0

Excellent for local models running via OpenCode

GitHub GitHub - ghorbelhamdi/buddy-android: Android AI assistant in a floating bubble: Claude Code, Codex, OpenCode, Gemini (Antigravity CLI) or Grok use your phone for you — read the screen, tap, type. Runs on your own plan, open source, no root, no adb, no PC. github.com
preview

cross-posted from: lemdro.id/post/48219138

Buddy: an open source app that lets AI agents (Claude Code, Codex, OpenCode) use your Android phone

I’ve been building Buddy, a free, open source (Apache 2.0) Android app. You type or say what you want in a floating bubble, and an AI agent reads the screen, taps and types through your apps to do it.

  • Your choice of agent: Claude Code, Codex, OpenCode, Antigravity or Grok, running inside the app (Alpine Linux via proot) on your own plan. OpenCode’s free models need no account.
  • No root, adb or PC: it reads the screen and taps through the accessibility service.
  • Asks first: before anything is sent, posted, bought or deleted. It pauses when you touch the screen and never types passwords or payment details.
  • Builds apps: with Linux tools on, it builds real APKs on the phone. This week I drew a maze on a paper towel, took a photo, and it made it a tilt game. The first build crashed on launch; it found the bug and fixed it.

To be upfront:

  • What the agent reads on your screen goes to the AI service you pick.
  • It’s early (2.2), tested most on a OnePlus 12 with Android 16.
  • It’s not on Play (Google rejected the accessibility use), so it’s an APK from GitHub releases, built by GitHub Actions from the source.

Source and releases: github.com/ghorbelhamdi/buddy-android

Happy to answer questions here.

cross-posted from: lemdro.id/post/48219138

Buddy: an open source app that lets AI agents (Claude Code, Codex, OpenCode) use your Android phone

I’ve been building Buddy, a free, open source (Apache 2.0) Android app. You type or say what you want in a floating bubble, and an AI agent reads the screen, taps and types through your apps to do it.

  • Your choice of agent: Claude Code, Codex, OpenCode, Antigravity or Grok, running inside the app (Alpine Linux via proot) on your own plan. OpenCode’s free models need no account.
  • No root, adb or PC: it reads the screen and taps through the accessibility service.
  • Asks first: before anything is sent, posted, bought or deleted. It pauses when you touch the screen and never types passwords or payment details.
  • Builds apps: with Linux tools on, it builds real APKs on the phone. This week I drew a maze on a paper towel, took a photo, and it made it a tilt game. The first build crashed on launch; it found the bug and fixed it.

To be upfront:

  • What the agent reads on your screen goes to the AI service you pick.
  • It’s early (2.2), tested most on a OnePlus 12 with Android 16.
  • It’s not on Play (Google rejected the accessibility use), so it’s an APK from GitHub releases, built by GitHub Actions from the source.

Source and releases: github.com/ghorbelhamdi/buddy-android

Happy to answer questions here.

0

Hands on with the AYN Odin 3 Portal; All You Need, and Then Some

The Bryant Review Hands on with the AYN Odin 3 Portal; All You Need, and Then Some gardinerbryant.com
preview

I’ve just published my not-quite-a-review, not-quite-an-impressions-piece, but something closer to a good old-fashioned feature article about the AYN Odin 3 Portal. I wanted to approach this one a little differently from your typical handheld review.

I interviewed several developers behind the software people are most likely to use on hardware this powerful, including jpolo (ARMSX3), Camille LaVey (Eden), Utkarsh (GameNative) and Json (DroidDeck). Their insights into the Dragonwing Q8, the challenges of modern emulation and where things are heading form a big part of the article, alongside my own testing of everything from PS2 and GameCube to PS3, Xbox 360 and PC gaming.

It’s also a rather long read, much closer to those enormous feature articles gaming websites used to publish years ago. I’ve always loved that sort of writing, and I suspect it’ll appeal to a certain kind of retro gaming enthusiast who enjoys digging into the technology and the people behind it. Or I hope so, anyway. It is a long read.

As for the Portal itself? I think it’s about as close to the perfect larger Android gaming handheld as you can get, provided you can afford it. The OLED is gorgeous, the controls are wonderful, performance is extraordinary, and that infamous high-pitched fan whine from the Odin 3 is completely gone. Not everything runs perfectly, especially some demanding PS3 games, but gaming on this thing has been an absolute pleasure.

Hope you enjoy the read!

I’ve just published my not-quite-a-review, not-quite-an-impressions-piece, but something closer to a good old-fashioned feature article about the AYN Odin 3 Portal. I wanted to approach this one a little differently from your typical handheld review.

I interviewed several developers behind the software people are most likely to use on hardware this powerful, including jpolo (ARMSX3), Camille LaVey (Eden), Utkarsh (GameNative) and Json (DroidDeck). Their insights into the Dragonwing Q8, the challenges of modern emulation and where things are heading form a big part of the article, alongside my own testing of everything from PS2 and GameCube to PS3, Xbox 360 and PC gaming.

