Booting into Debian with most devices fully functional is great.
What I'd like to know is what software runs adequately under it in 4 GB RAM. Web browsing should definitely be possible, but I suppose it's limited to very few tabs. Some very lightweight DE could likely make it more usable. Running something like WezTerm + tmux as the DE could be even more economical, leaving some room for e.g. development tools.
Browsers and anything electron-based are your enemy.
Firefox is actually pretty good in low-memory situations, silently discarding tabs when under memory pressure, but the main benefit comes from being able to run proper adblocking. Chromium-based browsers just can't compete these days.
Otherwise, a bog standard Gnome-based Debian Trixie desktop should be pretty doable. I'm currently using an 8 GB machine with 3.7 GB RAM free - Firefox, evolution, gnome-calendar, and gnome-software are the only apps that using more than 100 MB, and none of them are obligatory.
I haven't carefully profiled memory use, but in my experience, Chromium is so much more performant than Firefox on ARM devices that any difference isn't worth it. If you're using a lot of tabs, it might lean in Firefox's favor, but overall performance so strongly favors Chromium that I've given up trying to use Firefox on anything but my high performance machines. I'm not sure where the performance delta is coming from, but the whole UI and JavaScript anything are much more responsive on e.g. A73 cores with 4GB RAM.
it's probably the "you only notice when it doesn't work" situation, but my experience with firefox on ram limit has been a lot about tabs forgetting the url in them
as in, I click "open in new tab", some time later I switch to them... only to get hit with "new tab", even though a moment ago it displayed tab name and I could right click -> bookmark to preemptively copy the address
Try the "Auto tab discard" extension. It allows me to have hundreds of tabs "open" and (in combination with Tree Style Tabs) largely blur the line between "browser sessions" and "bookmarks".
Yeah, agreed. The built-in tab discarder only kicks in when there's actual memory pressure, so can sometimes be a bit precarious. Auto tab discard happens way before that, so tends not to be affected in the same way. I guess it uses more i/o in total, but it's not noticeable on a system with a fast-ish SSD.
It can still be a bit iffy when memory's really tight, but even then a simple tab reload is usually enough to fix things.
Haven't had that happen, but what I have had happen is that I open in a new tab, and it just displays this spinner in the middle of the window while on the tab. It never loads. I take the URL from the address bar and drag it into yet a newer tab and there it loads. Then I close the original new tab. Sometimes I gotta do that a few times for the thing to load. I tend to open in new tab with middle click, if it makes a difference.
>[Firefox runs] proper adblocking. Chromium-based browsers just can't compete
Any familiarity with Safari and blocking performance? uBlock Origin Lite is a simple option, AdGuard can do more (injection?) though uBO feels more trustworthy still…
Seconding ad-blocking. I have a low-end phone (4GB ram, and a mediatek processor from 2018), and setting up DNS-based ad-blocking made a lot of sites go from unusable to usable.
I was thinking about Brave too while reading this thread. I’m not on a memory constrained system exactly but Brave seems to be tons snappier due to its as blocking. I wonder too if Brave is a case where you can pull it off and still take advantage of chromium based.
I run Ubuntu on my Chromebook. It's what I'm using to read this now. Web browsing works just fine. There's a limit to how many sites I can have open at a time, but since I regularly view sites that use over 1 GB of ram in Chromium, that's the case on all my machines.
Most of the games I play run in 4 GB, but since my Chromebook only has 32 GB of storage, There are some I can't install and I generally only have four or five installed at any given time.
Some time ago I got myself a similarly priced x86-64 Windows tablet on Amazon (Celeron N4020 + 4 GB RAM). I installed Linux Mint on it with a slightly customized kernel (some extra quirks were needed).
I connected an old SSD to it with a SATA2USB adapter, and I use it as a home file server and HTPC. It has a micro HDMI output, and it is connected to my TV. During the day it is playing music non-stop, in the evening it is playing some movies. It has no problem with high bitrate full HD movies, the CPU doesn't even break a sweat. I think it could also play 4K content, if I had any.
(Previously I used a Mac Mini with VLC for this for a few years, but I'm happier with my current setup, it's more stable)
I have 8GB, which I've had since 2012. Never had a problem - I run a lean Nixos with just xmonad and dmenu, chrome, emacs, and about a dozen open pdfs and video tutorials.
Y’all are embarrassing me with Lubuntu and Chrome on a 2013 Dell with 16GB and an SSD. Not fast enough for all I need to do but covers 80% of my needs. It’s my road laptop and the home desktop handles the rest.
Mine is a 2012 mac book air, I've replaced the battery early this year, and last month I upgraded the ssd to 1tb. I expect this computer will be a family heirloom after the apocalypse.
I'm curious, how much to replace the battery? I bought my kid a used 2015 MacBook Air like 5 years ago and it needs a new battery, otherwise it's totally fine.
having many tabs is perfectly fine - it's having many *youtube* tabs is troublesome
main trouble to me has been caused by unity games - those are the big ram devourers, even most basic 2D ones (I still don't understand how that happens, why such regression since KSP days)
and plenty of 2D games work perfectly fine (devs really overestimate minimal requirements)
> main trouble to me has been caused by unity games
Generally it's probably just bad optimization. But that only gets you so far because Unity's asset streaming is designed to work with level-based games. It will only let you unload assets if you package them per-level and then swap them in and out at load screens between levels. Absolutely useless for games like KSP.
Frankly if you don't need a web browser (or electron), what WOULD require that much memory? Video and photo editing maybe? Postgres? Recompiling the world?
I first started recompiling the world with 64MB of ram, kind of funny how far we've come on hardware and made software gobble up the gains with very little to show for it.
Since it seems AI is pretty good at reverse-engineering stuff like this, is there any educational material on how to use it for that purpose? Seems like it could really help port things like postmarketOS to new devices (and improve support on existing ones)?
I have claude code hooked up to deepseek, I hooked up my spare cheapo android tablet, installed adb and fastbook with my package manager and asked the AI to jailbreak the tablet.
It discovered the tablet was running a unisoc t606, found a CVE from a couple years ago, and unlocked the bootloader for me. I was the meat puppet holding the "volume up" button and plugging in the usb cable a bunch of times. Like most of my experiences with this stuff, it was pretty eerie.
Next step for me is to attempt mainline linux, there seems to be some postmarketOS devs playing with it. We've probed most of the tablet's hardware except the exact display.
I have some experience on this and could make an article if you are interested.
The key is to have downstream sources and be very very conservative with the AI, slowly build step by step.
You also have to know C and have a spider sense of what's acceptable or not.
Another key is to ask for approval before editing any source with a patch of what it intends to do. This way you can judge what it wants to do and ask for a double check of the patch. Go quality over quantity.
This isn't web frontend with Tailwind, you have to be very strict and somewhat knowledgeable. Nobody can use AI to write kernel code without some good low level and engineering knowledge.
Please do write an article! I've wanted to get into reusing old android hardware for quite some time now, but never knew where to look for good instructions to get started. Especially PostmarketOS seems very interesting, but rather underdocumented in some places.
I will then, didn't know it would be interesting for other people.
