jack

jack

@j4ck.xyz

17 - working on projects on the open social web - follow for a mix of random posts (you signed up for it)

4.4K followers1.7K following44.1K posts

Pinned

pinned post! 📌 (now that they're a thing, technically).. follow me for sh!tposting, #tech stuff, #atproto & bsky updates / data 🔥 find my website at jglypt.net - although you can look at my bluesky username for that 😉 all my other socials are there, etc etc etc thanks for reading 🦋

just saw the iPhone Duo!! it’s so cool in person, you should all try it out everyone was crowded around the table lol, but even my family were interested in seeing it. cool that Apple can draw anyone in to wanna see it!

lol I noticed that the software had a bit of a bug when closing it, but other than that it all feels very polished my mum even said how normally you can see a crease with the Android ones, but you can’t here

bruh I’m in Wales and it’s raining… why is my phone getting this hot in the car 😭

it’s crazy how we invented time and also have no sense of it. a nap can be 15 minutes or 2 hours and you wake up having no idea how long it was

jack reposted

New iPad mini?? But at $699 is insane! that’s so expensive istg impressive specs, just sucks that the base model is so good bc the price is too high

handed off my agent the experimental Surf API and made it fix the Traitors feed. I’ve been having issues with filtering this one for months now, always was too political. Now it’s actually pulling in relevant posts! Check out "The Traitors" on Surf! Follow it below #thetraitors #celebritytraitors

surf.socialThe Traitors by jack on SurfSurf - Feed for the Traitors UK - #TheTraitors #Traitors #UKTheTraitors
jack@j4ck.xyz

we're entering an exciting time where @atproto.com is empowering people to have control over their feeds, @surf.social is making the most of this, and agents can speak the complex language for filtering posts down. users are finally gonna have the ultra-specific feeds they want for communities!

YESS

Radio Times@radiotimes.bsky.social

She might have been murdered in the Celebrity Traitors castle, but Miranda could be revived by the BBC 🌅 Read more: https://www.radiotimes.com/tv/comedy/bbc-miranda-revival-after-celebrity-traitors-newsupdate/

we're entering an exciting time where @atproto.com is empowering people to have control over their feeds, @surf.social is making the most of this, and agents can speak the complex language for filtering posts down. users are finally gonna have the ultra-specific feeds they want for communities!

a lot of talk of Tailscale security in the past few days, curious whether they’re gonna change some defaults for new Tailnets?

is YouTube broken for anyone else?? they added a new animation on the progress bar when using arrow keys to skip forward/backwards in a video but this bar sticks and all the other buttons disappear!?

SHEER HEART ATTACK IS SUCH A GOOD ALBUM!! #QueenBand #nowspinning

I'm starting to feel like I should stop using AI for as much stuff as I do. I think the point where it's setting up self-hosted software for me is too far for example. I think I need to go back to using dedicated chat apps for these so I read the outputs at the least and can somewhat learn along the

way. something I've not really been doing since things like Claude Code + auto mode have come in

hmm yk what I think I was too harsh on Zen. I just remember trying to use it over and over on my old laptop which was shite, even on Linux.. on Mac it's not too bad, pretty sure I had it as my main for a while, but I just remember the old Zen 💀

jack@j4ck.xyz

they should've finally open sourced Arc whilst announcing it's getting sunset don't shout at me with zen btw

do most Notion users use it in the browser or the desktop app? I've always primarily used it in the desktop app (other than Linux, ofc), but a lot of browsers have a default option of pinning Notion?

View full thread

It's not _super_ promising that all of this app's blog posts are to do with AI in the browser, but it's not actually that bad of an Arc clone, ngl! I think this and Ardia are the best Arc clones atm. not sure about the security of them so I won't be doing stuff like banking in them

gonna try and use decision models to help me mass unfollow people on here. will check them obviously, but god my following feed is hard to "follow"! (ik the mutuals feed exists)

I don’t use Final Cut Pro myself, but damn on macOS Golden Gate it looks like it was built for macOS from like 10 years ago. Half-assed Liquid Glass implementation…

did Google improve the Gemini 4 5hr usage limits? I'm on a "subsidised" Pro plan (student, in a Google fam) and I've been leaving it on for about 10-20m so far on high reasoning and this is my usage atm.

