Lu Ta Yauk

[OmedTheVibeCoder @OmedVibeCodes]

Cheapest AI coding options right now:

• Freebuff — Free
• Gemini CLI / AI Studio — Free
• Command Code Go — $1

• OpenCode — Free / Go: $5 → $10
• Cline — Free / Pass: $5 → $10

• Qoder — $20 ⭐
• ChatGPT Plus — $20 ⭐
• Cursor — $20 ⭐

What’s your favorite right now? 👇

[Eric Cole @erichustls]

My lawyer works for some of the richest folks in New York.

Yesterday he asked me:

Do you have a "death folder" for your family.

I said no.

He replied:

If you love your family, prepare these death documents right now:

(Every smart millionaire has these)

[Rimsha Bhardwaj @heyrimsha]

A guy named Max at Carnegie Mellon just open-sourced an autonomous AI company for free.

It's called Auto-Company. You start a loop. Fourteen AI agents wake up, read their shared memory, form a small squad, and get to work. Then they sleep briefly and do it all again. Forever.

Each agent has a real mental model behind it, not a generic "you are a helpful assistant" prompt.
> The CEO thinks like Jeff Bezos.
> The CTO thinks like Werner Vogels.
> The engineer thinks like DHH.
> The marketer thinks like Seth Godin.
> A dedicated "inversion" role thinks like Charlie Munger and runs pre-mortems on every idea.

Fourteen roles in total, plus a library of 30+ reusable skills like deep research, scraping, financial modeling, SEO, and security audits that any agent can load on demand.

The workflow is designed to converge, not just talk.
> Cycle 1: the team brainstorms ideas.
> Cycle 2: Munger runs a pre-mortem, Thompson checks the market, Campbell checks the unit economics. GO or NO-GO.
> Cycle 3 onward: build, deploy, ship. Pure discussion loops are forbidden.

The memory between cycles is a single markdown file called consensus dot md. No vector database. No orchestrator bloat. Just a plain text baton the team hands off after every cycle. You edit that one file to steer the whole company.

It runs on Claude Code or Codex CLI, with a native daemon on macOS, Windows, and Linux. There's a local dashboard that shows live cycle status, model spend, and which agents are currently active.

Auto-Company is still early. It costs real money per cycle, stability isn't guaranteed, and it can't compete with a real company. But the idea is huge.

For years, only VC-backed labs could afford to experiment with autonomous agent stacks like this. Now the blueprint fits in a git clone.

[z @zaynmcps]

NVIDIA ESTÁ REGALANDO ACCESO A MÁS DE 80 MODELOS DE IA Y CASI NADIE SE HA DADO CUENTA

a través de sus APIs alojadas tienes acceso gratis a:

→ MiniMax M2.7
→ GLM 5.1
→ Kimi 2.5
→ DeepSeek 3.2
→ GPT-OSS-120B

se conecta directo con Hermes, Cursor y OpenCode

configuración:

1/ saca tu api key en https://t.co/TLcyPycgrb
2/ base_url = “https://t.co/eVRBmtO2A1”
3/ api_key = “$NVIDIA_API_KEY”
4/ elige modelo (ej. minimaxai/minimax-m2.7)

tantos proyectos están regalando acceso estos días, solo hay que aprovecharlo

[Jun Song @jun_song]

Download these models from HF right now :

> GLM-5.2
> Qwen3.6-27B/35B
> Minimax-M3
> Deepseek-V4-Flash
> HY3

[Siddhant Garg @siddhantgarg33]

People think learning AI takes months.

It's really just a couple of hours.

And I wrote 17 free guides to start right away:

Claude 101: https://t.co/Puwp6c7006
Claude Code: https://t.co/2xdVTzC7SJ
Claude Skills: https://t.co/DhSrLuVe0H
Nano banana 2: https://t.co/zo5mfb5lEr
Claude in Excel: https://t.co/EX4I3fFvvx…
Best AI for Search: https://t.co/GB6eWmrAIk
1M followers with AI: https://t.co/w9O5QkDcmI
Claude for your team: https://t.co/EuuT0yAz7Q…
No prompt saves you: https://t.co/LZIcLQEMRN
AI Slides (PPT in 2026): https://t.co/wWdQD3TIdk
Set up Claude Cowork: https://t.co/ESJo766gqI…
Claude to sound like you: https://t.co/X0dSJSBAeb…
Claude interactive charts: https://t.co/3oSNfVOEMu
Claude as your computer: https://t.co/JxBRQPFWOa…
Claude Cowork + Project: https://t.co/ESJo766gqI…
You're an AI workaholic: https://t.co/2kvVezAj5B
Setup AI before prompting: https://t.co/Fn2czdGYz6…
___

1. Save this list for later (three dots, top right).
2. Share it with a friend by ♻️ reposting this image.

[smrati tiwari @smratitiwa86867]

Everyone is paying for AI voice tools...