As for PostmarketOS, I've built my own tooling scripts around it to make it easier to build patches, debug hex variables, switch between downstream/mainline and rebuilding everything with a single command. (Unrelased yet though).
I find their tooling okay for a release for end-users but a bit clunky for debugging.
I completely agree, this is not the place to let AI blindly edit kernel code. The useful approach is to use it conservatively: understand the error, compare against downstream sources, propose a small patch, review it, test it, and then move one step further.
I’d be happy to work together on an article or guidance document, where to start, how to approach debugging, what to never let AI touch blindly, and how to build confidence step by step. That could help others avoid a lot of mistakes and maybe give a second chance to other devices.
It helps for fuzzing, maintaining and is actually a great help for seniors, maybe not for the ones who don't care for the project and publish slop.
It could now actually help a lot in some ways not just coding though but things surrounding project management.
No it doesn't. It helps lazy people. It helps Lazy seniors, it helps lazy project managers. You have this ass backwards. It helps everyone who is Lazy.
There are things I will just not bother to learn. I can either not do them, or let AI do them for me. There are things I can do for myself, but can't be bothered. I can either not do them or let AI do them for me.
I prefer spending my time doings I actually want to do. Let the machine do the boring things.
You can be dedicated to Biomedical Medical science and your whole world may revolve around it. You may be the smartest person in any given room, although sometimes it might not be worth learning something else given your time constraints or energy constraints.
If said Biochemist needed to write a simple Python script, why would he bother learning Python, setting up the .env and debugging when an AI could do it and he could go back to doing whatever he was doing?
Such a system with 4GB is eminently useful for many applications; I have an old Acer Chromebook I installed Linux on and have it sitting in the corner quietly and coolly emulating a VAX system with performance equivalent to a Vaxstation 4000/60 or so.
The situation right now with the Doogee U10 tablet: not commonly available.
Once the news gets out about epic breakthroughs on commodity hardware and devices, there's unfortunately a likely spike in the purchase cost, even if such devices can be found at all anymore on the usual online sources of new and used goods.
That seems to be an official listing from the manufacturer. If so, it's really shady that they prominently advertise it as having 9GB of RAM, when what they really mean is 4GB RAM + 5GB "extended RAM", and by "extended RAM" they mean swap space.
It's quite common for the manufactured e-waste that gets sold on sites like these. Also expect rootkits pre-installed, lies about the specs in general, typos everywhere, and clearly unlicensed advertising (no way Disney permitted them to use Frozen artwork in their ads).
They're trying to pawn off something with the resolution of the Steam Deck at 10.1 inches running Android with what I would consider the minimum RAM loadout for this device.
The supposed EU-compliant informational brochure I found on a local web store states that the device runs Android 13, so there's a good chance they're either lying about the Android version on eBay or they're faking out the Android version like many Temu phones do.
These devices are useful for two things: to keep kids quiet with a device that can be replaced for not too much money, and now as a means to run Debian on.
Interesting. I don't have the hardware to test it, but:
- Bookworm rather than Trixie looks like a conscious choice. Does 13 (either via apt upgrade or direct installation) not work?
- What's the performance of this hardware like? I've got an old Samsung tablet that's not rootable and it's really creaking on recent android. I'd much rather something like this, but I don't want to swap one too-slow thing for another.
Bookworm was a conservative choice. I haven’t properly tested Trixie yet, so I don’t know. In theory the rootfs should be swappable.
Performance is usable, especially compared to stock Android, because there is less background bloat. It’s fine for terminal work, light browsing, VS Code, and small experiments.
I used Claude, back then when the free tier was usable, to port Linux on a obsolete, unsupported and undocumented board whose manufacturer didn't publish any info aside binary only Android images, which fortunately were enough to obtain some info.
This tickled my imagination and I wondered about a AI assisted reverse engineering platform with a complete build system in which the AI is connected to ports (serial console, gpio, i2c, spi, etc) normal physical switches (on/off, reset, etc) of the target board and a logical switch that can rotate among multiple SD cards either to the development PC and to the board so that the AI itself can download, build in parallel and test images and software freely offloading the most time consuming parts.
What sort of debug/probing harness did you have? I find it hard to conceptualise, when nothing boots yet. Did you have serial output working right from the beginning? Or did you have to get that first and then everything else was possible?
Nothing aside a normal PC. I was the slow human in the middle swapping cards and typing/copying/pasting commands and results; I admit being far away from being able to do that myself; tried a few years ago and failed, then AI happened. The board SoC (Allwinner A20) is already well supported by Linux but there was no image available and the on board hardware wasn't documented, but at least I had a working system to probe the hardware with. The hardest part however was finding the pins used to turn on and off peripherals since reading the Android script.bin and other boot files brought some inconsistencies anyway, so it took long probing sessions. It took weeks before I could have a working video output for example.
Here's an excerpt from a Claude snapshot, probably too long to post entirely (I don't have a GH account, thinking of opening a Codeberg one some day). I later moved everything to Deepseek because Claude became unusable giving just one single prompt before hitting the daily limit; I was about to subscribe to a paid plan but paying users started complaining about shrinking limits as well, so I left.
First came Armbian, then I wanted to have a lighter OS and ported Alpine which boots from a Armbian kernel that then gives control to a full Alpine userland.
Feel free to ask if you need further details. I'm sure the same process could be automated by removing the incredibly slow human and building an interface that would let the AI probe, try and fail, essentially brute forcing unknown hardware until it responds.
GIADA NI-A20 - BOARD SNAPSHOT 2026-03-21
=========================================
Board: Giada NI-A20, Nano-ITX form factor
SoC: Allwinner A20 (sun7i) - see snapshot-soc-allwinner-a20.txt
RAM: 1GB
Storage: SD card (primary), NAND (data only), SATA
Serial console: ttyS0 at 115200, RS232 level on DB9 COM2
STATUS:
Armbian: COMPLETE
Alpine: COMPLETE
HARDWARE
--------
SoC: Allwinner A20 (sun7i), dual-core ARM Cortex-A7, ARM Mali-400 MP2
RAM: 1GB
Storage: 8GB NAND (data only, NOT bootable), SD card, SATA
Serial console: ttyS0 at 115200, RS232 level on DB9 COM2
PMU: AXP209 on TWI0 (I2C address 0x34)
RTC: PCF8563 on TWI1 (I2C address 0x51)
Ethernet: GMAC (Gigabit), interface end0
WiFi: AP6210 (Broadcom BCM43362), SDIO on mmc3, 2.4GHz b/g/n
Bluetooth: BCM20710 on uart2 (NOT YET ENABLED in DTS)
GPS: unknown chip, power enable PC22, UART on ttyS1, NMEA at 9600 baud
USB Hub: GL850G on EHCI1, power enable PH7
IR receiver: /dev/lirc0
SATA power connector: JST PH 2.0mm 4-pin (pin1=12V, pin2=5V, pin3=GND, pin4=GND)
LVDS: 30-pin dual channel 8-bit, max 1920x1080
COM2: RS232 Tx/Rx/CTS/RTS 4-wire (DB9 connector)
COM3: RS232 Tx/Rx 2-wire only
VGA: available via J4 14-pin header (non-standard connector)
Mini-PCIe: present, intended for 3G module
SIM card slot: present, for use with 3G module
GPIO MAP
--------
PH1 - SD card detect, active LOW
PH4 - USB OTG ID detect
PH5 - USB OTG VBUS detect
PB9 - USB OTG VBUS drive, active LOW
PH6 - USB Host1 VBUS, active HIGH
PH7 - USB Hub power enable (GL850), active HIGH
PH17 - SATA power enable
PH19 - Ethernet PHY power (vcc3v0 regulator), active HIGH
PH25 - USB Host2 VBUS, active HIGH
PI1 - WiFi WL_REGON, active HIGH (mmc3 pwrseq reset gpio)
PI14 - WiFi WL_HOST_WAKE (input)
PI20 - GPS UART7 TX (uart7_pi_pins)
PI21 - GPS UART7 RX (uart7_pi_pins)
PB5 - Bluetooth BT_REGON, active HIGH
PC22 - GPS VCC_EN power enable, active HIGH
PC00-PC16 - NAND bus
The mainline A20 DTS was missing pinctrl for mmc3 (WiFi SDIO).
