Daily Journal 📓

Short dated entries, links, and microblog-style notes.


I wrote a reply in the atmosphere community forum about how pushing people to pick a PDS isn’t what we should be doing.

It’s long enough I want to keep a copy in my journal posts.

I need to write more on this topic, of a minimum viable community product, but more concisely than this.


The PDS is a commodity. Which is excellent. It’s cheap to run and we can think of it is a database. Databases have powered a lot of things for a long time, but they are boring tech. The starting point is absolutely not “which server should hold my account”. Pick any of them, they work everywhere, and it’s not the thing that gets you excited by itself: we’ll hand you a PDS wherever you start.

Products. Well built, designed, and articulated. With a point of view that fits a set of people. And a plan to tell people about it, and a plan to fund its operations.

I think a PDS and a member list and a couple of integrated apps and maybe a custom feed or three could be a product that is in and around groups and communities and so on.

EuroSky has gotten there and has more room to experiment. The PDS is there, and their portal shows apps, and Mu gives them a unique surface that they can showcase more things.

AtStore is one piece we can use to direct people. Even there, like EuroSky has done a great job not of saying “here are 1000 apps” but rather here are 8 or 12, that are useful, that are made in Europe, that you should be proud of.

And other than that: we do marketing. We write things. We share experiences. We cross promote. We try things. We try one thing in one culture and another in another and sometimes we’ll swap them.

We have a meetup not of our own (well, we can do a couple of those, too), but people inside every interest, who show up with the atmosphere in their pocket and a URL to get started with.

Yes, there will be a PDS buried somewhere at the bottom. But mostly, there will be people. Not platforms.


Umans has Cloud Agents, which are pre-paid hosted VMs similar to what exe.dev provides by default.

Right now they support pre-loading with OpenCode or OpenDesign, with Fusion and pi-web on the backlog.

I paid $10 for a month of a Small VM (2 vCPU · 4 GB RAM · 40 GB). There are $20 and $40 options as well.

After that, I immediately wanted to be running OpenCode2 which is in beta and what I’m running my laptop.

I used the agent itself to install, configure and troubleshoot this setup, and have been using it all day through a mobile web browser. I added it to my iPhone Home Screen as a web app so I could easily switch to web pages as well.

I’ve prepaid (not actually sure if it recurs automatically) so this gives me 30 days to set this up on my own on BringYourOwn.Computer OVH bare metal Proxmox.

I need to figure out a proper Ansible setup so that my agents can deploy their own sandboxed apps on Proxmox.

Exe.dev is great by the way. I got started down this path because they have an iOS app that I tried out last weekend with their Shelley agent. They also have a whole set of secure integrations like Discord, Slack, Tailscale etc.

So much stuff happening. I don’t NEED hosted compute anywhere as an individual, but of course that’s always been the case for personal / homelab style work.

And of course, next up, multiplayer.








BMC Next website thoughts:

Localfirst: HappyView running on desktop with SQLite. Keep my own atproto records synced & indexed. Run a search across that, locally. And other search endpoints? Also: private data.

Sync the SQLite db to website server? Extract links? Sync with Semble?


Manton has added independent PDS hosting for Micro.blog users:

“When new users opt to use Micro .blog’s PDS, their domain name becomes their Bluesky handle. They can then immediately sign into Bluesky with their own domain name without creating an account on Bluesky.”



Aargh. Finding Takeout Tools after I’ve gotten this Saved Places.json from Gmaps. Now going to get the CSV version so I can get separate lists.

“Google’s Takeout export for saved places is incomplete by design, but the gaps aren’t obvious until you’ve already gone through the process ”




Cloudflare launching wallets for agents with x402 support.

They’re doing the handles landgrab at cloudflare.pay - but looks like it uses subdomains, so could compose with atproto.



I subscribed to the Umans founders plan at the end of June. Below is the email they sent canceling it as of August 1st.

I’m happy with the usage I had, which wasn’t 24/7 agents, and two weeks of travel. I guess I’ll try pay-per-token for a while, and of course our local AI compute ZAI project continues.


We built the founding plans around a simple picture: one developer, working intensively, free from token anxiety. That picture is gone. Agentic usage exploded. Many of you now run fleets of agents 24/7 across multiple machines, and you expect more models, availability in every time zone, lower latency, higher throughput. We serve thousands of developers and several trillion tokens a month, as a small self-funded team.

A flat unlimited plan cannot carry that world. Not because the usage is wrong, but because the only way to keep such a plan alive is to slowly fill it with throttles, hidden limits and fair-use enforcement. We would spend our energy policing usage instead of doing what we are actually good at: running large open-weight models fast, reliably and at low cost.

So we are choosing the clean reset over the slow degradation.

What this means for you, as a Founding Max subscriber

  • Your plan ends around August 1, 2026 00:00 UTC. Your plan API keys stop working at that time.
  • As an annual subscriber, you will receive a prorated refund of the unused remainder, to your original payment method. Refunds land within 5-10 business days, nothing to do on your side.
  • 1 perk ($10 credit) is on its way to your account, redeemable toward anything on Umans: top-up or cloud-agent time.
  • Concurrency packs (if any) end with the plan, no further charges. Cloud-agent slots are not affected.
  • Founding subscribers keep free access to future experimental windows for new models, and priority access to whatever plan comes next. A survey is coming soon to design it together.

Lots of people have been asking about the atproto alternative to Buzz, the Jack Dorsey-supported Nostr chat for agents & humans to chat in.

The answer is freeq, which is atproto + IRCv3 extensions, including CRDTs with Automerge, QUIC, iroh. Follow @freeq.at.