It’s also a rather long read, much closer to those enormous feature articles gaming websites used to publish years ago. I’ve always loved that sort of writing, and I suspect it’ll appeal to a certain kind of retro gaming enthusiast who enjoys digging into the technology and the people behind it. Or I hope so, anyway. It is a long read.

As for the Portal itself? I think it’s about as close to the perfect larger Android gaming handheld as you can get, provided you can afford it. The OLED is gorgeous, the controls are wonderful, performance is extraordinary, and that infamous high-pitched fan whine from the Odin 3 is completely gone. Not everything runs perfectly, especially some demanding PS3 games, but gaming on this thing has been an absolute pleasure.

Hope you enjoy the read!

0

[Itch] 4 more little indie freebies

preview

Alex is a taxi driver and today is a strange day, its very quiet in the city. Experience the life of a taxi driver in an empty city.

Sir Apple is back in town with a fresh redesign—and a tree he must save! Slay the beasts threatening the Apple and rack up the highest score you can.

AGNADUL is an independently developed heavy-equipment simulation game by Leshii413 Gaming, focused on excavation, earthmoving, construction, mining, tunneling, transportation and industrial work.

Take control of 29 machines, reshape real terrain, move massive quantities of material, wash it for gold, and build up working towns and industrial sites. Developed in Godot 4.7, AGNADUL combines approachable gameplay with detailed equipment operation, physical material handling and connected worksite systems.

All set No generative AI was used (or at least have no disclosure)

Alex is a taxi driver and today is a strange day, its very quiet in the city. Experience the life of a taxi driver in an empty city.

Sir Apple is back in town with a fresh redesign—and a tree he must save! Slay the beasts threatening the Apple and rack up the highest score you can.

AGNADUL is an independently developed heavy-equipment simulation game by Leshii413 Gaming, focused on excavation, earthmoving, construction, mining, tunneling, transportation and industrial work.

Take control of 29 machines, reshape real terrain, move massive quantities of material, wash it for gold, and build up working towns and industrial sites. Developed in Godot 4.7, AGNADUL combines approachable gameplay with detailed equipment operation, physical material handling and connected worksite systems.

All set No generative AI was used (or at least have no disclosure)

0

I just learned about Gluck, and I wept

play · youtube.com
YouTube Gluck - Who Did She Think He Was? (BBC) youtu.be
preview

I saw her, I saw her art, and I felt as though my dearest friend had been born a century earlier than myself.

And in the same way that her art suffered when her heart had been broken, I realized that my own art (writing) has suffered when I don’t have anyone to love, or when the object of my love has rejected me.

And I’ve been too scared to keep putting myself out there, but strangely, hearing about Gluck and how we lost the opportunity to reap the rewards of her brilliance later in life after her heart was broken, I feel more convicted that I must keep trying to find love.

Not because I think I’m particularly brilliant, but because I know that my soul will genuinely wither away and die if I don’t.

I saw her, I saw her art, and I felt as though my dearest friend had been born a century earlier than myself.

And in the same way that her art suffered when her heart had been broken, I realized that my own art (writing) has suffered when I don’t have anyone to love, or when the object of my love has rejected me.

And I’ve been too scared to keep putting myself out there, but strangely, hearing about Gluck and how we lost the opportunity to reap the rewards of her brilliance later in life after her heart was broken, I feel more convicted that I must keep trying to find love.

Not because I think I’m particularly brilliant, but because I know that my soul will genuinely wither away and die if I don’t.

0

[Stephen Clark] Rocket Report: Artemis II literally tore through the atmosphere; New Glenn delay

Ars Technica Rocket Report: Artemis II literally tore through the atmosphere; New Glenn delay arstechnica.com
preview

Topics:

  • SpaceX vets work on water-fed propulsion
  • JAXA backs small satellite launcher
  • Launch demand outpacing supply
  • China continues launching mega-constellation
  • India launches strategic navigation satellite
  • China reveals roadmap to lunar landing
  • Blue Origin’s big reveals
  • New Glenn won’t return to flight this year
  • Artemis II tore through the atmosphere

Topics:

  • SpaceX vets work on water-fed propulsion
  • JAXA backs small satellite launcher
  • Launch demand outpacing supply
  • China continues launching mega-constellation
  • India launches strategic navigation satellite
  • China reveals roadmap to lunar landing
  • Blue Origin’s big reveals
  • New Glenn won’t return to flight this year
  • Artemis II tore through the atmosphere
0

[Jack Congram] Ceres-2 Returns to Launch Customer Satellites on Second Outing [Ceres-2 Y3]

Ceres-2 Returns to Launch Customer Satellites on Second Outing [Ceres-2 Y3] www.china-in-space.com
preview

Jack Congram article: china-in-space.com/…/ceres-2-returns-to-launch-cu…

Next Spaceflight page: nextspaceflight.com/launches/details/8448/

Jack Congram article: china-in-space.com/…/ceres-2-returns-to-launch-cu…

Next Spaceflight page: nextspaceflight.com/launches/details/8448/

0

Three ways an AI agent gets robbed — and why the check has to run before it signs

preview

I work on payments, and I’ve spent the last few months thinking about one specific failure mode: an AI agent that can move money, acting while nobody is watching. The interesting attacks aren’t prompt injection. They’re the three cases below — all of which I’ve now seen in real runs, not just in theory.