Meanwhile, this free app clones your voice in just 10 seconds.

It runs completely on your computer.

Here's why it's blowing up:

• Clone your voice in ~10 seconds
• Multilingual voice cloning
• Local speech-to-text (dictation)
• MCP support for AI agents
• Works with Claude Code, Cursor & Cline
• Open source (MIT licensed)
• No monthly subscription

Once you try it, paying for basic AI voice tools starts to feel unnecessary.

Repo link 👇

[Julian Goldie SEO @JulianGoldieSEO]

Kimi K3 is the most powerful open-source AI model ever, and you can use it 100% free right now.

Go to https://t.co/EljZZ8tKe0. Sign in. Switch the dropdown from K2.6 to K3.

That's it. No credit card. No API setup.

On my own tests, it beat GPT 5.6 and Opus 4.8.

It builds games. 3D worlds. Full websites. Even videos.

It goes fully open source in July 2026.

Use it off-peak hours and you get even more free tokens.

Most people will never even try it.

The ones who do get frontier-level AI for $0.

[Jun Song @jun_song]

Best Local AI models by VRAM size (7/18) :

Phone (~4gb) : Bonsai-27b
Nano (~12gb) : Gemma4-12b
Micro (~36gb) : Qwen3.6-27b
Small (~128gb) : Deepseek V4 Flash ds4
Medium (~256gb) : GLM-5.2-2bit
Large (~512gb) : GLM-5.2-NVFP4

[Jun Song @jun_song]

Best AI models by use cases (7/18) :

Frontend : Kimi-K3
Backend : Fable
Debug : GPT-5.6-SOL
Image gen : GPT
Translation: Gemini-3.5-Flash
Search : Grok-4.5
Video gen : Seedance 2.0
Best price : Deepseek V4 Pro
Local AI (Large) : GLM-5.2
Local AI (Small, 128gb) : HY-3, DSV4 Flash

Use models accordingly

[Dr. Priyam Bordoloi @DocPriyamMD]

Pancreatic cancer is notoriously silent. After diagnosis, the average survival time is often just 6 to 12 months.

​Because it hides so deep in the body, catching it early is your only real chance.
​5 EARLY WARNING signs you can't ignore.

1. ​Painless Jaundice.

[hunter @hxxntrr]

Hackers leaked your social security number in 2024. Yours and 272 million others

Everyone panicked about identity theft. Nobody noticed the leak came with a weapon attached

That breach can wipe collections off your credit report in 4 business days

It's called FCRA Section 605B and the bureaus pray you never learn it exists

Here's the mechanism:

When your identity is compromised, federal law gives you the right to BLOCK information on your report that resulted from it. A block is a completely different animal from a dispute

A normal dispute under Section 611 gives the bureau 30 days to "investigate." They outsource this overseas and rubber stamp whatever the collector says

A 605B block gives them 4 business days to remove the item. No investigation. No back and forth with the collector. The law says block it, so they block it

The play:

Step 1: Go to identitytheft. gov and file an FTC identity theft report. The National Public Data breach exposed 2.9 billion records. Your name, address history, and SSN are being sold in bundles for less than a Chipotle order. You have standing. Use it

Step 2: Pull all 3 reports. Flag every account, collection, and inquiry you don't recognize as yours

Step 3: Before you send anything, freeze the shadow bureaus. LexisNexis, Innovis, SageStream. These are the databases the big 3 quietly use to re-verify data. Frozen source, nothing to verify against

Step 4: Send each bureau the 605B block request with your FTC report attached. Certified mail

Step 5: Count to 4 business days

"that sounds too easy"

The bureaus built the dispute system to wear you the fuck down. 30 days per round, form letter responses, "verified as accurate" with zero explanation. They win by attrition

605B skips the entire maze. Congress wrote it after realizing bureaus kept re-reporting fraud victims' accounts over and over. The 4 day clock has teeth. Miss it and the bureau is liable

The collectors spent years profiting off data they bought for pennies. Then a hacker sold the entire country's data for pennies and accidentally handed 272 million people the strongest deletion tool in the FCRA

The breach was the bad news. This is the half nobody read

(i fix credit in 30-90 days. link in bio)

[Jason @Jasoncoderx]

iOS 26 update is making your iPhone battery drain faster?

