Definitely Interested

[talard]

Yes, me too. 🤞

[tombstone]

Yes I in anyone building one yet then we  need a good Ui with plugin 

[Reptar]

I was able to make a docker image with gotty for web based terminal access but don't really have any idea how to make a nice UI for auto-gpt using something like streamlit for example.... Has anyone else been able to make something?

 

I've been trying to get auto-gpt to do it (lol) but it's not working out too well.

[maverhick]

Is it possible to install autogpt without it being in the apps section? Can anyone share how to accomplish that?

 

[Reptar]

10 minutes ago, maverhick said:

Is it possible to install autogpt without it being in the apps section? Can anyone share how to accomplish that?

 

 

yes, search what youre looking for in apps, then click "Click here to get more results from dockerhub"

 

 

Once you find what you're looking for you can click Install. It'll ask you if you want unraid to  attempt to determine applicable paths and ports etc.
You'll probably need to manually enter variable names/values.

[tombstone]

On 4/25/2023 at 12:08 AM, Reptar said:

 

yes, search what youre looking for in apps, then click "Click here to get more results from dockerhub"

 

 

Once you find what you're looking for you can click Install. It'll ask you if you want unraid to  attempt to determine applicable paths and ports etc.
You'll probably need to manually enter variable names/values.

Can you share what variable you put in? 

[Reptar]

1 hour ago, tombstone said:

Can you share what variable you put in? 

That depends on the docker image you're using.

Here's the .env from the auto-gpt github though:

https://github.com/Significant-Gravitas/Auto-GPT/blob/master/.env.template

[tombstone]

19 hours ago, Reptar said:

That depends on the docker image you're using.

Here's the .env from the auto-gpt github though:

https://github.com/Significant-Gravitas/Auto-GPT/blob/master/.env.template

I am considering trying this. Only if you could provide me with a video or a how-to guide on how to do this to a Docker image?

[Reptar]

3 hours ago, tombstone said:

I am considering trying this. Only if you could provide me with a video or a how-to guide on how to do this to a Docker image?

I was able to make a docker image that uses gotty for a web based terminal (so you dont have to open auto-gpt in the console window every time)

docker image: autogpt_gotty

 

Docker Images needed:

• autogpt_gotty (reptard one -- You'll need to click "Click here to get more results from dockerhub" for it to show up)
  Note: you can allow unraid to detect config (that's what I did)
• redis (jj9987's repository)

 

API Keys Required:

• OpenAI key, see: https://docs.agpt.co/setup/ (Getting an API Key section)

Note: The instructions below also assumed you have set up a google api/custom search engine. you can omit those if you dont want to use google search. Otherwise, see here for instructions on how to acquire: https://docs.agpt.co/configuration/search/#google-api-keys-configuration

 

1. Install redis (default port is 6379) -- leave everything other than port as is. Port can be changed to whatever
2. Install autogpt_gotty
   A. Set container port 1 value. This can be any unused port.
   B. Add the following variables along with their values:
       - OPENAI_API_KEY
       - GOOGLE_API_KEY
       - CUSTOM_SEARCH_ENGINE_ID
       - MEMORY_BACKEND (value: redis)
       - REDIS_HOST (Value: your unraid IP)
       - REDIS_PORT (value: port set for redis container in step 1)
       - SMART_LLM_MODEL (value: gpt-3.5-turbo -- this assumes you do not have access to the gpt-4 API)
       Note: you can use add any of the other variables from the Auto-GPT repo
   C. Optional: If you want to save data written by Auto-GPT, add the following Path:
      - Name: Workspace Path
      - Container Path: /app/autogpt/auto_gpt_workspace
      - Host Path: Your desired unraid share/directory
      - Access Mode: Read/Write

[tombstone]

On 5/3/2023 at 11:49 PM, Reptar said:

That depends on the docker image you're using.

Here's the .env from the auto-gpt github though:

https://github.com/Significant-Gravitas/Auto-GPT/blob/master/.env.template

 

I am considering trying to make my our from an Docker image but Only if you could provide me with a video or a how-to guide on how to do this to a Docker image? Will be good to k ow also... What you have done above looks cool too

Quote

[Asmithcompute]

On 5/4/2023 at 10:05 AM, Reptar said:

I was able to make a docker image that uses gotty for a web based terminal (so you dont have to open auto-gpt in the console window every time)

docker image: autogpt_gotty

 

Docker Images needed:

• autogpt_gotty (reptard one -- You'll need to click "Click here to get more results from dockerhub" for it to show up)
  Note: you can allow unraid to detect config (that's what I did)
• redis (jj9987's repository)

 

API Keys Required:

• OpenAI key, see: https://docs.agpt.co/setup/ (Getting an API Key section)

Note: The instructions below also assumed you have set up a google api/custom search engine. you can omit those if you dont want to use google search. Otherwise, see here for instructions on how to acquire: https://docs.agpt.co/configuration/search/#google-api-keys-configuration

 

