[–] 3 points 2 months ago (1 child)

Swap is rarely hit, especially if set up with zram.

This is not a good thing btw. Any unused anonymous page takes up space that could instead be used for file-backed pages that make your system faster.

Multiple swap tiers beyond zram/zswap

Swap is not tiered storage!

Priorities control order of preference, not tiers. If you run out of space on a higher priority, it will not move that swap's data to a lower priority swap. It will keep all of it exactly where it is and new data will hit the lower prio swap instead, no matter how hot it is.

Intel Optane

Cool tech but it's dead and was quite niche even when it was alive.

zram write-back device

Not a thing you actually want to use for swap. It's not an automatic writeback that is integrated into the Linux MM in any way. (Probably has some use-case for non-swap zram purposes though.)

Large HDD swap partition used as a third tier swap disk

This makes no sense at all unless you are extremely space-constrained on the NVMe and absolutely must not OOM – even if progress stalls to an absolute crawl.

swapped pages can be moved around by the user (or a tool), by turning swap devices off and on.

This is neither feasible nor desirable. You don't have enough granularity to do anything useful by doing so.
Even if you had, it'd work against the MM because it resurrects pages as "hot" that have been cold for a long time.
In any situation where swap is important, making the kernel think cold pages are hot is the very last thing you want.


I too wish it were but tiered/transcedental memory is not a thing in Linux and these hacks do not change that fact; they merely look similar if you don't look close enough.

I cannot think of a single use-case where this would be preferable to a decently sized physical swap with zswap XOR just zram swap (if physical swap is infeasible).

  • source
  • parent
  • context
  • [–] 2 points 2 months ago (1 child)

    This testing compares apples to oranges. Differently sized swap and quite obviously different workloads. Given how very much compress ratios depend on the specific data that is compressed, this experimental setup cannot produce valid results.

    This is exacerbated by your swap being full. Zswap is more of a cache in front of your actual swap; it requires physical swap to function. If the physical swap is full, it cannot receive more data! Zswap not doing very much when the swap is full is totally expected behaviour because it simply doesn't. The solution to that is to size your swap sensibly. (Admittedly, this does not appear to be documented clearly.)

    zswap uses the exact same allocator as zram these days (zsmalloc). It'd be very surprising if it had different space efficiency characteristics. It's not impossible (could be a bug) but claiming so would require quite certain evidence IMHO.

    RE: LRU inversion: the problem with not caring about it is that it's not a visible problem until it very suddenly is. Your system will not gradually degrade but very suddenly and unpredictably hit a wall that it cannot get itself over.

  • source
  • parent
  • context
  • [–] 2 points 2 months ago

    Inconsistent time before sleep is likely the kernel waiting for IO to sync to disk.

    Not going to sleep at all implies a serious issue that prevented the kernel from doing so. Check your logs!

    Suspend reliability greatly depends on the firmware of your specific device as it controls the wakeup.

    On my Framework 16 for instance, it is very reliable these days. I can't remember the last time it woke up unexpectedly and I'm not sure whether it happened even once since the firmware update that fixed a major oversight in keyboard wakeup (waking when a key was pressed through a closed lid).

  • source
  • [–] 1 point 2 months ago

    In Hyprland, how it works is that you define the screen lock command (e.g. swaylock) in hypridle and that gets wired into logind somehow such that you can loginctl lock-session.

    I then (also in hypridle) define a before sleep hook that locks the session.

    I believe the proper way to do this would be to run your lockscreen as a systemd unit that is WantedBy and Before suspend.target. This would in theory take care of making it idempotent and reliably show the lockscreen before initiating suspend.
    A friend of mine has this sort of setup and, while it is reliable, the lockscreen is still racy and sometimes only triggers after suspend. Might be the lockscreen unit not signalling its readiness to systemd properly though.

  • source
  • [–] 2 points 5 months ago (1 child)

    Thank you!

    I've found the Seedstudio thing after posting this too and it looks like the thing I'd be looking for!