This isn’t normal.
After the update, Apple quietly turned several important privacy settings back ON without telling you.

These settings are draining your battery in the background.

[Mr. Buzzoni @polydao]

NVIDIA is giving away free access to 140+ AI models for a full year

> most people building AI agents are paying $50-200/month for API access

NVIDIA just made that argument irrelevant

models you get: GLM 5.2, MiniMax M3, Nemotron-3-Ultra-550B-A55B, Kimi K2.7, and 130+ more

setup:

> step 1 - get your free key
> go to https://t.co/9OAPVB6SrM
> register -> bind phone -> copy API key

> step 2 - add to Hermes agent
> open Settings -> Model Provider -> Custom

base_url = "https://t.co/ndvYuRyWSB"
api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx"

> step 3 - pick a model

model = "z-ai/glm-5.2"
model = "minimaxai/minimax-m3"
model = "nvidia/nemotron-3-ultra-550b-a55b"
model = "moonshot-ai/kimi-k2.7"

> Hermes already has NVIDIA set as default base_url
> paste the key and you're running instantly
> works the same in Cursor and OpenCode

> cost: $0
> limit: 40 req/min
> expires: 1 year

while everyone is paying for API access,
this is sitting there for free

with this many models in one place, it is also the easiest way to test which one actually fits your agent before you commit to a paid provider

[阿泽 AZe @Chenzeze777]

Windows 要慌了

有人写了一个开源 PowerShell 脚本，51 K+ stars，把 Windows 自带的所有垃圾——预装应用、广告追踪、Copilot、Recall、Bing 搜索——一条命令全清干净

微软往你电脑里塞 Candy Crush、塞 OneDrive 弹窗、塞 Bing 网页搜索、塞 Copilot 侧边栏、塞 Xbox、塞定向广告。Win11 Debloat 一条远程命令全干掉了：

· 一键卸载所有预装应用和推广软件，干净
· 彻底禁用遥测追踪、诊断数据、广告 ID，干净
· 移除 Copilot、Recall、Click to Do 等 AI 监控功能，干净
· 禁用 Bing 网页搜索和 Edge 的 AI 功能，干净
· 恢复 Windows 10 经典右键菜单，不折腾
· 启用 Windows Sandbox 和 WSL，不花钱
· 支持命令行参数批量部署，导出配置跨设备同步
· MIT 开源，不收费不捆绑不追踪

说白了，微软把 Windows 当成广告牌和遥测收割机的时候，Win11 Debloat 反着来——把你的电脑真正还给你，不收一分钱不留一行隐私数据不弹一个广告

🔗 https://t.co/r0BeIn9PaS

[Arzoo Ai @Arzoo12sh]

THERE'S AN IPHONE SETTING THAT FEDERAL AGENTS CHECK BEFORE EXECUTING A WARRANT FOR YOUR ICLOUD DATA.

IF IT'S TURNED ON, THERE'S NOTHING THEY CAN DO.

MOST PEOPLE HAVE NEVER HEARD OF IT.

HERE'S WHAT IT IS AND HOW TO TURN IT ON:

[Nav Toor @heynavtoor]

Your grandmother has one photo from 1962.

It is faded. Torn at the corner. Her face is blurry. The details are gone. You have looked at it a hundred times and wished you could see her clearly.

Professional photo restoration costs $50 to $500 per photo. MyHeritage bundles photo enhancement into a $119 to $259 per year genealogy subscription. Remini adds a watermark on the free tier and limits your daily uses. Adobe Photoshop costs $55 a month.

A researcher at Tencent ARC Lab in Shenzhen named Xintao Wang built an AI that restores damaged face photos. For free.

