Existing Code Modification, & Multi Model Intelligence

[Toocky]

Love the tool and can see it growing into something really powerful.

We have built a similar tool though much simpler for our own internal code updates and I can see 2 things that may be of use for this project.

1 - Modifying existing code, from what I can see this just builds new code so you probably already have this on the cards, it would be great to modify an exiting project, we currently have an analyse task in our agent that reads the existing code base and builds out the framework or in your case it would populate the DB with everything and then we work on the new requirements with the "Product Owner" before it cycles through the coding process and finishes. In this case it would be great to define an existing codebase, branch it and then work off that branch for the updates.

2 - We also ran into issues especially with GPT where it would struggle with code and just cycle trying to build the right thing and get stuck i.e. your "TODO" issue we have found a pretty successful way around this and have broken our agent into 2 Groups, "Coordinator" + "Consultants", where the Coordinator is the primary intelligence behind the solution and each round of queries it asks a group of consultants in our case that is GPT-4o, Gemini 1.5 Pro, and Claude3 (though that will now change to Claude3.5) the coordinator then takes the 3 results provides it to the coordinator along with the original requirements and it builds the best solution from the consultants answers. We have maintained GPT-4o as the Coordinator as it consistently puts the code in the correct formatting (even if its own answer originally was poor), we consistently saw formatting issues from other models. That said this can significantly increase cost but on the flip side it drastically reduces after code care, and reduces cycles.

We have been lurking around the community for a while and built one of the First AutoGPT UI's a year ago https://github.com/neuronic-ai/autogpt-ui

But for a while we have been focused on our API service and are currently in a beta phase https://apipie.ai/ we are currently in the process of implementing functions/Tools support for our models.

As I mentioned we are in Beta so we have some credit we can give out if anyone's interested in testing or using APIpie.ai in this tool to support different models then hit us up on https://discord.gg/fnVjanuCzB