Video breakdown of AgentForge #76
Replies: 3 comments 2 replies
-
|
THIS IS GREAT! Note that we've changed the format of the prompt file. If you look in the dev branch, you'll see that you don't need the 'vars' field for each prompt any more. We're going to be working on documentation for that soon. I really love the graphs. Excalidraw is a great tool and looks really good. Your presentation/speech cadence is good, but it comes off a bit rigid. A good way to sound more natural and less like you're reading from a script is to use bullet points instead. One note about frustration, is it adjusts the threshold of the vector query on the database, allowing the system to use a broader range of tools. (If the screwdriver doesn't work, try the hammer) The other thing you missed is that the system is set up so that you can design your own agents. If you look in https://github.com/DataBassGit/AgentForge/tree/main/Examples/CustomAgents you can see how this is done. Custom agents load the Agent superclass and leverage those functions unless they are overridden, but you can also create new functions that are specific to that agent, or build a library of new functions and import them to your custom agents. We've also set the system up to use custom tools. There are tools prompts in the .agentforge folder that you can add to. Documentation on custom tools will be coming soon, but you can use any python library as a tool! Overall, great video and thank you for doing this. You hit all the major points and we thank you for your interest. Keep us posted on anything you build and feel free to reach out with any questions. |
Beta Was this translation helpful? Give feedback.
-
|
Hey all, This isn't directly related to AgentForge, but I've been going through the code for AceAF and I plan to make a video going over on a high level how each of the python based demos work. This is the diagram I've made that I plan to use as the guide when going through the AceAF project If you have any critiques or suggestions, please let me know |
Beta Was this translation helpful? Give feedback.
-
|
Of note, each layer of ACE sends its bus messages to a different tab in the Kivy Chat App. The messages are also in the logs, but we found this to be easier for debugging than going through a big log. Another key thing that needs to be pointed out. Chat messages go directly to layer 3. That agent controls things like memory, and made the most sense for a target. A major issue we ran into was that layers 4, 5, and 6 are overriding the response from layer 3. This shouldn't be happening, and is a control problem we didn't have time to resolve. Be sure to send Dave a link to your video. I'm sure he'd love to see your interpretation as well. We recorded a meeting where each member of the ACE team presented their work. I bet Dave would send you a copy if you asked for it. |
Beta Was this translation helpful? Give feedback.

Uh oh!
There was an error while loading. Please reload this page.
-
Hey all, appreciate the work done for this project. I've recently made a video doing a breakdown of the this project.
Would love to get any feedback or suggestions about the video
https://www.youtube.com/watch?v=bu2QAocymu0
Beta Was this translation helpful? Give feedback.
All reactions