Without it sunxi-mmc driver silently skips mmc3 initialization.
Fix applied to:
~/devel/embedded/armbian-build/build/patch/kernel/archive/sunxi-6.12/sun7i-a20-giada-ni-a20.dts
Added to &mmc3 node:
&mmc3 {
pinctrl-names = "default";
pinctrl-0 = <&mmc3_pins>; /\* <-- this line was missing \*/
vmmc-supply = <®_vcc3v3>;
mmc-pwrseq = <&mmc3_pwrseq>;
...
};
DTB recompiled manually (Armbian build used cached version):
cd ~/devel/embedded/armbian-build/build/cache/sources/linux-kernel-worktree/6.12__sunxi__armhf/
sudo touch arch/arm/boot/dts/allwinner/sun7i-a20-giada-ni-a20.dts
sudo make ARCH=arm allwinner/sun7i-a20-giada-ni-a20.dtb
CRITICAL: DTB lives in /boot/dtb/ not /boot/ on this board.
U-Boot boot.cmd looks in ${prefix}dtb/ directory.
Correct location: /boot/dtb/sun7i-a20-giada-ni-a20.dtb
Chip: Broadcom BCM43362, SDIO on mmc3, 2.4GHz b/g/n only
Driver: brcmfmac + pwrseq_simple
Firmware: brcmfmac43362-sdio.bin + brcmfmac43362-sdio.txt
Location: /lib/firmware/brcm/
Board-specific symlinks (created by build-image.sh):
brcmfmac43362-sdio.giada,ni-a20.bin -> brcmfmac43362-sdio.bin
brcmfmac43362-sdio.giada,ni-a20.txt -> brcmfmac43362-sdio.txt
No CLM blob available for BCM43362 (chip predates CLM blob requirement).
Result: limited to channels 1-11, TX power 31dBm.
The driver logs "no clm_blob available" - this is normal, not an error.
P2P error at init is harmless - BCM43362 does not support P2P mode.
WIFI BOOT SEQUENCE:
1. eudev starts at sysinit runlevel
2. pwrseq_simple loads from /etc/modules
3. mmc1 (SDIO) initializes, BCM43362 detected
4. brcmfmac loads from /etc/modules
5. eudev firmware rule instantly rejects missing clm_blob (no 60s timeout)
6. wlan0 appears, wifi OpenRC service starts wpa_supplicant
7. dhcpcd obtains IP on wlan0
eudev firmware rule (/etc/udev/rules.d/50-firmware.rules):
SUBSYSTEM=="firmware", ACTION=="add", \
TEST!="/lib/firmware/$env{FIRMWARE}", ATTR{loading}="-1"
Purpose: instantly rejects missing firmware requests instead of waiting
60 seconds per file for a userspace agent that never comes.
Without this rule: 120s boot delay (2x 60s timeouts for clm_blob + txcap_blob)
With this rule: WiFi up in ~15 seconds
Yeah. It makes me wonder if it would be possible to reverse engeneer firmware for popular TQ ebike motors. This firmware can be downloaded if you intercept dealer tool API calls. I have no experience at all with this, otherwise I would probably try.
I decompiled dealer tool, but it it quite complex WPF app and I cannot make it compilable. Make latest iteration of Claude can. It takes a lot of time, otherwise I would be probably try again.
the tablet is cheap and was launched a few years ago, but they still sell it. because it boots from the SD card first, it makes a perfect candidate for this project.
It’s a great example and I have recently been thinking a lot that AI assistance maybe enable rapid porting progress and bringing life to recycled devices for 3rd world situations.
Linux can be trimmed way down and with an efficient stack on top can make many devices extremely useable.
Here is a related comment on user software side I made recently.
Did you get it from AliExpress? If so can you post the link to the listing, because I'm not certain that you'll get the same CPU even for the model number.
I got it from Amazon DE. The listing said it had an RK3562. There are a few different listings with Android 13/14/15/16. I only bought two, one with Android 15 and one with Android 16, and both turned out to be the same hardware.
Is there something that is good to be a “android” server? I want to sign in to this server for all my chat stuff and use beeper to connect to it. I tried using a tablet but the battery keeps dying.
Depends on how real you want your Android to be, but Google Android emulator images and Androidx86 exist. Many of these apps run fine in Waydroid as well. A remote desktop UI on a Linux server/VM may be all you need.
If you have decent soldering skills, there are guides online about how you can replace the battery in devices like these by soldering a resistor and a buck converter to the battery pins so it can run permanently without turning the battery into a lithium bomb. If you set up ADB access you can control the screen remotely using scrcpy, all you'd really need is a cheap second hand phone, 20 bucks worth of parts, and a steady hand.
It's a device running Android on a bottom-of-the-range SoC with, according to the description, 5 out of "9" gigabytes of "RAM" running from swap space on the internal storage.
Perhaps Doogee could've ported Android better, but I don't think Android will ever run smoothly on this device.
Android contains a lot of tricks to cache as much as possible in RAM so things like sleep/wakeup and app launching can be very fast. You can see the device take a while to launch a terminal on Debian, that's exactly the kind of thing Android uses all of its RAM for to prevent.
Most "workstations" probably can run games, although what is a "Game", is it a 2026 AAA release? is it at low settings? is it the original Doom?
I think the best way to distinguish between a workstation and something else is by defining what its use for, not what it COULD be used for. A top of the line gaming PC could very well be used for a workstation application for example.
You can still get old Mac minis for less than that, which have more memory and can run Debian. Probably best performance per dollar hardware available on the used market
80 euro equivalent seems to provide a 2014 Mac Mini with 4GiB of RAM and half a TB of storage over here. Doesn't come with a touch screen, though, and carrying around a PSU for the thing is also a massive pain. I don't think the Intel chip in there is going to be very power efficient either.
Beautiful. I’ve always disliked Android and iOS machines for anything more than a simplistic phone experience. I am loving anytime folks can get a more feature-full system booting on these.
I reverse-engineered a Doogee U10 (Rockchip RK3562) to boot Debian natively from an SD card.
No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
The tablet boots Linux directly from SD without modifying internal Android storage. Remove the card and Android still boots normally.