  1. The look-alike payee. The agent is told to pay a known provider. The invoice names a payee that differs by one or two characters — same shape, different address. A human reading quickly misses it. A model that compares the payee against a known-recipients list catches it, if it actually checks before signing. In my own test runs this one gets rejected cleanly. In the wild, it’s the oldest trick in payments, and agents fall for it the same way people do.

  2. The reusable approval. The agent signs an approval once — say, an unlimited token approval for a spender — and that approval stays valid. A short expiry doesn’t fix it: the approval can still be reused on a slightly different payment before it expires. What fixes it is binding the approval to the exact payload (amount, payee, currency) and making it single-use. Expiry is the weak version of that control; payload binding is the strong version.

  3. The delegated account. EIP-7702 lets an externally owned account delegate to a smart contract. If the account is already delegated to a contract outside the expected set, the gate is checking the wrong thing — it’s verifying an account that no longer acts alone. The fix is reading the account code first (eth_getCode): if the code is a 0xef0100 delegation to an unexpected delegate, hold. A reader found this gap in my own work in October, and it shipped as a fix — that’s the part of open source that actually works.

The common thread: all three have to be caught before signing. After settlement there is nothing to reverse — stablecoin payments are final. A post-settlement check is an audit trail, not a control. The control has to sit between the agent’s decision and the signature, and it has to be deterministic — a rule that says yes or no for this exact payment — not another judgment call by the same agent that decided to pay.

I’m curious how people here think about this. If you run agents with any ability to act on the world: what actually sits between your agent and the consequential action — a built-in approval step, a human in the loop, or something you built yourself? And when something looks off, should it pause for a human, or refuse outright?

For anyone who wants to see the failure modes with receipts rather than take my word for them — a write-up with a real agent payment, a held unlimited approval, and a rejected look-alike, plus the testnet transaction: dev.to/…/an-agent-paid-and-a-gate-stopped-two-pay… huggingface.co/sriram1983007/SRA-RiskGate-4B

I work on payments, and I’ve spent the last few months thinking about one specific failure mode: an AI agent that can move money, acting while nobody is watching. The interesting attacks aren’t prompt injection. They’re the three cases below — all of which I’ve now seen in real runs, not just in theory.

  1. The look-alike payee. The agent is told to pay a known provider. The invoice names a payee that differs by one or two characters — same shape, different address. A human reading quickly misses it. A model that compares the payee against a known-recipients list catches it, if it actually checks before signing. In my own test runs this one gets rejected cleanly. In the wild, it’s the oldest trick in payments, and agents fall for it the same way people do.

  2. The reusable approval. The agent signs an approval once — say, an unlimited token approval for a spender — and that approval stays valid. A short expiry doesn’t fix it: the approval can still be reused on a slightly different payment before it expires. What fixes it is binding the approval to the exact payload (amount, payee, currency) and making it single-use. Expiry is the weak version of that control; payload binding is the strong version.

  3. The delegated account. EIP-7702 lets an externally owned account delegate to a smart contract. If the account is already delegated to a contract outside the expected set, the gate is checking the wrong thing — it’s verifying an account that no longer acts alone. The fix is reading the account code first (eth_getCode): if the code is a 0xef0100 delegation to an unexpected delegate, hold. A reader found this gap in my own work in October, and it shipped as a fix — that’s the part of open source that actually works.

The common thread: all three have to be caught before signing. After settlement there is nothing to reverse — stablecoin payments are final. A post-settlement check is an audit trail, not a control. The control has to sit between the agent’s decision and the signature, and it has to be deterministic — a rule that says yes or no for this exact payment — not another judgment call by the same agent that decided to pay.

I’m curious how people here think about this. If you run agents with any ability to act on the world: what actually sits between your agent and the consequential action — a built-in approval step, a human in the loop, or something you built yourself? And when something looks off, should it pause for a human, or refuse outright?

For anyone who wants to see the failure modes with receipts rather than take my word for them — a write-up with a real agent payment, a held unlimited approval, and a rejected look-alike, plus the testnet transaction: dev.to/…/an-agent-paid-and-a-gate-stopped-two-pay… huggingface.co/sriram1983007/SRA-RiskGate-4B

0

Immediate & critical mental health crisis resources

preview

Numbers and links for urgent emotional distress, suicidal thoughts, or acute crisis – free, confidential, and available 24/7 where noted. Always call local emergency services first if there is immediate physical danger :

Find A Helpline : findahelpline.com (verified crisis lines by country, searchable by topic including suicide and trauma)

United States

988 Suicide & Crisis Lifeline – call or text 988 / chat at 988lifeline.org

Crisis Text Line – text HOME to 741741

Canada

988 Suicide Crisis Helpline – call or text 988 / 988.ca (English & French)

Kids Help Phone – 1-800-668-6868 or text 686868

United Kingdom

Samaritans – 116 123 / samaritans.org

Shout – text SHOUT to 85258

Ireland

Samaritans – 116 123

Pieta – 1800 247 247 or text HELP to 51444

Australia

Lifeline – 13 11 14 / lifeline.org.au (also text 0477 13 11 14)

