Forgejo

a repo is just too big to put in a pds so they came up with this software called a knot and a knot hosts repositories on behalf of atmosphere identities so when signing up you can pick the tangled hosted knot or if you are hosting your own knot and have configured that knot such that it is willing to host repositories on behalf of your identity then when you sign into Tangled, it will create repositories in your knot, wherever that may be, on your behalf, through all of this cryptographic identity verification.

So if the knot is the repo, then the knot can be any, up to a point, right? You can make your knot be a Forgeo instance.

If the Forgeo instance is compatible with the knot API, it has to implement some particular features, but there's no reason you couldn't make Forgeo a knot host, I don't think.

Yeah.

Specifically, what Anat has to do is listen to this big global stream of events that are happening in the atmosphere, and in response to events from particular identities, allow particular actions.

So, Forgeo would need to listen to the atmosphere stream of events and then just do stuff in response to those identities.

I don't know anything about at Proto or at Proto, but is it literally just a single stream for everything?

Like...

Because of that reproducibility, I can just go in, modify the base image, and hit deploy.

And within a few minutes between Bootsy, Forgeo, and Ansible, I've got a brand new server that already has all the containers, because I just add the container file into the image.

Or the... sorry, the quadlet file into the image.

And then that quadlet is actually the one that does the, the pod man poll.

This whole Bootsy project started with the idea of I really don't want to have to log into my home lab every few days and run DNF update and reboot all my servers automatically.

And from that, I ended up rebuilding every one of the servers in my home lab.

I stood up a Forgeo Git server.

I spun up Forgeo runners.

For context, LinkedIn's own research found 41% of long-form posts were fully AI-generated.

Jake Saunders documented a home setup using Coolify, Forgeo, and OpenAI Codex that takes a single prompt through code, tests, continuous integration, deployment, and HTTPS, with Docker sandboxing between the LLM and the host.

Skeptics pointed out it isn't really self-hosted if it depends on Codex, local coding models still aren't good enough, and verification via test cases is, quote, the same agent validating its own assumptions.

Also worth reading today, DeepMind's retrospective on 15 years of game AI tracing the arc from Atari to SEMA 2.