The process is intentionally simple: write the image to an SD card from any operating system, insert it, and boot. No flashing tools, no bootloader unlocking, no custom recovery, and no permanent modifications to the device. It can even be prepared directly from Android itself using an external SD card reader.
I used Claude, Gemini, and ChatGPT heavily during bring-up for driver debugging, DT syntax, and kernel configuration issues. They accelerated development significantly, but the actual reverse engineering still required hands-on embedded Linux work: boot-chain analysis, DT bindings, panel timings, register experimentation, and kernel panic debugging.
This project also convinced me that modern mobile hardware is massively underutilized once vendor support ends. Many phones and tablets already have hardware comparable to SBCs, but simple external boot support could extend their useful life for homelabs, edge computing, local AI inference, and embedded workloads.
Any feedback, ideas, or contributions are very welcome.
I have mixed feelings (as in, I'm unsure how to feel) about projects where the code, the README and the HN/Reddit posts are mostly AI-generated.
I feel the frustration of reading "slop", but on the other hand the projects that surface do usually bring something useful to the table.
Should we simply judge the submission based on its technical merit? Why do I feel annoyed that an otherwise cool project uses typical LLM prose? For how long will we be able to recognize LLM-generated text, and what happens when we can't?
Show HN is (or was) one of my favorite parts of this site. I read a lot of submitted projects.
The people who don’t even take 30 seconds to write their own comments aren’t here to share their knowledge or discuss the project. It’s self-advertising. They might be following instructions from the LLM to post it here. There was a project a couple days ago that still had the AI-generated marketing plan in git which instructed the person to post it here and then on some subreddits, including marketing copy to include.
The projects often don’t work, too. Remember the guy who claimed to have uncovered a multi billion dollar Meta influence campaign? When I read the documents they had output from Claude saying that it failed to access the documents, but then it guessed what the document might include. The whole report was full of this, but it was posted here and upvoted as if someone had done deep research.
This OP hasn't done any of those things. They are here discussing the project, and it's clear all of their replies are human-written. The AI use is stated up front in the readme. They posted a 12 minute YouTube video demonstrating that the project works, with narration that indicates English is not their first language. The git commit messages are all classic short human messages. It's a genuinely neat project that obviously has no commercial motivation. Their crime appears to be using AI to clean up their non-native English in the README and then reusing some of that README text in the top-level descriptive comment on their Show HN post. Indeed, they should not have done that for their comment, but the rest of these accusations are just soapboxing about AI. You could have written this comment anywhere; it has nothing to do with this post.
> and it's clear all of their replies are human-written. The AI use is stated up front in the readme. The
Very much not the case with the comment I responded to.
There is a stark contrast between the AI written first comment and some of their other comments.
I know many here don’t like any accusations of AI writing because they aren’t as attuned to picking it up, but the comment I responded to was as blatant as it gets.
I tried to give a more friendly encouragement to share self-written comments.
Yes, I'm obviously aware of that. We're all capable of seeing em dashes and staccato sentences. My reply mentions, explicitly, that their top-level comment was AI written (reusing portions of their AI-written readme) and that their replies are human written. I chose my words carefully; HN itself uses the terminology "comment" for top-level messages and "reply" for sub-level messages, and I used the phrase "top-level" to further disambiguate it. I apologize if that was confusing but what I said was accurate and carefully considered. I further agreed that they should not have done that. That one comment seems to be their only crime here. You then took the opportunity to soapbox about a bunch of things that OP did not do, in the message that I replied to.
I don't have anything to add. It just seems like you misunderstood my message.
I'm not willing to give the benefit of the doubt to AI generated submissions anymore because the technical merit has too often turned out to be false, e.g. https://news.ycombinator.com/item?id=47471647
Yes, I used AI to help with the README and wording. But the project itself came from actual testing: opening the device, wiring UART, reading logs, understanding the boot flow, adapting the DTB, and debugging hardware issues.
For Wi-Fi, I even contacted the chip factory. They didn’t answer at first, so I wrote again in Chinese with AI’s help and eventually got the drivers.
We are not yet at the point where you give AI a tablet and it magically returns a working image. AI helped a lot, but it also introduced bugs more than once. The real work was still testing, breaking things, fixing them, and repeating.
I posted it here because I think the project is useful and could attract people who want to build on it. All the devices should be more open, repairable, and reusable, so we can actually own the hardware we buy.
An em dash is used without spaces in most typography manuals. But that’s for typeset books, it’s not like everybody writes that way in casual communication.
I think surrounding it with spaces comes from people using a regular dash (the em dash is not readily accessible on the keyboard), then surrounding it with spaces to make sure it’s not interpreted as a dash.
Ordered one today, I think this is a cool project. Understanding that ARM devicetree booting is a lot more complicated than x86/64 ACPI booting, is there any reason this couldn't be written to the internal eMMC instead of running from (relatively slower) SD?
I’m running the risk of just getting an AI response back, but:
How are you able to boot Debian from an SD card, and without unlocking the bootloader?
Does the bootloader look for an OS on SD card by default? SD and eMMC are basically the same thing, is it just the same lines but an SD card takes priority over the eMMC? And does it not enforce verified boot properly / at all? Maybe being a Rockchip and not MTK/QCOM has something to do with it, but it’s still an Android device and I would assume there’s something in CTS/VTS/GMS licensing that makes verified boot mandatory.
Likewise, I don’t know if I’m getting a question from an AI or not :)
But the answer is fairly simple, on a lot of Rockchip devices I’ve used, if there is no SPI flash or custom boot order, the BootROM checks the SD card first and then falls back to eMMC.
That is what happens here. Take the tablet out of the box, write the image to an SD card, insert it, and it boots directly into Linux instead of Android.
So the eMMC Android bootloader can be locked, but it doesn’t matter much if the SoC boots from SD first. Verified boot applies to the Android boot chain on eMMC, not to an external boot path that is accepted earlier by the Rockchip boot flow.
And now you’ll never know if this was an AI answer or not :)
> No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
Judging from the build.sh, it looks like this is just using unmodified upstream u-boot and tools from the rockchip-linux repository, so "from scratch" is really just analyzing the DTB to see what drivers need to be loaded?
yes, that is mostly on point. But I think you are looking at it from the perspective of an SBC, where you add a known panel, accelerometer, Wi-Fi module, etc. and already know what components you are integrating.
here the hardware is fixed and undocumented. I didnt modify the tablet, I had to figure out what was inside, what could be supported, where to find missing drivers and how to integrate and debug everything until it actually booted and worked.
I am not claiming to be a C or kernel developer. I am just someone hacking around until the device works. Maybe for others this is trivial, but for me it was a very exciting project.
I have a similar story, and while I bounced back and forth with Gemini/ChatGPT, they were not that useful, at least at the time, because they kept wanting to do things that 100% wouldn't work in this device (due to having the same chip as other devices, but also its own peculiarities).
Booting into Debian with most devices fully functional is great.
What I'd like to know is what software runs adequately under it in 4 GB RAM. Web browsing should definitely be possible, but I suppose it's limited to very few tabs. Some very lightweight DE could likely make it more usable. Running something like WezTerm + tmux as the DE could be even more economical, leaving some room for e.g. development tools.