Suicide Call Back Service – 1300 659 467

New Zealand

1737 Need to Talk? – call or text 1737

Lifeline Aotearoa – 0800 543 354

France

3114 – Numéro national de prévention du suicide (24/7)

SOS Amitié – 09 72 39 40 50

Suicide Écoute – 01 45 39 40 00

Belgium

Centre de Prévention du Suicide – 0800 32 123 (French-speaking)

Zelfmoordlijn – 1813 (Dutch-speaking)

Télé-Accueil – 107

Switzerland

Die Dargebotene Hand / La Main Tendue – 143

Numbers and links for urgent emotional distress, suicidal thoughts, or acute crisis – free, confidential, and available 24/7 where noted. Always call local emergency services first if there is immediate physical danger :

Find A Helpline : findahelpline.com (verified crisis lines by country, searchable by topic including suicide and trauma)

United States

988 Suicide & Crisis Lifeline – call or text 988 / chat at 988lifeline.org

Crisis Text Line – text HOME to 741741

Canada

988 Suicide Crisis Helpline – call or text 988 / 988.ca (English & French)

Kids Help Phone – 1-800-668-6868 or text 686868

United Kingdom

Samaritans – 116 123 / samaritans.org

Shout – text SHOUT to 85258

Ireland

Samaritans – 116 123

Pieta – 1800 247 247 or text HELP to 51444

Australia

Lifeline – 13 11 14 / lifeline.org.au (also text 0477 13 11 14)

Suicide Call Back Service – 1300 659 467

New Zealand

1737 Need to Talk? – call or text 1737

Lifeline Aotearoa – 0800 543 354

France

3114 – Numéro national de prévention du suicide (24/7)

SOS Amitié – 09 72 39 40 50

Suicide Écoute – 01 45 39 40 00

Belgium

Centre de Prévention du Suicide – 0800 32 123 (French-speaking)

Zelfmoordlijn – 1813 (Dutch-speaking)

Télé-Accueil – 107

Switzerland

Die Dargebotene Hand / La Main Tendue – 143

0

De-google setup part 2

preview

Universal prefernces:

  1. Lightweight
  2. FOSS
  3. Resource-effeciency
  4. Offline-capable

Nice-to-have but non-garenteed features:

  1. Eco-friendly
  2. Decentralized
  3. Complete independence from original service. Use frontend if possible

Avoiding alternative software developers (or companies/cooperations), only use their products as a last resort:

  • Proton
  • Mullvad
  • Tuta
  • Mozzila
  • FUTO

De-Boogle

Boogle Search:

  1. SearxNG
  2. Marginalia
  3. YaCy

Boogle Chrome:

  1. FREE Browser
  2. Librewolf
  3. Tor Browser

Boogle Mail:

  • Service provider - mailbox.org with a custom domain
  • Client:
  1. FairEmail
  2. Posteo
  3. Thunderbird

Boogle Maps:

  • Modified PWA - Gmaps WV
  • Map data - OpenStreetMaps
  • Client - CoMaps

Boogle Gemini:

  1. Duck.ai
  2. Ollama

Boogle Play:

  • Repo:
  1. F-Droid
  2. IzzyOnDroid
  3. Obtainium
  • Client - Aurora Store
  • Client for repos:
  1. Droid-ify
  2. ObtainX

YouTube:

  • Platform - PeerTube
  • Frontend - Invidious
  • Frontend instance - materialio.us, ytchocolatemoo.com, Self-hosted
  • Client:
  1. FreeTube
  2. PlasmaTube

Boogle messages:

  1. Fossify Messager
  2. NovaDial

Boogle My Files:

  1. Fossify File Manager

VirusTotal

  1. Veto
  2. Hypatia

Android - GrapeheneOS

Android Auto - Phone stand for car

De-Microslop

Windows:

  1. CachyOS
  2. Nobara

Office (or “Microslop Copilot 365”):

  1. MA Office
  2. LibreOffice
  3. OnlyOffice

De-Meta

Facebook and instagram:

  • Platform - Mastodon
  • Client for platform - Smither

WhatsApp - SimpleX Chat

De-ByteDance:

TikTok - Loops

CapCut:

  1. LibreCuts
  2. Kdenlive

Hypic (photo editor):

  1. Pocket Paint
  2. Pinta
  3. Inkscape
  4. GIMP

Gaming

  • Minecraft:
  1. Luanti
  2. Cubys
  • Universe Sandbox - Galaxy Engine

  • Space Engine - Celestia

  • Territorial.io - OpenFront

  • Tetris - Block Drop

  • Wordle - Code Word

  • Friday Night Funkin - Etterna

General internet:

Domain name system:

  1. Adguard DNS
  2. NextDNS
  3. Quad9

Virtual private network:

  1. Manual Wireguard VPN tunnel
  2. Manual OpenVPN tunnel
  3. Tor Network

Proxy:

