Scott McCarty and Jorge Castro joined me this week to clear up something I kept getting wrong myself: Hummingbird is not one thing.
Two separate things launched at Summit. Red Hat hardened images went generally available, a set of container images plus an RPM channel. Fedora Hummingbird Linux was announced separately, a bootable image built on the same channel with a Fedora kernel. The distinction Scott landed on is that a container image will never boot. Add a kernel and it becomes a bootable container.
Other ground we covered:
Composition is roughly 95 percent Fedora Rawhide, about 1 percent from the major Fedora releases, and the rest pulled forward or wider when Fedora does not have it yet
Risk based versus compliance based patching, and why Scott will not call any of this “more secure,” only different
Why time to exploit collapsing is the actual story behind all of it
Jorge on Utah, Bluefin built on Hummingbird, and what it took to make it shippable
Switching an existing bootc home lab turns out to be a pointer change
Both guests closed with the same request, so I will pass it along: what would you actually use Fedora Hummingbird Linux for? They are reading, and they say they can pivot fast toward real use cases.
Example issue - Contributors connect their agents to a contributor relay which then sends them work items to complete. Here’s a typical hive submitted issue from Kyle Rankin, his Ornith model does the work and then goes through a review process.
Visualization of our journey - here’s a git history of the project starting from “manual mode” to partial autonomomy.
At around 4:20 James kicks off Utah development, it’s the blob in the bottom middle of the screen. On the right side of the visualization you see Dakota, which is a sibling Bluefin that is currently in “manual mode”. This little “burst” captures when everyone tossed their agents at it. This one will look way more interesting in a few months.
Note how many more red dots exist by the end. Those are all docs. The entire thing doesn’t work without extensive and up to date documentation. I spent most of July preparing for October.
I’m tuning in to every Fedora podcast, and it’s great to see how enthusiastic you all are about what you are doing. I often wish, though - especially with this episode - that we would return to conveying more information and fewer “sales pitch” on the podcast. I don’t mind hearing “for the container world”, “cloud native”, “zero CVE”, “AI”, “agentic” a few times - but really “few” is enough for me, then I want to know what this is about concretely: what and who is it for, where do I get it, how do I use it.; how do things work together (e.g. where are agents used and what for) . Positive example here: Scott’s explanation of two valid update strategies, putting “zero CVE” in its proper, well-motivated place without over-selling it.
I don’t expect the podcast to be a tutorial. But if I hadn’t known anything about Hummingbird before then I wouldn’t have known much more after either. Really, the summary above is at least as informative.
Again, it’s great you all are fans of what you’re doing, and we’re certainly all fans of Fedora. Be as enthusiastic as you are, just don’t assume everyone’s on board already; convince them with facts and information - yes, including fun facts
I didn’t really look around, there’s a bunch of these things and lots of interesting things happening in this space. I’ve been using the term ADLC for Agentic Development Lifecycle to describe these.
Hive spun out from KubeStellar and just got the votes to enter the CNCF Sandbox as it’s own project, so it will be a peer to bootc and podman. A bunch of Bluefin contributors ended up over there too so it just grew organically. Paperclip looks neat I’ll check it out!