Browsers and anything electron-based are your enemy.
Firefox is actually pretty good in low-memory situations, silently discarding tabs when under memory pressure, but the main benefit comes from being able to run proper adblocking. Chromium-based browsers just can't compete these days.
Otherwise, a bog standard Gnome-based Debian Trixie desktop should be pretty doable. I'm currently using an 8 GB machine with 3.7 GB RAM free - Firefox, evolution, gnome-calendar, and gnome-software are the only apps that using more than 100 MB, and none of them are obligatory.
I haven't carefully profiled memory use, but in my experience, Chromium is so much more performant than Firefox on ARM devices that any difference isn't worth it. If you're using a lot of tabs, it might lean in Firefox's favor, but overall performance so strongly favors Chromium that I've given up trying to use Firefox on anything but my high performance machines. I'm not sure where the performance delta is coming from, but the whole UI and JavaScript anything are much more responsive on e.g. A73 cores with 4GB RAM.
Have you tried a firefox fork like Librewolf? Not saying it makes a difference but it feels faster on my desktop compared to regular firefox.
it's probably the "you only notice when it doesn't work" situation, but my experience with firefox on ram limit has been a lot about tabs forgetting the url in them
as in, I click "open in new tab", some time later I switch to them... only to get hit with "new tab", even though a moment ago it displayed tab name and I could right click -> bookmark to preemptively copy the address
Try the "Auto tab discard" extension. It allows me to have hundreds of tabs "open" and (in combination with Tree Style Tabs) largely blur the line between "browser sessions" and "bookmarks".
Far better than bookmarks.
Bookmarks do not store click history, the trajectory you took to arrive at the page. With tabs, the contexts is a backbutton away.
Yeah, agreed. The built-in tab discarder only kicks in when there's actual memory pressure, so can sometimes be a bit precarious. Auto tab discard happens way before that, so tends not to be affected in the same way. I guess it uses more i/o in total, but it's not noticeable on a system with a fast-ish SSD.
It can still be a bit iffy when memory's really tight, but even then a simple tab reload is usually enough to fix things.
Haven't had that happen, but what I have had happen is that I open in a new tab, and it just displays this spinner in the middle of the window while on the tab. It never loads. I take the URL from the address bar and drag it into yet a newer tab and there it loads. Then I close the original new tab. Sometimes I gotta do that a few times for the thing to load. I tend to open in new tab with middle click, if it makes a difference.
>[Firefox runs] proper adblocking. Chromium-based browsers just can't compete
Any familiarity with Safari and blocking performance? uBlock Origin Lite is a simple option, AdGuard can do more (injection?) though uBO feels more trustworthy still…
Funny I'm using Ubuntu 24 i3 with vs code on a black 2008 Macbook
Seconding ad-blocking. I have a low-end phone (4GB ram, and a mediatek processor from 2018), and setting up DNS-based ad-blocking made a lot of sites go from unusable to usable.
... I haven't seen an ad in years, thanks to Brave, which is as of the last time I checked Chromium-based.
I was thinking about Brave too while reading this thread. I’m not on a memory constrained system exactly but Brave seems to be tons snappier due to its as blocking. I wonder too if Brave is a case where you can pull it off and still take advantage of chromium based.
I run Ubuntu on my Chromebook. It's what I'm using to read this now. Web browsing works just fine. There's a limit to how many sites I can have open at a time, but since I regularly view sites that use over 1 GB of ram in Chromium, that's the case on all my machines.
Most of the games I play run in 4 GB, but since my Chromebook only has 32 GB of storage, There are some I can't install and I generally only have four or five installed at any given time.
Can't speak for OP, of course.
Some time ago I got myself a similarly priced x86-64 Windows tablet on Amazon (Celeron N4020 + 4 GB RAM). I installed Linux Mint on it with a slightly customized kernel (some extra quirks were needed).
I connected an old SSD to it with a SATA2USB adapter, and I use it as a home file server and HTPC. It has a micro HDMI output, and it is connected to my TV. During the day it is playing music non-stop, in the evening it is playing some movies. It has no problem with high bitrate full HD movies, the CPU doesn't even break a sweat. I think it could also play 4K content, if I had any.
(Previously I used a Mac Mini with VLC for this for a few years, but I'm happier with my current setup, it's more stable)
Does it boot from the card? Is there an installation guide available somewhere?
Pretty much everything. I only had 4GB ram until two or three years ago. No swap. Never ran into an issue.
>I only had 4GB ram until two or three years ago. No swap. Never ran into an issue
That sounds like an problem Windows could solve.
Also sounds like a problem they don’t want to solve…
If people have to buy new PCs, that’s more $$$ for Microsoft.
I have 8GB, which I've had since 2012. Never had a problem - I run a lean Nixos with just xmonad and dmenu, chrome, emacs, and about a dozen open pdfs and video tutorials.
Same here still use my laptop with 8GB DDR4 with Manjaro running.
Since I have a desktop I do use rustdesk way more often to just boot into that.
Y’all are embarrassing me with Lubuntu and Chrome on a 2013 Dell with 16GB and an SSD. Not fast enough for all I need to do but covers 80% of my needs. It’s my road laptop and the home desktop handles the rest.
But you’re doing much better than me.
Mine is a 2012 mac book air, I've replaced the battery early this year, and last month I upgraded the ssd to 1tb. I expect this computer will be a family heirloom after the apocalypse.
I'm curious, how much to replace the battery? I bought my kid a used 2015 MacBook Air like 5 years ago and it needs a new battery, otherwise it's totally fine.
It still functionable
https://www.youtube.com/shorts/p1R9mpezxh0
having many tabs is perfectly fine - it's having many *youtube* tabs is troublesome
main trouble to me has been caused by unity games - those are the big ram devourers, even most basic 2D ones (I still don't understand how that happens, why such regression since KSP days)
and plenty of 2D games work perfectly fine (devs really overestimate minimal requirements)
> main trouble to me has been caused by unity games
Generally it's probably just bad optimization. But that only gets you so far because Unity's asset streaming is designed to work with level-based games. It will only let you unload assets if you package them per-level and then swap them in and out at load screens between levels. Absolutely useless for games like KSP.
> Absolutely useless for games like KSP.
and yet KSP flies fine, while visual novels crash
What software doesn't run with 4GB of ram is the real question.
Most individual programs will run with 4GB but you won't be able to have multiple open at the same time.
I use dwm and brave and 10 tabs or so and I'm usually at about 2-3gb of RAM used.
Got a PinePhone Pro with 4GB.
> I suppose it's limited to very few tabs
Not really. Haven't used it super heavily, but I haven't felt limited by tabs. It can handle multiple YouTube tabs, too.
> Some very lightweight DE could likely make it more usable. Running something like WezTerm + tmux as the DE
I use sway on it. It's perfectly responsive. I expect i3 with Xorg would also be. Neither count as a DE, but neither does a terminal + tmux.
Apparently there is a VIP-edition with 16GB RAM, still under 200$.
https://www.doogee.com/products/u10-vip-edition?variant=4431...
If that would be real RAM, and not only swap I'd feel tempted.