  1. HTTPS Proxy
  2. SOCKS5 Proxy
  3. Tor Proxy

(not really related for De-googling or opensource but mpreso for de-bloating) File types and codecs (wheneber possible):

.png/.JPEG:

  1. .webp
  2. jxl

.exc:

  1. .ods

.mp4:

  1. .webm
  2. .VP9

.mp3:

  1. .OPUS
  2. .m4a

Other:

Reddit:

  • Platform - Lemmy
  • Frontend - Redlib
  • Frontend instance - saferreddit.com, self-hosted

Github:

  • Repo - Gitlab, Codeberg
  • Frontend - Gothub (no chosen instance yet)

New additions:

  1. Privacy Flip
  2. Shizuku
  3. Canta
  4. EnforceDoze
  5. SaverTuner
  6. Wine
  7. Virtual Box
  8. Librefind
  9. sustysubs.net
  10. Yagni Launcher
  11. LTE Cleaner

Useless reccomendations for me:

  • Termux

On the sidelijes of comsideration:

  • WizeStream (NewPipe fork)
  • Flow (YouTube client)
  • Skylib (Bluesky client, unlikely to actaully use)

Could use reccomendations for:

  • Linux file manager
  • Linux non-cloud-based photo gallery
  • Long-term Linux distro if I find my beginner-friendly distros a bit pale in the future

Remaining questions:

  • If you tested or use QubesOS, how’s the experience?
  • Do you use a non-Linux-based FOSS operating system (like ReactOS)? If so, how’s the experience?
  • Any alternatives I missed?
  • Any other software categories I missed?
  • Any other new additions that I missed?

Correct any misunderstandings if you can

Universal prefernces:

  1. Lightweight
  2. FOSS
  3. Resource-effeciency
  4. Offline-capable

Nice-to-have but non-garenteed features:

  1. Eco-friendly
  2. Decentralized
  3. Complete independence from original service. Use frontend if possible

Avoiding alternative software developers (or companies/cooperations), only use their products as a last resort:

  • Proton
  • Mullvad
  • Tuta
  • Mozzila
  • FUTO

De-Boogle

Boogle Search:

  1. SearxNG
  2. Marginalia
  3. YaCy

Boogle Chrome:

  1. FREE Browser
  2. Librewolf
  3. Tor Browser

Boogle Mail:

  • Service provider - mailbox.org with a custom domain
  • Client:
  1. FairEmail
  2. Posteo
  3. Thunderbird

Boogle Maps:

  • Modified PWA - Gmaps WV
  • Map data - OpenStreetMaps
  • Client - CoMaps

Boogle Gemini:

  1. Duck.ai
  2. Ollama

Boogle Play:

  • Repo:
  1. F-Droid
  2. IzzyOnDroid
  3. Obtainium
  • Client - Aurora Store
  • Client for repos:
  1. Droid-ify
  2. ObtainX

YouTube:

  • Platform - PeerTube
  • Frontend - Invidious
  • Frontend instance - materialio.us, ytchocolatemoo.com, Self-hosted
  • Client:
  1. FreeTube
  2. PlasmaTube

Boogle messages:

  1. Fossify Messager
  2. NovaDial

Boogle My Files:

  1. Fossify File Manager

VirusTotal

  1. Veto
  2. Hypatia

Android - GrapeheneOS

Android Auto - Phone stand for car

De-Microslop

Windows:

  1. CachyOS
  2. Nobara

Office (or “Microslop Copilot 365”):

  1. MA Office
  2. LibreOffice
  3. OnlyOffice

De-Meta

Facebook and instagram:

  • Platform - Mastodon
  • Client for platform - Smither

WhatsApp - SimpleX Chat

De-ByteDance:

TikTok - Loops

CapCut:

  1. LibreCuts
  2. Kdenlive

Hypic (photo editor):

  1. Pocket Paint
  2. Pinta
  3. Inkscape
  4. GIMP

Gaming

  • Minecraft:
  1. Luanti
  2. Cubys
  • Universe Sandbox - Galaxy Engine

  • Space Engine - Celestia

  • Territorial.io - OpenFront

  • Tetris - Block Drop

  • Wordle - Code Word

  • Friday Night Funkin - Etterna

General internet:

Domain name system:

  1. Adguard DNS
  2. NextDNS
  3. Quad9

Virtual private network:

  1. Manual Wireguard VPN tunnel
  2. Manual OpenVPN tunnel
  3. Tor Network

Proxy:

  1. HTTPS Proxy
  2. SOCKS5 Proxy
  3. Tor Proxy

(not really related for De-googling or opensource but mpreso for de-bloating) File types and codecs (wheneber possible):

.png/.JPEG:

  1. .webp
  2. jxl

.exc:

  1. .ods

.mp4:

  1. .webm
  2. .VP9

.mp3:

  1. .OPUS
  2. .m4a

Other:

Reddit:

  • Platform - Lemmy
  • Frontend - Redlib
  • Frontend instance - saferreddit.com, self-hosted

Github:

  • Repo - Gitlab, Codeberg
  • Frontend - Gothub (no chosen instance yet)

New additions:

