I keep thinking about having a server where I’d run ai coding agents.
Why: always on for recurring tasks (that would use my discount tokens 🙃) + I can send messages to them from my phone + better isolation vs running things directly on my laptop (but still some scoped access)
Do people do this/what’s your setup?
(I’m vaguely aware of the Mac mini OpenClaw meme, but I’m not particularly interested in running local models, so doesn’t seem like the best fit?)
I wrote an HTTP server for my home computer that wraps Claude Code sessions running in containers (using the Agent SDK), and then I access it from my phone using Tailscale.
In retrospect, I’d advise something simpler: Just use a local user account to run it instead of containers. Mine works well enough that I can’t be bothered to rewrite it though.
Just wrapping Claude Code to have HTML output was 90% of the work, and adding things like scheduling would be pretty easy.
Ty for links! I feel like for Claude—native remote access is getting solid already, and they’re clearly working on it. I loathe Codex Web though and hope “happy” can provide a better UX for that (though on initial trial it just got stuck after first message 🙈).
Did you already have a desktop that you used for this, or did you do a new hw setup?
Over the last week I’ve been iterating on my SRS practice in @RoamResearch I present to you Rememo—a roam/memo (by @JoseBrowneX) compatible standalone app. It connects to your graph and lets you review your cards without waiting for Roam to load 🙃
I’ve historically used daily-page oriented SRS approach, I’ve implemented in roam-toolkit, but over time I’ve become more overwhelmed with the amount of things showing up on the daily page. I still use and endorse this approach for Programmable attention items.
But for actually memory oriented cards that don’t have a reason to be on a specific day page it’s less of a fit. So I’ve been experimenting with migrating those to roam/memo. But the fact that Roam takes forever to load is a big point of friction for doing incremental practice.
Hence Rememo—loads fast & allows you to sneak in reviews on the go
I keep thinking about having a server where I’d run ai coding agents. Why: always on for recurring tasks (that would use my discount tokens 🙃) + I can send messages to them from my phone + better isolation vs running things directly on my laptop (but still some scoped access)
Do people do this/what’s your setup? (I’m vaguely aware of the Mac mini OpenClaw meme, but I’m not particularly interested in running local models, so doesn’t seem like the best fit?)
I wrote an HTTP server for my home computer that wraps Claude Code sessions running in containers (using the Agent SDK), and then I access it from my phone using Tailscale.
In retrospect, I’d advise something simpler: Just use a local user account to run it instead of containers. Mine works well enough that I can’t be bothered to rewrite it though.
Just wrapping Claude Code to have HTML output was 90% of the work, and adding things like scheduling would be pretty easy.
https://happy.engineering/ seems like a more polished app that works similarly (and doesn’t require Tailscale).
Ty for links! I feel like for Claude—native remote access is getting solid already, and they’re clearly working on it. I loathe Codex Web though and hope “happy” can provide a better UX for that (though on initial trial it just got stuck after first message 🙈).
Did you already have a desktop that you used for this, or did you do a new hw setup?
+1 especially keen to hear from people who didn’t already start with a desktop setup, since I only have a laptop rn.
Over the last week I’ve been iterating on my SRS practice in @RoamResearch I present to you Rememo—a roam/memo (by @JoseBrowneX) compatible standalone app. It connects to your graph and lets you review your cards without waiting for Roam to load 🙃
Find it at https://stvad.github.io/rememo/
I’ve historically used daily-page oriented SRS approach, I’ve implemented in roam-toolkit, but over time I’ve become more overwhelmed with the amount of things showing up on the daily page. I still use and endorse this approach for Programmable attention items.
But for actually memory oriented cards that don’t have a reason to be on a specific day page it’s less of a fit. So I’ve been experimenting with migrating those to roam/memo. But the fact that Roam takes forever to load is a big point of friction for doing incremental practice.
Hence Rememo—loads fast & allows you to sneak in reviews on the go