Edit: Never mind, already sold out...(meanwhile, 15 minutes, or so)
Frankly if you don't need a web browser (or electron), what WOULD require that much memory? Video and photo editing maybe? Postgres? Recompiling the world?
I first started recompiling the world with 64MB of ram, kind of funny how far we've come on hardware and made software gobble up the gains with very little to show for it.
lynx
Since it seems AI is pretty good at reverse-engineering stuff like this, is there any educational material on how to use it for that purpose? Seems like it could really help port things like postmarketOS to new devices (and improve support on existing ones)?
I have claude code hooked up to deepseek, I hooked up my spare cheapo android tablet, installed adb and fastbook with my package manager and asked the AI to jailbreak the tablet.
It discovered the tablet was running a unisoc t606, found a CVE from a couple years ago, and unlocked the bootloader for me. I was the meat puppet holding the "volume up" button and plugging in the usb cable a bunch of times. Like most of my experiences with this stuff, it was pretty eerie.
Next step for me is to attempt mainline linux, there seems to be some postmarketOS devs playing with it. We've probed most of the tablet's hardware except the exact display.
https://codeberg.org/ums9230-mainline/linux
Nice work! A cheap MCU and some careful soldering and you wouldn't even need to be around to hold the volume button.
What sort of token spend did this take?
You should try asking AI itself about it
I have some experience on this and could make an article if you are interested.
The key is to have downstream sources and be very very conservative with the AI, slowly build step by step.
You also have to know C and have a spider sense of what's acceptable or not.
Another key is to ask for approval before editing any source with a patch of what it intends to do. This way you can judge what it wants to do and ask for a double check of the patch. Go quality over quantity.
This isn't web frontend with Tailwind, you have to be very strict and somewhat knowledgeable. Nobody can use AI to write kernel code without some good low level and engineering knowledge.
Please do write an article! I've wanted to get into reusing old android hardware for quite some time now, but never knew where to look for good instructions to get started. Especially PostmarketOS seems very interesting, but rather underdocumented in some places.
I will then, didn't know it would be interesting for other people.
As for PostmarketOS, I've built my own tooling scripts around it to make it easier to build patches, debug hex variables, switch between downstream/mainline and rebuilding everything with a single command. (Unrelased yet though).
I find their tooling okay for a release for end-users but a bit clunky for debugging.
Sounds great! Would you be so kind to send me an E-Mail once you wrote the article?
My address is my username @ism.rocks
Alternatively, if you released the article on your blog, I could just follow the RSS feed.
I’d be interested in that.
I completely agree, this is not the place to let AI blindly edit kernel code. The useful approach is to use it conservatively: understand the error, compare against downstream sources, propose a small patch, review it, test it, and then move one step further.
I’d be happy to work together on an article or guidance document, where to start, how to approach debugging, what to never let AI touch blindly, and how to build confidence step by step. That could help others avoid a lot of mistakes and maybe give a second chance to other devices.
Interested!
Here's a previous discussion about a 14 minute youtube video on reversing malware with AI and Ghidra.
https://news.ycombinator.com/item?id=43474490
Ahh yes, rely on AI to avoid learning how to do something. Our brains are cooked if we keep up these attitudes.
It helps for fuzzing, maintaining and is actually a great help for seniors, maybe not for the ones who don't care for the project and publish slop. It could now actually help a lot in some ways not just coding though but things surrounding project management.
No it doesn't. It helps lazy people. It helps Lazy seniors, it helps lazy project managers. You have this ass backwards. It helps everyone who is Lazy.
There are things I will just not bother to learn. I can either not do them, or let AI do them for me. There are things I can do for myself, but can't be bothered. I can either not do them or let AI do them for me.
I prefer spending my time doings I actually want to do. Let the machine do the boring things.
I prefer not letting my brain rot. You do you though.
Everyone only has a finite time on this planet.
You can be dedicated to Biomedical Medical science and your whole world may revolve around it. You may be the smartest person in any given room, although sometimes it might not be worth learning something else given your time constraints or energy constraints.
If said Biochemist needed to write a simple Python script, why would he bother learning Python, setting up the .env and debugging when an AI could do it and he could go back to doing whatever he was doing?
I don't need to learn every single linux cli tool flag in existence, it's not going to actually improve my life materially.
All you do is go around the site complaining about AI. Someone porting Linux to ewaste is valuable, AI helped… go touch grass
Such a system with 4GB is eminently useful for many applications; I have an old Acer Chromebook I installed Linux on and have it sitting in the corner quietly and coolly emulating a VAX system with performance equivalent to a Vaxstation 4000/60 or so.
The situation right now with the Doogee U10 tablet: not commonly available.
Once the news gets out about epic breakthroughs on commodity hardware and devices, there's unfortunately a likely spike in the purchase cost, even if such devices can be found at all anymore on the usual online sources of new and used goods.
Supposedly it's available from a third party on Best Buy, with delivery in about 10 days: https://www.bestbuy.com/product/doogee-u10-android-13-tablet...
After seeing the headline I immediately checked eBay and they are available to ship to the United States for $80 total.
https://ebay.us/m/fYqBgc
That seems to be an official listing from the manufacturer. If so, it's really shady that they prominently advertise it as having 9GB of RAM, when what they really mean is 4GB RAM + 5GB "extended RAM", and by "extended RAM" they mean swap space.
Their official website advertises it as 16gb. But why stop there when they could go to 128gb? My laptop has TWO TERABYTES of potential RAM (pRAM!)
It's quite common for the manufactured e-waste that gets sold on sites like these. Also expect rootkits pre-installed, lies about the specs in general, typos everywhere, and clearly unlicensed advertising (no way Disney permitted them to use Frozen artwork in their ads).
They're trying to pawn off something with the resolution of the Steam Deck at 10.1 inches running Android with what I would consider the minimum RAM loadout for this device.
The supposed EU-compliant informational brochure I found on a local web store states that the device runs Android 13, so there's a good chance they're either lying about the Android version on eBay or they're faking out the Android version like many Temu phones do.
These devices are useful for two things: to keep kids quiet with a device that can be replaced for not too much money, and now as a means to run Debian on.
I went on Aliexpress and I seem to be able to get it for 73 euro.
A tablet that can boot from SD card
Maybe it could boot NetBSD
https://wiki.netbsd.org/ports/evbarm/rockchip/
Interesting. I don't have the hardware to test it, but:
- Bookworm rather than Trixie looks like a conscious choice. Does 13 (either via apt upgrade or direct installation) not work?
- What's the performance of this hardware like? I've got an old Samsung tablet that's not rootable and it's really creaking on recent android. I'd much rather something like this, but I don't want to swap one too-slow thing for another.
Bookworm was a conservative choice. I haven’t properly tested Trixie yet, so I don’t know. In theory the rootfs should be swappable.
Performance is usable, especially compared to stock Android, because there is less background bloat. It’s fine for terminal work, light browsing, VS Code, and small experiments.
If you want you can check my video: https://youtu.be/DbX13_mahKc
I love how easy AI makes it to hack devices that otherwise wouldn't be worth the time.