1. Install redis (default port is 6379) -- leave everything other than port as is. Port can be changed to whatever
2. Install autogpt_gotty
   A. Set container port 1 value. This can be any unused port.
   B. Add the following variables along with their values:
       - OPENAI_API_KEY
       - GOOGLE_API_KEY
       - CUSTOM_SEARCH_ENGINE_ID
       - MEMORY_BACKEND (value: redis)
       - REDIS_HOST (Value: your unraid IP)
       - REDIS_PORT (value: port set for redis container in step 1)
       - SMART_LLM_MODEL (value: gpt-3.5-turbo -- this assumes you do not have access to the gpt-4 API)
       Note: you can use add any of the other variables from the Auto-GPT repo
   C. Optional: If you want to save data written by Auto-GPT, add the following Path:
      - Name: Workspace Path
      - Container Path: /app/autogpt/auto_gpt_workspace
      - Host Path: Your desired unraid share/directory
      - Access Mode: Read/Write

so it works and boots!, but it seems im missing something. i keep getting a "redis.exceptions.ResponseError: unknown command 'FT.INFO', with args beginning with: 'auto-gpt'" error when its about to work. i have mapped /app/auto/gpt/auto_gpt_workspace. also redis keeps shooting this error: "Possible SECURITY ATTACK detected. It looks like somebody is sending POST or Host: commands to Redis. This is likely due to an attacker attempting to use Cross Protocol Scripting to compromise your Redis instance. Connection aborted"

[Reptar]

43 minutes ago, Asmithcompute said:

so it works and boots!, but it seems im missing something. i keep getting a "redis.exceptions.ResponseError: unknown command 'FT.INFO', with args beginning with: 'auto-gpt'" error when its about to work. i have mapped /app/auto/gpt/auto_gpt_workspace. also redis keeps shooting this error: "Possible SECURITY ATTACK detected. It looks like somebody is sending POST or Host: commands to Redis. This is likely due to an attacker attempting to use Cross Protocol Scripting to compromise your Redis instance. Connection aborted"

Forgot to mention, use this redis docker:

redis/redis-stack-server

[Asmithcompute]

this worked! awesome!

[Reptar]

14 hours ago, Asmithcompute said:

this worked! awesome!

Glad it all worked out!

[Asmithcompute]

hey, not sure how well versed you are in all of this, but I had autogtp going for a bit to create something, and after a while it accidently, sent too many tokens and just crashed. do you know of the way to alleviate this?

[Reptar]

16 hours ago, Asmithcompute said:

hey, not sure how well versed you are in all of this, but I had autogtp going for a bit to create something, and after a while it accidently, sent too many tokens and just crashed. do you know of the way to alleviate this?

 

Do you mean it just locked up and stopped showing text on the screen? or did you see errors pop up?

[Asmithcompute]

I actually fixed this issue by telling autogpt to keep token counts below the 4k thresholds, and if it thinks it may go over, to chunk up the text before thinking! it worked lol.

 

I have been getting a few "cannot save action to json", other than that it works great!

 

 

I do have another question off topic of this one. could you point me into a direction of a guide for making my own docker image of another project? you seem to know what you're doing, and this project here id love to be able to have as a docker image. unless it doesnt make sense to do so

[Reptar]

2 hours ago, Asmithcompute said:

I actually fixed this issue by telling autogpt to keep token counts below the 4k thresholds, and if it thinks it may go over, to chunk up the text before thinking! it worked lol.

 

I have been getting a few "cannot save action to json", other than that it works great!

 

 

I do have another question off topic of this one. could you point me into a direction of a guide for making my own docker image of another project? you seem to know what you're doing, and this project here id love to be able to have as a docker image. unless it doesnt make sense to do so

 

I actually have no idea what I'm doing, I basically looked at multiple random dockerfiles and tried to pull bits and pieces together until things worked. If errors popped up, I would research them and attempt fixes based on the research.

 

Unfortunately, I didn't use just one guide... I pulled information from various sources...

I watched this a while ago and I found her videos helpful in general:

 

For compiling the autogpt_gotty image, I used some of these sites to pull bits of information:

https://docs.docker.com/build/building/base-images/

https://devopscube.com/build-docker-image/

https://www.stereolabs.com/docs/docker/creating-your-image/

 

I also found someone else's project which did the same thing. At the time, it wasn't working at all though so I pulled snippets from their dockerfile and the main auto-gpt github repo and cobbled something together:

https://github.com/claytondukes/autogpt-docker

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

[MrGrey]

When this works, I'll stop using UnRAID... Maybe two weeks before GPT figures it out?

 

DAMN... I was looking forward to zfs.

 

MrGrey.

[Nano]

I setup AUTOGPT on a windows Server and wow, its really awesome at first, Then you realise nothing got done and you spent $1 on API calls in the space of 30 minutes  Its fun but its nothing ground breaking. 

Edited June 3, 2023 by Nano

[Reptar]

18 hours ago, Nano said:

I setup AUTOGPT on a windows Server and wow, its really awesome at first, Then you realise nothing got done and you spent $1 on API calls in the space of 30 minutes  Its fun but its nothing ground breaking. 

 

Yeah, it's not there yet but it sure is cool to look at!

 

I saw one guy who was mainly using to research and summarize things automatically. He basically asked it to look at what his company offered, look up a potential client and come up with marketing material that would look good to that specific potential client.

 

I think stuff like that probably works well... if you ask it to build something though, it just goes full herrdurr