    What's your experience w.r.t. coverage?
    Obviously that highly depends on where exactly you are – you certainly aren't going to have coverage in the outback – but I'm mostly concerned with places where people actually go and would take my bag/laptop/bicycle to. 'Stralia is going to generally be quite different from Germany too of course but it would be a good reference point from which I could extrapolate.

  • source
  • parent
  • context
  • [–] 4 points 6 months ago (1 child)

    Well, they have – I think. When you download an edited image, it supposedly downloads an image with edits applied. The original is optionally available too.

    If you download the edited image, this is effectively equivalent to the status quo of image editing.

  • source
  • parent
  • context
  • [–] 2 points 6 months ago

    The issue is not the instruction set of the processors. That's actually quite well standardised with ARM (albeit unfree) and there is plenty of generic support for it because of that.

    The issue is all the "peripheral" devices such as WiFi, WWAN, display etc. that are wired up in extremely bespoke device-specific ways. They are usually implemented in vendor kernels with millions of lines of divergence to mainline at best and/or proprietary blobs at worst.

    Changing the ISA from one well-supported closed standard to a less well-supported open one will not change that issue one bit.

  • source
  • parent
  • context
  • [–] 2 points 6 months ago (3 children)

    Are there any (ideally waterproof) compact devices with long battery life (months~years)?

    On the website I only found a long list of supported devices with brand name search and protocol type. grep showed no LoRaWAN devices though?

    My use-case is theft tracking. I only need the device to be able to locate itself after a theft actually occurred and I request it remotely. (Perhaps also periodically with very low frequency.)

  • source
  • parent
  • context
  •  

    Next chapter in the Kagi Assistant experience

    With this release, we're introducing a new sidebar, designed to streamline your workflow and keep everything you need within reach. This is just the beginning - the new sidebar lays the foundation for even more powerful features to come.

    We're also rolling out Claude 3.7 Sonnet with extended thinking to tackle even more complex challenges.

    As always, we’d love to hear your thoughts - join the conversation on Discord or share your feedback through our forum at https://kagifeedback.org/

    Kagi, lost in translation, in Tokyo!

    From April 21 to 25, a dedicated team from Kagi including CEO, Vlad, will travel to Tokyo 🇯🇵 to meet with companies interested in our advanced translation solutions. If you are based in Japan and your company is interested, please get in touch with Gillian.

    We'd also love to meet our local user community while we're there. If you're based in Tokyo, we'd love to organise a dinner and connect in person. Looking forward to great conversations and good food with fellow Kagi fans!

    Please contact Gillian at gillian@kagi.com to arrange.

    Jobs, jobs, jobs

    We are looking for a Head of Business Development and a Technical Architect (basically CTO equivalent in Kagi world).

    Head out to Kagi hiring to apply and check out other open positions.

    Improvements and bug fixes

    Search

    Based on the community's feedback, we’ve added a setting to always hide AI-generated images by default (suggested by #5998 @keyboardJones). You can enable it on Settings>Search>AI.

    Kagi Assistant

    Kagi Translate

    • Add token and other URL parameters to make it easier to use translate in private mode aminomancer
    • Fix translation of Catalan traditional time turly
    • IP URLs not being detected Hanbyeol
    • Inaccurate one-word translations nichu42 JW
    • Copied translation incorrectly removes line breaks LunarWatcher
    • Sync translation theme and language from Kagi Search frin
    • Add section in language picker for most commonly used languages Marcin
    • Fix some words being incorrectly identified as URLs nichu42
    • Add support for markdown goulot_situez
    • Clicking a link from a translated website doesn't redirect to translation of that page Rexios
    • Alternative translations getting cut off Peter
    • Page flashes in light mode theme for a second before turning dark nichu42
    • Source textarea steals focus while user is typing elsewhere tuesday
    • Selection of output text is unreliable on Firefox bkrein
    • Load alternative translations for user-selected text in the output box nichu42
    • Text alignment: When user hovers on input text, highlight the corrisponding part in the output RoxyRoxyRoxy

    Orion browser release: smoother, smarter, and more secure

    This week's Orion release brings big upgrades:

    🔐 3rd party Password Managers & Adblock
    We have updated support for uBlock Origin, 1Password, and Bitwarden extensions.