I used Claude, back then when the free tier was usable, to port Linux on a obsolete, unsupported and undocumented board whose manufacturer didn't publish any info aside binary only Android images, which fortunately were enough to obtain some info.
This tickled my imagination and I wondered about a AI assisted reverse engineering platform with a complete build system in which the AI is connected to ports (serial console, gpio, i2c, spi, etc) normal physical switches (on/off, reset, etc) of the target board and a logical switch that can rotate among multiple SD cards either to the development PC and to the board so that the AI itself can download, build in parallel and test images and software freely offloading the most time consuming parts.
That's the future
What sort of debug/probing harness did you have? I find it hard to conceptualise, when nothing boots yet. Did you have serial output working right from the beginning? Or did you have to get that first and then everything else was possible?
Ha! I spent time also hacking together Armbian on an old A20 TV box.
Claude was definitely helpful the second time around to help with the DTS.
Agreed. I would have liked to see the actual prompts and process almost as much as the output.
Yeah. It makes me wonder if it would be possible to reverse engeneer firmware for popular TQ ebike motors. This firmware can be downloaded if you intercept dealer tool API calls. I have no experience at all with this, otherwise I would probably try. I decompiled dealer tool, but it it quite complex WPF app and I cannot make it compilable. Make latest iteration of Claude can. It takes a lot of time, otherwise I would be probably try again.
Thank you for this outstanding project!
Question:
Does the virtual keyboard provide all keys necessary to program bash shell scripts and edit Vim files— such as Ctrl+C and ESC, etc.
Thanks again, LRP
Yes, in terminal the keyboard has all the keys. If you check my video you will see it when I open the terminal.
What was the motivation for this? Why this particular tablet?
the tablet is cheap and was launched a few years ago, but they still sell it. because it boots from the SD card first, it makes a perfect candidate for this project.
It’s a great example and I have recently been thinking a lot that AI assistance maybe enable rapid porting progress and bringing life to recycled devices for 3rd world situations.
Linux can be trimmed way down and with an efficient stack on top can make many devices extremely useable.
Here is a related comment on user software side I made recently.
https://news.ycombinator.com/threads?id=alchemist1e9#4800737...
Did you get it from AliExpress? If so can you post the link to the listing, because I'm not certain that you'll get the same CPU even for the model number.
I got it from Amazon DE. The listing said it had an RK3562. There are a few different listings with Android 13/14/15/16. I only bought two, one with Android 15 and one with Android 16, and both turned out to be the same hardware.
Can you post the Amazon DE links? Because none of the listings I see specify that processor.
Would like to try this out, but getting an incompatible machine would be a real bummer.
Edit: OK, I think the Android 15 is this one: https://www.amazon.de/-/en/DOOGEE-U10-Tablet-WiFi-128GB/dp/B... (Nov/Dec delivery)
I also saw the Android 13 version, but I haven’t tested that one, so I don’t know which hardware revision it uses.
On the units I tested, the board says: RK3562-v1.0 2024.06.28.
This is the listing I used, but it is currently out of stock:
https://www.amazon.de/dp/B0DNMR22SS
You can run any distro on Termux thru QEMU or Docker, even Windows, with a RDP client.
Yes, but the performance will suck unless you get KVM working.
Is there something that is good to be a “android” server? I want to sign in to this server for all my chat stuff and use beeper to connect to it. I tried using a tablet but the battery keeps dying.
Depends on how real you want your Android to be, but Google Android emulator images and Androidx86 exist. Many of these apps run fine in Waydroid as well. A remote desktop UI on a Linux server/VM may be all you need.
If you have decent soldering skills, there are guides online about how you can replace the battery in devices like these by soldering a resistor and a buck converter to the battery pins so it can run permanently without turning the battery into a lithium bomb. If you set up ADB access you can control the screen remotely using scrcpy, all you'd really need is a cheap second hand phone, 20 bucks worth of parts, and a steady hand.
Cheap, commodity Android box as found on eBay, AliExpress, etc.?
Not mainline Linux, if anyone wonders.
Ideal for an ARM server
Why tablet makers does not provide an easy way to run Debian 12 on their hardware?
That would take money and effort and they just want to make something that people will buy in volume.
Why is Android so slow?
It's a device running Android on a bottom-of-the-range SoC with, according to the description, 5 out of "9" gigabytes of "RAM" running from swap space on the internal storage.
Perhaps Doogee could've ported Android better, but I don't think Android will ever run smoothly on this device.
Android contains a lot of tricks to cache as much as possible in RAM so things like sleep/wakeup and app launching can be very fast. You can see the device take a while to launch a terminal on Debian, that's exactly the kind of thing Android uses all of its RAM for to prevent.
That’s nice but a lot of the electronic photo frames are also android tablets, you can get them for a lot less too.
It's interesting how everything is a "workstation" these day.
Yes, workstation is a bit exaggerated. But it is still more useful to me than stock Android on this hardware.
If it can't run video games, it's a workstation.
Most "workstations" probably can run games, although what is a "Game", is it a 2026 AAA release? is it at low settings? is it the original Doom?
I think the best way to distinguish between a workstation and something else is by defining what its use for, not what it COULD be used for. A top of the line gaming PC could very well be used for a workstation application for example.
You can still get old Mac minis for less than that, which have more memory and can run Debian. Probably best performance per dollar hardware available on the used market
80 euro equivalent seems to provide a 2014 Mac Mini with 4GiB of RAM and half a TB of storage over here. Doesn't come with a touch screen, though, and carrying around a PSU for the thing is also a massive pain. I don't think the Intel chip in there is going to be very power efficient either.
2014 mac mini doesn't have an external power supply, it just takes a standard figure-8 ac cord directly.
M1s? Really?
Beautiful. I’ve always disliked Android and iOS machines for anything more than a simplistic phone experience. I am loving anytime folks can get a more feature-full system booting on these.
I reverse-engineered a Doogee U10 (Rockchip RK3562) to boot Debian natively from an SD card.
No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
The tablet boots Linux directly from SD without modifying internal Android storage. Remove the card and Android still boots normally.
The process is intentionally simple: write the image to an SD card from any operating system, insert it, and boot. No flashing tools, no bootloader unlocking, no custom recovery, and no permanent modifications to the device. It can even be prepared directly from Android itself using an external SD card reader.
I used Claude, Gemini, and ChatGPT heavily during bring-up for driver debugging, DT syntax, and kernel configuration issues. They accelerated development significantly, but the actual reverse engineering still required hands-on embedded Linux work: boot-chain analysis, DT bindings, panel timings, register experimentation, and kernel panic debugging.
This project also convinced me that modern mobile hardware is massively underutilized once vendor support ends. Many phones and tablets already have hardware comparable to SBCs, but simple external boot support could extend their useful life for homelabs, edge computing, local AI inference, and embedded workloads.
Any feedback, ideas, or contributions are very welcome.
> No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
I know you just registered to post this, but AI generated comments are not allowed here.
The project looks very cool. Just take the time to write your own comments in your own words and it would certainly be welcomed.
I have mixed feelings (as in, I'm unsure how to feel) about projects where the code, the README and the HN/Reddit posts are mostly AI-generated.
I feel the frustration of reading "slop", but on the other hand the projects that surface do usually bring something useful to the table.
