[Misc] Should we store agent metadata in a SQL database or a GitHub repository?

[joshprk]

There was a discussion on 11/12/24 that brought possible considerations to not go along with splitting the monorepo (#304).

Currently, the AgentHub agents are stored on our own SQL database on the cloud, along with the files as blobs. The proposal to make AgentHub a separate repository was made with the thought that agent metadata would be stored on GitHub, which would contain static links to downloads of agents which are verified through cryptographic hashes.

We currently need to investigate where it is better to stick with a SQL database--which we will have to maintain--or use a GitHub database. An example of a GitHub database can be found here: https://github.com/NixOS/nixpkgs/blob/master/pkgs/applications/editors/bluej/default.nix

While the current master branch (v1) will likely not change from its current model, we should investigate which way would be better for the v2 branch.

In favor of SQL database:

• We control the server
• Agent developers may have an easier time simply uploading files to our servers

In favor of GitHub database:

• We do not need to worry about maintaining server infrastructure
• Agent developers will have to supply their own download sources, which will be verified using cryptographic hashes. This may be their own GitHub repository.

Please feel free to share your opinion on this issue below.

[dvsLick]

I propose allowing database mirrors, allowing anyone to host it, giving extra freedom to developers, and the possibility for support if any of these services are downed or stopped. for security, agents would be marked for their validity using their hash to warn of any possibly unsafe downloads for example, see opnsense