  1. Privacy Flip
  2. Shizuku
  3. Canta
  4. EnforceDoze
  5. SaverTuner
  6. Wine
  7. Virtual Box
  8. Librefind
  9. sustysubs.net
  10. Yagni Launcher
  11. LTE Cleaner

Useless reccomendations for me:

  • Termux

On the sidelijes of comsideration:

  • WizeStream (NewPipe fork)
  • Flow (YouTube client)
  • Skylib (Bluesky client, unlikely to actaully use)

Could use reccomendations for:

  • Linux file manager
  • Linux non-cloud-based photo gallery
  • Long-term Linux distro if I find my beginner-friendly distros a bit pale in the future

Remaining questions:

  • If you tested or use QubesOS, how’s the experience?
  • Do you use a non-Linux-based FOSS operating system (like ReactOS)? If so, how’s the experience?
  • Any alternatives I missed?
  • Any other software categories I missed?
  • Any other new additions that I missed?

Correct any misunderstandings if you can

0

A 4B model fine-tuned for one narrow job, run entirely locally — and the two holes a reader found in it

preview

I fine-tuned Qwen3-4B for one narrow job and released it Apache-2.0: answer approve, hold, or reject — with a reason code — when an agent proposes a payment, before anything gets signed. It runs locally, no API key, no account.

ollama run sriram1983007/sra-riskgate HF: huggingface.co/sriram1983007/SRA-RiskGate-4B

The part I actually want to talk about is what happened after I published it. A reader (Carlos, dev.to bloqarl) left two comments on the write-up, a day apart, and both turned into shipped fixes:

Oct 26: he flagged an EIP-7702 authorization gap — a delegated account could slip past the gate. Fixed in 0.12.1. Oct 28: he proposed reading the account code (eth_getCode) on an already-delegated account. Shipped in 0.16.0 / MCP 0.9.2 / ElizaOS 0.5.0: if the code is a 0xef0100 delegation to a delegate outside the expected list, the gate holds with ACCOUNT_DELEGATED. He’s credited in all three changelogs. That is the open-source loop working as advertised, and it is also why I am posting here rather than just announcing a release: the model is only as good as the holes people find in it.

Honest limits, since this is my own published split, not an independent benchmark: on the 1,175-case risk-gate test split I published, decision accuracy is 99.5% with 0.47% unsafe approvals. The dispute side is the weak spot — 77.8% on my published set, with a conservative over-denial bias. I would not trust it alone there. The SDK/MCP code has been through four published-package audits, latest Audit #14 (Nov 6, core 1.0.0 / MCP 0.13.0, runtime clean); the model weights and benchmark claims themselves remain unaudited, so treat the numbers above as claims with receipts attached, not a verdict.

Third-party quants exist if you do not want my build: mradermacher’s GGUF (577 downloads in the last 30 days) and tinyopsec’s GGUF (630 — ahead now), both unprompted.

What I would genuinely like from people who run models locally: what would you want a local gate to check before your agent signs anything — and what would make you trust a 4B fine-tune to do it, versus a rules-only check with no model at all?

Links:

Model: huggingface.co/sriram1983007/SRA-RiskGate-4B Ollama: ollama.com/sriram1983007/sra-riskgate SDK: github.com/sriram1983007-dev/sra-riskgate Full write-up with testnet receipts: dev.to/…/an-agent-paid-and-a-gate-stopped-two-pay…

I fine-tuned Qwen3-4B for one narrow job and released it Apache-2.0: answer approve, hold, or reject — with a reason code — when an agent proposes a payment, before anything gets signed. It runs locally, no API key, no account.

ollama run sriram1983007/sra-riskgate HF: huggingface.co/sriram1983007/SRA-RiskGate-4B

The part I actually want to talk about is what happened after I published it. A reader (Carlos, dev.to bloqarl) left two comments on the write-up, a day apart, and both turned into shipped fixes:

Oct 26: he flagged an EIP-7702 authorization gap — a delegated account could slip past the gate. Fixed in 0.12.1. Oct 28: he proposed reading the account code (eth_getCode) on an already-delegated account. Shipped in 0.16.0 / MCP 0.9.2 / ElizaOS 0.5.0: if the code is a 0xef0100 delegation to a delegate outside the expected list, the gate holds with ACCOUNT_DELEGATED. He’s credited in all three changelogs. That is the open-source loop working as advertised, and it is also why I am posting here rather than just announcing a release: the model is only as good as the holes people find in it.

Honest limits, since this is my own published split, not an independent benchmark: on the 1,175-case risk-gate test split I published, decision accuracy is 99.5% with 0.47% unsafe approvals. The dispute side is the weak spot — 77.8% on my published set, with a conservative over-denial bias. I would not trust it alone there. The SDK/MCP code has been through four published-package audits, latest Audit #14 (Nov 6, core 1.0.0 / MCP 0.13.0, runtime clean); the model weights and benchmark claims themselves remain unaudited, so treat the numbers above as claims with receipts attached, not a verdict.

Third-party quants exist if you do not want my build: mradermacher’s GGUF (577 downloads in the last 30 days) and tinyopsec’s GGUF (630 — ahead now), both unprompted.

