@microcosm.blue
Open-source APIs to kick-start your next app on ATProto. Community-supported infrastructure ready for production. relay.fire.hose.cam constellation.microcosm.blue slingshot.microcosm.blue more: https://microcosm.blue https://tangled.org/@microcosm.blue
The blacksky.community web application now uses our own API servers to load posts, timelines and profiles.
When data is missing (accounts on the edges of the network, outside of our primary community, etc) we leverage @microcosm.blue
slingshot also has a new `hydrateQueryResponse` thing, which can inject record contents into the response from constellation in one go
and there’s a full manyToMany query ready to ship for constellation, which would let you inject all actual vote records if you wanted to verify CIDs!
constellation has getManyToMany counts and picks up anything that looks like a uri (not just at-uris), so if you put uri-looking things on the vote field you might be able to just use it directly (even silly uri-like stuff like `vote:yes` might even get picked up!)
/plugin marketplace add tangled.org/zzstoatzz.io...
/plugin install protopack
build <cool atproto app>
us-west.plc.wtf has completed backfill after about 72h. you can use it!
added a psuedo-comments-section to my site, which simply lists posts that link to a page (through @microcosm.blue for bsky, and searching over my own instance for fedi) :3
Okay! this probably should have been a leaflet! thanks for reading this far!
you can support more work like this from microcosm: github.com/sponsors/uni... / ko-fi.com/bad_example
or/especially, if you want to fund independent work around PLC specifically, DM me! there's a lot we can do.
What about the /export endpoint(s)?
it's unfortunate they're not in the reference read-replica!
for a public replica to be credible, its contents should be publicly auditable. if bluesky doesn't intend to add/accept them, we'll fork + do it ourselves. nobody uses it right now, but it is important.
What about being PLC-migration-ready?
instead of continuously running an expensive database cluster *before* it's ever needed, i'd love to build replica-to-authority migration tools into Allegedly, that can rapidly convert a mirror into a cluster like that if/when needed instead.
and in fact, @ptr.pet has been making amazing progress on a new *even-lighter-weight* plc read-replica implementation in Allegedly! more on this soon!
and of course we'll get back to updating public op archives again, so backfilling any read-replica can be a sub-1hr task instead of a 24h+ one
and Allegedly is not out of the picture! four months ago we shipped `wrap` mode, which makes it a reverse-proxy for *any* PLC mirror implementation, replacing nginx or caddy, with PLC-specific bonus features like forwarding writes upstream to plc.directory! (automatic tls certs included, caddy fans)
embedded sqlite means plc.wtf can let go of the expensive managed postgres, and afford to run more replica instances (in more places!) instead. to start, we'll keep one in finland and add one in `us-west`.
automated deployment details here (ansible, the horror): github.com/at-microcosm...
awesome things about the new reference read-replica:
- correct op validation and nullification!
- uses the newer more-robust sequence-cursor-polling and `/export/stream`!
- can use a local embedded sqlite database!
- very easy to set up and run!
so that's where Allegedly's mirror is at. until now!
the new reference read-replica server (by @retr0.id!) makes different choices, which don't all serve every original goal of plc.wtf (namely: it's not a migration-ready reference PLC server). but i think it is a *great* fit for plc.wtf today.
and lastly, the cost to run the managed postgresql database from UpCloud actually exceeds their sponsorship amount. big production-grade databases are expensive!
- nullified ops don't always update their `nullified` state correctly
- behind on re-synchronizing the directory since the last spam removal
- no weekly bundles published since end of Nov 2025 (+ no spam removal)
- never shipped the `zip` command to diff two PLCs to detect tampering
with sustainability uncertain and momentum waning, i refocused my main efforts back to other microcosm projects. Allegedly's mirror mode has kept plc.wtf going, but it's a bit behind on maintenance:
- runs an old version of plc reference server
- date-cursor-polls plc's /export, which can miss ops
at that point i had spent weeks building PLC tooling and the migration-ready replica mode for Allegedly, but hadn't focused on public communications to translate any of that effort into support for microcosm.
the posts and announcements i *did* make: didn't translate into new support.
the (eventual) Swiss Association is really really good news for the future of PLC and the atmosphere at large! but even though being prepared for adversarial migration will *remain important* after control is handed over, the announcement did unfortunately dampen the momentum behind plc.wtf.
this all took quite a bit of effort and coordination to pull together. Bluesky dropped their Swiss Association announcement (timeline/concrete details still TBD), right when that initial leaflet i put together was in its final draft.
docs.bsky.app/blog/plc-dir...
the design was informed through conversations with other players in the ecosystem, including folks from Blacksky, Bridgy Fed, Tangled, the AT Protocol Community Fund, and selfhosted.social.
its managed postgres database is sponsored by UpCloud, ready to scale into a serious-scale HA cluster.
the allegedly mirror-mode replica at plc.wtf is meant to be a technically-plausible candidate for PLC authority successor in the event of adversarial PLC migration. It's meant for us to have *at least* one option actually ready!
(again, updates.microcosm.blue/3lz7nwvh4zc2u)
and not to get too off track, but it also
3. supports the `/export` endpoint
which was a big motivator, since none of the other existing PLC mirror projects supported it, and it's hard to audit without.
losing it is a little sad, even if it wasn't used! i'm hopeful that support will be added.
running the full-blown reference PLC inside is very intentional! the motivations laid out here: updates.microcosm.blue/3lz7nwvh4zc2u
an allegedly mirror-mode replica is
1. fully compliant (it's running the plc.directory code)
2. ready to become a PLC authority at the push of a button
...and copying new ops directly into its database by polling an upstream PLC server.
that is: allegedly's mirror is *not* a lightweigth PLC replica, it's a full-blown PLC directory. (sorry for the llm-ism, i don't know other words to underscore this point)
there are a few reasons for the change, but first, a little read-replica context. Bluesky's own PLC read-replica service was released one week ago, but Allegedly's mirror mode turns five months old today!
mirror mode *runs the reference typescript PLC server inside*, delegating read requests to it–
the change is otherwise good news! plc.wtf will run the reference plc-replica server, instead of Allegedly in mirror mode.
like the new us-west.plc.wtf instance, it will still have Allegedly's experimental endpoints enabled (you can send PLC writes directly to it! free yourself from plc.directory!)
deprecation notice:
the `/export` endpoint at plc.wtf will be removed in the next update.
metrics show no current use in the expected patterns (backfill or polling) so no impact expected. too many details follow :)
eu.plc.wtf (second microcosm reference replica instance, finland this time)
the Allegedly wrapper also has experimental op forwarding enabled, of course. (experimental.us-west.plc.wtf)
go-didplc/cmd/plc-replica 🤝 Allegedly. bffs.
oh and as the domain implies, this instance is in the `us-west` region.
thanks as always @upcloud.com for sponsoring plc.wtf !!!