    👓 User Interface
    Several improvements concern vertical tabs, the sidebar, and favicons.

    📺 Youtube and PiP
    Several improvements enhance the use of Youtube and the Picture-in-Picture feature.

    🪲 Crash fixes & Stability
    Bye bye crashes on launch, instability if the browser stays open for a long time, and freezes.

    For all the details, please refer to Orion's changelog.

    Kagi on Socials

    This week's featured social media post:

    Tag our accounts or use #Kagi when mentioning us in your posts!

    Kagi in the News

    • This TechCrunch article mentions Kagi as an option for users who prioritize not just a better search experience but also a more privacy-focused search.
    • OMG! Ubuntu covered Orion's expansion to Linux.
    • t3n, a German tech magazine, also featured Kagi, highlighting it as a privacy-friendly and ad-free alternative. Read with Kagi Translate here.

    Kagi Community Reviews

    We are deeply grateful to everyone who takes the time to share their experiences, tips, and feedback about Kagi, helping to spread the word and inspire others to discover its value. Here are some recent reviews to highlight:

    Have a Kagi review you'd like to feature? Please share it and ping us on any of our various socials when you do!

    The Kagi Way

    We partnered with artist Chaz Hutton on a graphic that illustrates how Kagi empowers you to search without the noise, trackers and distractions. Help us share it widely!

     

    We're thrilled to announce that development of Orion Browser for Linux has officially started! Our team is working hard to bring the same speed, privacy, and innovation that Mac users love to the Linux platform.

    This is an ambitious project that we expect will take approximately one year to complete. Our target is to achieve feature parity with the current macOS version by March 2026.

    Want to stay updated on Orion for Linux?
    Register here to receive news and early access opportunities throughout the development year.

     

    Kagi Assistant adds Sonnet 3.7 and the preview of multi-step reasoning assistant called Ki

    We're happy to unveil the latest updates to our assistant experience. This release brings a smoother, smarter, and more intuitive interface designed to make your interactions simpler. Try it out and feel the difference!

    We also added Claude 3.7 Sonnet to our model lineup! It's now powering our !code assistant and plays a key role in our advanced multi-step reasoning model, Kii. Currently in early testing, Ki is available exclusively to our Discord community members - join now to get early access (Ultimate account users only)!

    And today OpenAI released GPT4.5 - we have already benchmarked it. Check Kagi LLM benchmark.

    TikTok video search

    Video search just got even sharper—you can now filter specifically for Tiktok videos. Find exactly what you're looking for, faster.

    We are hiring a Flutter developer!

    We just opened a position for a skilled Flutter Developer. Apply now or send someone our way.

    Improvements and bug fixes

    Search

    Assistant

    Translate

    Try Kagi Translate at https://translate.kagi.com/

    • If on 'manual' translate mode, do a translate when the user changes target language #6037 @Thibaultmol
    • Improve Korean Localization #5305_334, #5305_329 @Hanbyeol
    • Don't show alternative translations in JP, if the only thing that's changed from original is the pronunciation being added. #5305_328 @frin
    • JP romanization is sometimes incorrect/overly formal #5305_328 #5305_323 @frin
    • German (Switzerland) translations using the wrong "ss" #5305_314 @psy-q
    • Translated text sometimes put into quotation marks #5305_306 @nichu42
    • Disable/export/delete translation history #5305_308 @ajimix
    • CTRL+A should only select the translated text, not whole page #5305_316 @RoxyRoxyRoxy
    • Add Hawaiian to list of supported languages #5305_320 @Bradh
    • Change URL params when translation is completed #5305_323 @frin
    • While proofreading hindi, strike/correct whole diacritic / ligature. @aldehyde.8578
    • Document translation: translate your .doc(x), .txt and .csv documents
    • Alternative translations: along the main translation, show multiple translation options
    • Word insights: fine-tune individual words or phrases
    • https://translate.kagi.com/iso_code now redirects to the main page, with the to language being set to [iso_code].
    • Make tooltips appear immediately, adjust style
    • Dynamic text size depending on length in input/output box
    • Increase max size of request header to allow longer romanization requests
    • [Forgot to credit last time] Add translation context field #5305_253 @Kate-Karui

    Kagi on Socials

    This week's featured social media post:

    Tag our accounts or use #Kagi when mentioning us in your posts!