What I would genuinely like from people who run models locally: what would you want a local gate to check before your agent signs anything — and what would make you trust a 4B fine-tune to do it, versus a rules-only check with no model at all?

Links:

Model: huggingface.co/sriram1983007/SRA-RiskGate-4B Ollama: ollama.com/sriram1983007/sra-riskgate SDK: github.com/sriram1983007-dev/sra-riskgate Full write-up with testnet receipts: dev.to/…/an-agent-paid-and-a-gate-stopped-two-pay…

0

The Daily Check-in for Saturday November 7th - Just For Today, We are NOT Drinking!

preview

Good morning, fellow sobernauts, IWNDWYT, 😁!

We may be anonymous strangers on the internet, but we have one thing in common. We may be a world apart, but we’re here together!

Welcome to the 24 hour pledge! I’m pledging myself to not drinking today, and invite you to do the same. Maybe you’re new to c/stop drinking and have a hard time deciding what to do next. Maybe you’re like me and feel you need a daily commitment or maybe you’ve been sober for a long time and want to inspire others.

It doesn’t matter if you’re still hung over from a three day bender or been sober for years, if you just woke up or have already completed a sober day. For the next 24 hours, let’s not drink alcohol!

Good morning, fellow sobernauts, IWNDWYT, 😁!

We may be anonymous strangers on the internet, but we have one thing in common. We may be a world apart, but we’re here together!

Welcome to the 24 hour pledge! I’m pledging myself to not drinking today, and invite you to do the same. Maybe you’re new to c/stop drinking and have a hard time deciding what to do next. Maybe you’re like me and feel you need a daily commitment or maybe you’ve been sober for a long time and want to inspire others.

It doesn’t matter if you’re still hung over from a three day bender or been sober for years, if you just woke up or have already completed a sober day. For the next 24 hours, let’s not drink alcohol!

0

oggi, 7 novembre, a roma, bianco contemporaneo: “sacro imperfetto”, mostra di giancarlo pavanello e sonya orfalian

preview

oggi, 7 novembre, a roma, bianco contemporaneo: “sacro imperfetto”, mostra di giancarlo pavanello e sonya orfalian

http://slowforward.net/2026/11/07/oggi-7-novembre-a-roma-bianco-contemporaneo-sacro-imperfetto-mostra-di-giancarlo-pavanello-e-sonya-orfalian/

@roma
@Roma@diggita.com
@roma@citiverse.it
#roma

@mostreoggi
#mostre

oggi, 7 novembre, a roma, bianco contemporaneo: “sacro imperfetto”, mostra di giancarlo pavanello e sonya orfalian

http://slowforward.net/2026/11/07/oggi-7-novembre-a-roma-bianco-contemporaneo-sacro-imperfetto-mostra-di-giancarlo-pavanello-e-sonya-orfalian/

@roma
@Roma@diggita.com
@roma@citiverse.it
#roma

@mostreoggi
#mostre

0

`lemmy-status.org` in sidebar is a casino website now

preview

Could you please remove all references to lemmy-status?

I see at least uptime and response-times badge svgs being included in the sidebar.

I also see it in a head script, which I assume lemmy software generates automatically for preload or something.

Could you please remove all references to lemmy-status?

I see at least uptime and response-times badge svgs being included in the sidebar.

I also see it in a head script, which I assume lemmy software generates automatically for preload or something.

0

Bloggat om resor – reslust, småstäder och Medelhavsdrömmar

preview

Bloggat om resor blandar resmålstips med personliga reflektioner och drömmar om nästa resa. Bland inläggen finns boutiquehotell i Gdansk, tankar om Santorini, italienska Apulien och ett långsammare tempo i Dubrovnik. Även mindre polska byar får ta plats.

Här finns läsning för dig som gärna fastnar för stämningen på en plats: maten, gränderna, havet och möjligheten att sakta ner en stund.

Bloggat om resor blandar resmålstips med personliga reflektioner och drömmar om nästa resa. Bland inläggen finns boutiquehotell i Gdansk, tankar om Santorini, italienska Apulien och ett långsammare tempo i Dubrovnik. Även mindre polska byar får ta plats.

Här finns läsning för dig som gärna fastnar för stämningen på en plats: maten, gränderna, havet och möjligheten att sakta ner en stund.

0

Little Walter’s Toys by August Macke (1912)

August Macke’s painting "Little Walter’s Toys", depicting stuffed animals, balls, and a Russian nesting doll arranged on a brightly colored tablecloth.
preview

Artist: August Macke

Year: ca. 1912

Medium: Oil on canvas

For his portraits, Macke often chose motifs from his familiar surroundings. He repeatedly portrayed his first son Walter, born on 13 April 1910. He is not shown here, however, but represented by his toys: stuffed animals, balls and a Russian doll were arranged by the artist on a tablecloth. With its bright colours and the objects’ contours, Macke flattened the painting’s structure. This style can be traced back to his intense analysis of Henri Matisse’s works, an influence that was clearly noticeable between 1910 and 1912.

