How can we access all the information/tasks the AI agent has completed for us?

[byc89]

So this may sound like a dumb question here, but where do we access all the work the AI agent has done for our goal?

I cant locate it anywhere.

[Julz19]

If you specifically told the ai to save data it created to a file and use see it use the Write To File command it the command is successful, then usually you can access the information it saves to a file or location for you by going to the main autogpt folder and then finding the workspace folder it as created for itself. Or if you want the auto got to direct to it then usually to it to use the search file command and search for either a specific file or all files. For a specific file the path to tell auto gpt to use will usually be “auto_gpt_workspace/(File name here along with the files extension type)” and for all files in the directory use the file path “auto_gpt_workspace”. I hope this helps if it doesn’t let me know so I can try helping a bit more if possible.

[byc89]

Wow, super helpful man. Thanks so much for this!

[Julz19]

Remember you may also need to watch for the commands and arguments it tries to use as sometimes it’s short term memory can cause it to forget the correct file paths throughout the process if not specified in the goals it has, if you choose to not use the file path info to much in the goals prompts then you can try doin it every other goal and then just watching and telling it to use the correct file path when it tries using the wrong path.

[byc89]

Yes thanks for that. I noticed that happening throughout it's tasks, even though we has specified it in the goals. It would save some information , but not all of it which was frustrating.

I had another question in regards to the agent actually carrying the tasks it wants to do.

When the agent goes to execute it’s suggested tasks, such as web scraping, google search keyword data and trends etc. It doesn’t seem to actually do it, because it obviously doesn’t have logins, or access to these tools to actually do what it wants to do.

So do we need to set up api keys linked to these types tools in autogpt so it can carry out these tasks?

So for google, would having a google search console api key help with this?

Are they any particular API keys/tools that an agent should have access to in order to carry out tasks that it wants to?

If so, what ones, and how would we set them up with auto gpt?

[Julz19]

to further specify down on your question, if you are talking about the agents the autogpt system can create then to my knowledge i am not sure the agents themselves can access files nor websites nor do i know a way for this to be possible. from what i do know though is throughout my running and usage of auto-gpt and its agents system i do know that if autogpt prompts the agents and create them a certain way then i have witnessed it give them full access and permisson to search the web and files that it can itself but i am honestly unsure of the prompts needed to cause this as my run ins with it have always seemed to be random happenings throughtout the process. it could be possible that through constantly giving it feedback rather than letting it run alone all the time it will get better with how it operates and then can allow this access to agents down the line of its progress but im not 100% with that info. but i would suggest giving it feedback and stuff as much as possible as i have realised that it operates and runs extremely better when constantly directed what to do than be autonomous.
the autonomous part of it that is nice is like saving info to files or writing prompts or scripts etc into files as well but other than that i have noticed that letting it run on its own it tends to constantly get more and more sidtracked down the road some human feedback is better for it constantly in my opinion. :) hope this helps answer some of your question, sorry i couldnt provide a detailed and explicit answer fully just off my knowledge.

[Julz19]

the system overall operates extremely well compared to chatgpt and stuff as it can carry out task for you but the fully autonomous part as of right now is what it struggles to handle correctly :/ looking into it tho myself while the people on this page look for ways to make it better in this aspect tho sooner down the line.

[Rollesy]

I have found asking it to make a script to write and execute codes and also a script for web scraping to assist in its functionality has made it more autonomous but it was making them in the agent workspace, mine had a parsing error and i couldn't get the agent i was using back up and running so started again but have asked it to make a folder we can both access the scripts with any agents and it seems to be working, it is writing its own python scripts now and i am going to make more agents within these scripts to help it function better, im sure there is a better way but that's what i have found so far.