    Kagi in the News

    Interested in covering Kagi on your outlet, newsletter or podcast? Hit us up! Our team is very approachable and we welcome any opportunity to engage with various communities about our latest features.

     

    Announcing Kagi Privacy Pass

    Kagi now supports Privacy Pass, an IETF-standardized protocol ensuring your searches are technically unlinkable to your account.

    Read all about why this matters and the details of implementation in our announcement blog post!

    Privacy Pass support is provided:

    • Natively for Orion browser users (macOS/iOS/iPadOS). On iOS, make sure to have version 1.3.17 and above (expected to roll out globally today) and update your macOS Orion to version 0.99.131.
    • Natively through Kagi App for Android (make sure to have version 0.29, expected to roll out globally today)
    • Browser extension for Firefox or Chrome

    A few important details:

    • Privacy Pass implementation is fully open-sourced for transparency and community collaboration
    • Kagi Privacy Pass is available for the Professional, Ultimate, Family, and Team plans.
    • Limitations:
      • It's not available for Trial/Starter plans
      • Privacy Pass mode disables account-specific features, like domain personalisaton (as we do not know which user is searching)
      • Privacy Pass mode is supported for Kagi Search in this initial phase, we will add support other services in the coming weeks. Check the blog post FAQ section for more details!

    Kagi is redefining privacy in search. Try it now!

    Kagi Tor Onion service

    Access Kagi securely and anonymously via our new Tor Onion Service.

    You can now access Kagi directly through the Tor network using our dedicated onion address:

    kagi2pv5bdcxxqla5itjzje2cgdccuwept5ub6patvmvn3qgmgjd6vid.onion

    See more information about Kagi Tor service in our documentation.

    An updated comprehensive privacy policy

    We're also excited to share our updated privacy policy, with simplified language and designed to clearly reflect our strong commitment to protecting your privacy.

    Improvements and bug fixes

    Search

    Kagi Assistant

    • Kagi Assistant now supports O3 Mini, Gemini 2.0 Flash, and R1 Distill Llama 70B!
    • Announce new results from assistant using ARIA Live #5996 @fastfinge
    • Allow to hide thought process for reasoning LLMs #6025 @KamilKurde
    • Add o3-mini model to Kagi Assistant #6135 @RegChien
    • Formatting issues in Reasoning for R1 in Assistant #6148 @hansihe
    • Assistant should wait with naming a thread until it's finished if not enough info #5828 @Thibaultmol
    • Space missing below reference list in Assistant downloaded transcript #5833 @thomasjsn
    • Allow for disabling the CTRL + SHIFT + Backspace shortcut #5550 @4P5mc
    • Images cannot be pasted into the Assistant on WebKit browsers #6193 @laiz
    • Info panel not visible for new Assistant UI #6143 @azdanov
    • The assistant prompt "think about </think> tag" breaks auto-hiding of reasoning text #6181 @ssg

    Kagi Android App

    Privacy Pass support! You can add the Privacy Pass shortcut by holding the Kagi Android icon.

    • The app returns to the main screen every time it re-renders (e.g., entering/leaving split screen) #5875 @Philippe_Choquette
    • Assistant via Android app doesn't work with multi file uploads #5959 @Thibaultmol

    Kagi on Socials

    Here is this week's featured social media post:

    Tag our accounts or use #Kagi when mentioning us in your posts!

    Kagi in the News

    Orion tops Apple's App Store's list of superpowered internet browsers "to seriously level up your web browsing"!

    And Android Police published an article about Kagi's new fair pricing model: "This ethical search engine will return your subscription money if you don't use it." The Verge also covered the news.

    view more: next ›