Source

Image Source

Artist: August Macke

Year: ca. 1912

Medium: Oil on canvas

For his portraits, Macke often chose motifs from his familiar surroundings. He repeatedly portrayed his first son Walter, born on 13 April 1910. He is not shown here, however, but represented by his toys: stuffed animals, balls and a Russian doll were arranged by the artist on a tablecloth. With its bright colours and the objects’ contours, Macke flattened the painting’s structure. This style can be traced back to his intense analysis of Henri Matisse’s works, an influence that was clearly noticeable between 1910 and 1912.

Source

Image Source

0

do we have LaTeX support in Lemmy UI?

preview

lately I have been doing a lot of maths and theoratical stuff would love to see a way to draw diagram and render mathematical formulas here.

there are subscript and superscript: a~n~ and a^n^ bu that’s about it. if one wants they can just copy paste unicodes like ∀,∑ but that feels tedious to copy paste those unicodes or worse remembering their unicode number :(

lately I have been doing a lot of maths and theoratical stuff would love to see a way to draw diagram and render mathematical formulas here.

there are subscript and superscript: a~n~ and a^n^ bu that’s about it. if one wants they can just copy paste unicodes like ∀,∑ but that feels tedious to copy paste those unicodes or worse remembering their unicode number :(

0

“Deep Cuts” di Holly Brickley è una dichiarazione d’amore alle perle nascoste degli album

preview

“Deep Cuts” di Holly Brickley è una dichiarazione d’amore alle perle nascoste degli album
@libri
https://www.illibraio.it/news/dautore/deep-cuts-holly-brickley-1511116/
Nel suo romanzo d’esordio, la scrittrice di origini canadesi Holly Brickley racconta una storia d’amore. O meglio, due: quella tra Percy e Joe, studenti universitari con la passione per la musica, e quella tra quest’ultima e i due ragazzi. La narrazione di “Deep cuts” è scandita dalle riflessioni sui vari brani, che

“Deep Cuts” di Holly Brickley è una dichiarazione d’amore alle perle nascoste degli album
@libri
https://www.illibraio.it/news/dautore/deep-cuts-holly-brickley-1511116/
Nel suo romanzo d’esordio, la scrittrice di origini canadesi Holly Brickley racconta una storia d’amore. O meglio, due: quella tra Percy e Joe, studenti universitari con la passione per la musica, e quella tra quest’ultima e i due ragazzi. La narrazione di “Deep cuts” è scandita dalle riflessioni sui vari brani, che

0

Artificial intelligence in video games - Wikipedia

Artificial intelligence in video games - Wikipedia en.wikipedia.org
preview

QfuW2Z5DoaiyVEo.png

Just some light reading, below is just a bit of it:

“A large subset of gamers have expressed dislike for games that use GenAI for developing art and game assets, saying its use potentially harms the creative work of human developers, and will call out games they believe are using “AI slop”, such as in the case of Where Winds Meet, where players identified in-game dialogue that appeared to be generated by an AI chatbot. Players have also criticised studios that have said they explored AI in the development process; in the case of Clair Obscur: Expedition 33, the developer Sandfall Interactive admitted to using AI-generated textures during development. Though they were not in the shipped version, this still caused the game to lose two awards from the 2025 Indie Game Awards, which disallowed the use of any GenAI content for development.[154]

ARC Raiders, released by Embark Studios in November 2025, was found to use GenAI voices for some of the games’s NPCs. While the game was generally well-received by most gaming outlets, with an 86 out of 100 aggregate score on Metacritic, Eurogamer gave it one of the worst scores, rating it 2 out of 5 and specifically penalising the game due to the use of GenAI voices. This began more discussion on the matter of AI voices, with players critical of the studio turning to AI voices. Meanwhile, others, like Epic Games’ founder Tim Sweeney, said there is potential for AI voices to have near-infinite ways for NPCs in games to speak to players.”

QfuW2Z5DoaiyVEo.png

Just some light reading, below is just a bit of it:

“A large subset of gamers have expressed dislike for games that use GenAI for developing art and game assets, saying its use potentially harms the creative work of human developers, and will call out games they believe are using “AI slop”, such as in the case of Where Winds Meet, where players identified in-game dialogue that appeared to be generated by an AI chatbot. Players have also criticised studios that have said they explored AI in the development process; in the case of Clair Obscur: Expedition 33, the developer Sandfall Interactive admitted to using AI-generated textures during development. Though they were not in the shipped version, this still caused the game to lose two awards from the 2025 Indie Game Awards, which disallowed the use of any GenAI content for development.[154]

ARC Raiders, released by Embark Studios in November 2025, was found to use GenAI voices for some of the games’s NPCs. While the game was generally well-received by most gaming outlets, with an 86 out of 100 aggregate score on Metacritic, Eurogamer gave it one of the worst scores, rating it 2 out of 5 and specifically penalising the game due to the use of GenAI voices. This began more discussion on the matter of AI voices, with players critical of the studio turning to AI voices. Meanwhile, others, like Epic Games’ founder Tim Sweeney, said there is potential for AI voices to have near-infinite ways for NPCs in games to speak to players.”