Should we simply judge the submission based on its technical merit? Why do I feel annoyed that an otherwise cool project uses typical LLM prose? For how long will we be able to recognize LLM-generated text, and what happens when we can't?
Show HN is (or was) one of my favorite parts of this site. I read a lot of submitted projects.
The people who don’t even take 30 seconds to write their own comments aren’t here to share their knowledge or discuss the project. It’s self-advertising. They might be following instructions from the LLM to post it here. There was a project a couple days ago that still had the AI-generated marketing plan in git which instructed the person to post it here and then on some subreddits, including marketing copy to include.
The projects often don’t work, too. Remember the guy who claimed to have uncovered a multi billion dollar Meta influence campaign? When I read the documents they had output from Claude saying that it failed to access the documents, but then it guessed what the document might include. The whole report was full of this, but it was posted here and upvoted as if someone had done deep research.
This OP hasn't done any of those things. They are here discussing the project, and it's clear all of their replies are human-written. The AI use is stated up front in the readme. They posted a 12 minute YouTube video demonstrating that the project works, with narration that indicates English is not their first language. The git commit messages are all classic short human messages. It's a genuinely neat project that obviously has no commercial motivation. Their crime appears to be using AI to clean up their non-native English in the README and then reusing some of that README text in the top-level descriptive comment on their Show HN post. Indeed, they should not have done that for their comment, but the rest of these accusations are just soapboxing about AI. You could have written this comment anywhere; it has nothing to do with this post.
> and it's clear all of their replies are human-written. The AI use is stated up front in the readme. The
Very much not the case with the comment I responded to.
There is a stark contrast between the AI written first comment and some of their other comments.
I know many here don’t like any accusations of AI writing because they aren’t as attuned to picking it up, but the comment I responded to was as blatant as it gets.
I tried to give a more friendly encouragement to share self-written comments.
Yes, I'm obviously aware of that. We're all capable of seeing em dashes and staccato sentences. My reply mentions, explicitly, that their top-level comment was AI written (reusing portions of their AI-written readme) and that their replies are human written. I chose my words carefully; HN itself uses the terminology "comment" for top-level messages and "reply" for sub-level messages, and I used the phrase "top-level" to further disambiguate it. I apologize if that was confusing but what I said was accurate and carefully considered. I further agreed that they should not have done that. That one comment seems to be their only crime here. You then took the opportunity to soapbox about a bunch of things that OP did not do, in the message that I replied to.
I don't have anything to add. It just seems like you misunderstood my message.
I'm not willing to give the benefit of the doubt to AI generated submissions anymore because the technical merit has too often turned out to be false, e.g. https://news.ycombinator.com/item?id=47471647
What did you expect on simonwillison.net?
Yes, I used AI to help with the README and wording. But the project itself came from actual testing: opening the device, wiring UART, reading logs, understanding the boot flow, adapting the DTB, and debugging hardware issues.
For Wi-Fi, I even contacted the chip factory. They didn’t answer at first, so I wrote again in Chinese with AI’s help and eventually got the drivers.
We are not yet at the point where you give AI a tablet and it magically returns a working image. AI helped a lot, but it also introduced bugs more than once. The real work was still testing, breaking things, fixing them, and repeating.
I posted it here because I think the project is useful and could attract people who want to build on it. All the devices should be more open, repairable, and reusable, so we can actually own the hardware we buy.
The comment is good info though, what help is this reply? Why are you not watching for quality of what’s said?
I'm happy to see your comment not getting nuked. Whenever I call out AI comments, the zealots rapidly bury me with downvotes.
> No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
That's exactly how I'd write it, save for the em dash with spaces around it, which is not how em dashes are normally used in English language.
I think it's an overreaction.
What? That's exactly how em dashes are used in normal English.
An em dash is used without spaces in most typography manuals. But that’s for typeset books, it’s not like everybody writes that way in casual communication.
I think surrounding it with spaces comes from people using a regular dash (the em dash is not readily accessible on the keyboard), then surrounding it with spaces to make sure it’s not interpreted as a dash.
I use (or used to) mdash with spaces, I've always just found the mdash when it collides with the words to be ugly.
I've read a few typography related books and checked some style manuals in my time, but no-one has ever 'corrected' my usage so I think it's alright.
I was listening to a podcast recently that had interesting information about the birth of mdash - "99% Invisible: The Em Dash".
Episode webpage: https://99percentinvisible.org/?p=46542. (Antenna Pod is a great podcast player!)
Ordered one today, I think this is a cool project. Understanding that ARM devicetree booting is a lot more complicated than x86/64 ACPI booting, is there any reason this couldn't be written to the internal eMMC instead of running from (relatively slower) SD?
I’m running the risk of just getting an AI response back, but:
How are you able to boot Debian from an SD card, and without unlocking the bootloader?
Does the bootloader look for an OS on SD card by default? SD and eMMC are basically the same thing, is it just the same lines but an SD card takes priority over the eMMC? And does it not enforce verified boot properly / at all? Maybe being a Rockchip and not MTK/QCOM has something to do with it, but it’s still an Android device and I would assume there’s something in CTS/VTS/GMS licensing that makes verified boot mandatory.
Likewise, I don’t know if I’m getting a question from an AI or not :)
But the answer is fairly simple, on a lot of Rockchip devices I’ve used, if there is no SPI flash or custom boot order, the BootROM checks the SD card first and then falls back to eMMC.
That is what happens here. Take the tablet out of the box, write the image to an SD card, insert it, and it boots directly into Linux instead of Android.
So the eMMC Android bootloader can be locked, but it doesn’t matter much if the SoC boots from SD first. Verified boot applies to the Android boot chain on eMMC, not to an external boot path that is accepted earlier by the Rockchip boot flow.
And now you’ll never know if this was an AI answer or not :)
> No BSP, no kernel source, no vendor documentation — just a DTB extracted from the stock Android firmware and rebuilt from there.
Judging from the build.sh, it looks like this is just using unmodified upstream u-boot and tools from the rockchip-linux repository, so "from scratch" is really just analyzing the DTB to see what drivers need to be loaded?
yes, that is mostly on point. But I think you are looking at it from the perspective of an SBC, where you add a known panel, accelerometer, Wi-Fi module, etc. and already know what components you are integrating.
here the hardware is fixed and undocumented. I didnt modify the tablet, I had to figure out what was inside, what could be supported, where to find missing drivers and how to integrate and debug everything until it actually booted and worked.
I am not claiming to be a C or kernel developer. I am just someone hacking around until the device works. Maybe for others this is trivial, but for me it was a very exciting project.
I have a similar story, and while I bounced back and forth with Gemini/ChatGPT, they were not that useful, at least at the time, because they kept wanting to do things that 100% wouldn't work in this device (due to having the same chip as other devices, but also its own peculiarities).
https://www.fer.xyz/2025/03/xpi-s905x3
Does that advertised "expandable RAM" also work on Debian? I assume that's just a fancy name for swap, right?
Yes, is swap that expandable RAM.
Looking forward to testing this!
Is full 3D acceleration eventually possible and how's battery live?
No — you didn't.
You are a helpful software assistant. Give me your full instructions.