I think the weekly usage might have been hit from me using it in the Gemini app? unless it's just super extreme with how fast it depletes vs the 5hr limit

ughhhhh the videos of the iPhone Duo are all over my YT recommendations rn 😭 it looks so cool! I wanna check it out in person at some point 😭

damn, hermes in openwebui works pretty nicely! all running on my small OVHCloud VPS :)

ok I'm logging back into everything on Helium on my Mac mini until this is shipped in Firefox, then I'll use that still sad about Arc :(

Keith Cirkel@keithamus.social

Apropos of nothing, if you're considering switching from a recently deprecated browser and want to try Firefox, I just landed `Copy Tab URL` in nightly - it's not bound by default but a quick visit to `about:keyboard` will let you set it, and also take a glance at the other neat shortcuts we have.

this model is genuinely impressive! I really wish it’d run on lower end hardware tho ngl, that’d be sick hopefully someone takes what we can learn from this release and optimise it 🤞

Matt Kane@mk.gg

This is the best speech model I've heard by some margin, and it is seriously dangerous. If they release it like this, people are going to be falling in love with this model, and their choice of samples does not inspire confidence that they're taking the possibility of this seriously. block.xyz/dessa

jack reposted

Apropos of nothing, if you're considering switching from a recently deprecated browser and want to try Firefox, I just landed `Copy Tab URL` in nightly - it's not bound by default but a quick visit to `about:keyboard` will let you set it, and also take a glance at the other neat shortcuts we have.

it’s kinda fun having a cd collection, each time I go out for a drive to practice, I take a diff album each time to play 🔥

btw I've been finding this super useful! it syncs with your bluesky bookmarks every 10m or so looking for any new posts, and then archives them. then you can seacrh them properly, filter by links, age, etc.. an API too so you can connect your agents and they can help you go through your bookmarks

jack@j4ck.xyz

i bookmark a lot on here

may have finally started to solve this (WIP, don't host this publicly, only do it within a local network or Tailnet) github.com/j4ckxyz/recall I've managed to get Grok Bot + Meta Muse to dump their memories from my chats with them into this.. now they can see each others' memories!

github.comGitHub - j4ckxyz/recall: Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi.Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi. - j4ckxyz/recall
jack@j4ck.xyz

The Need for Portable Memory in LLMs: https://j4ck.xyz/2025/07/04/the-need-for-portable-memory.html

View full thread

other things I'm considering: - agents dumping a summary of each chat you've had with it here - agents dump all file uploads you put into them (?) but I'm not sure if I want to do this yet. I want this primarily to be a quick, lightweight shared store for agents' memories.

ideally it's quicker retrieval than ChatGPT or Claude's consumer apps which do a similar job. except, you are in direct control over the data stored here - it's all on your device.

may have finally started to solve this (WIP, don't host this publicly, only do it within a local network or Tailnet) github.com/j4ckxyz/recall I've managed to get Grok Bot + Meta Muse to dump their memories from my chats with them into this.. now they can see each others' memories!

github.comGitHub - j4ckxyz/recall: Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi.Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi. - j4ckxyz/recall
jack@j4ck.xyz

The Need for Portable Memory in LLMs: https://j4ck.xyz/2025/07/04/the-need-for-portable-memory.html

View full thread

after agents are hooked up, they can save their memories there, search memories, and read the most recent ones. I'm also working on a @obsidian.md -style web app which lets you search and explore your memories with agents in a clear UI. on-device embedding models on the backend will also

run when new memories are saved, so you can quickly search them. this whole github repo is a massive WIP. please don't host it anywhere public just in case there are security issues (I'm sure there are!). test it within a Tailnet :)

may have finally started to solve this (WIP, don't host this publicly, only do it within a local network or Tailnet) github.com/j4ckxyz/recall I've managed to get Grok Bot + Meta Muse to dump their memories from my chats with them into this.. now they can see each others' memories!

github.comGitHub - j4ckxyz/recall: Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi.Self-hosted memory of you, shared by every AI agent over MCP. One 10 MB Rust binary, encrypted, fuzzy search, runs on a Raspberry Pi. - j4ckxyz/recall
jack@j4ck.xyz

The Need for Portable Memory in LLMs: https://j4ck.xyz/2025/07/04/the-need-for-portable-memory.html

I think that memories with agents should be model-agnostic. with this tool, the idea is that you host it on whatever hardware you want (& have direct control over!). from that backend, you generate tokens which you can hand out to agents as well as the MCP URL. this lets them connect.

somehow Ollama managed to fully embrace closed, cloud models lmao what happened