It is called GFPGAN. 37,000+ stars on GitHub. Apache 2.0 licensed.

You upload the photo. The AI studies the face. It reconstructs every detail. Eyes. Teeth. Hair. Skin texture. Wrinkles. Expression. The things that make her face HER face.

The result looks like the photo was taken yesterday.

Upload any old photo. See the result in seconds. In your browser. No download. No account.

Here is what GFPGAN restores:

Old family photos from the 1940s to 1990s. Faded, yellowed, damaged.
Low-resolution photos from early digital cameras and flip phones.
Blurry photos where faces are barely recognizable.
Compressed photos from old social media downloads.
Damaged photos with scratches, tears, and water stains.

The AI does not invent a new face. It uses facial priors learned from millions of faces to reconstruct the most likely original detail. The identity stays. The damage disappears.

Here is the math:

Professional photo restoration: $50 to $500 per photo.
MyHeritage: $119 to $259/year. Bundled with genealogy.
Remini: watermark on free tier. Paid plans start at $6.99/week.
Adobe Photoshop: $660/year.

GFPGAN: $0. Unlimited photos. No watermark. No subscription. Apache 2.0.

Runs on CPU. No GPU required. Can be combined with Real-ESRGAN to restore the background too.

37,000+ stars. Built by Xintao Wang at Tencent ARC Lab in Shenzhen. Apache 2.0.

The photo of your grandmother you wished was clearer.

It can be. Right now. For free.

[Amarillo @anyelamarillo]

🚨 UN PROGRAMADOR SOLO ACABA DE LOGRAR LO QUE GOOGLE LLEVA AÑOS IGNORANDO

Desarrolló Obscura, un navegador headless en Rust creado específicamente para agentes de IA, scraping masivo y automatización.

Solo usa 30 MB de RAM.
Las p��ginas cargan en 85 ms.
Bloquea automáticamente 3.520 trackers.
Elimina anuncios, analíticas y todo script de rastreo.

Se llama Obscura y tiene lo que Chrome nunca podrá darte:
Cada sesión genera una huella digital completamente distinta.
GPU, canvas, audio, batería… todo se randomiza en tiempo real.

Ningún detector de bots lo identifica porque se comporta exactamente como un Chrome real.

Es el reemplazo directo de Puppeteer y Playwright.
Sin Node.js. Sin dependencias pesadas. Un solo binario ligero.

Ya supera las 19k estrellas en GitHub. 100% open source. Totalmente gratis.

Si trabajas con scraping, IA o automatización… esto no es solo una mejora. Es un cambio de juego.

Guarda este post antes de que se te olvide. Es una de las mejores herramientas que han salido este año 💎

[Kanika @KanikaBK]

🚨 CHINA IS ON FIRE!

Tencent open-sourced TencentDB Agent Memory.

Long-term memory for AI agents that runs 100% locally.

↳ No Pinecone or cloud APIs
↳ You don't need to repeat yourself every session
↳ 61% fewer tokens
↳ PersonaMem accuracy: 48% → 76%
↳ Runs on plain SQLite
↳ FREE. Open source.

5.1k stars in 48 hours.

Most people paying $70/month for vector databases will miss this.

[Nico @nicos_ai]

GANA $1.2M AL AÑO Y TRABAJA SOLO

Cuando le preguntaron a este ingeniero senior de Anthropic cómo rinde como un equipo entero, no explicó nada.

Simplemente compartió su carpeta .claude/ completa.

Mismo modelo. Resultados radicalmente diferentes.

La mayoría sigue eligiendo entre Opus y Sonnet como si el modelo fuera el límite.

No lo es.

La verdadera ventaja no está en qué modelo usas.

Está en qué despiertas dentro del modelo.

Este es el stack que usan los que operan a otro nivel:

→ CLAUDE.md: el contrato maestro
→ settings.json: permisos y configuración
→ hooks/: los reflejos automáticos
→ agents/verifier: el policía que revisa todo
→ skills/: 33+ memorias musculares reutilizables
→ .mcp.json: las herramientas conectadas
→ MEMORY.md: el registro vivo de cada turno

Ya no chateas con el modelo.

Escribes la carpeta una sola vez y la carpeta ejecuta el modelo por ti.

Este es el sistema que separa a los que “usan Claude” de los que lo tienen trabajando como un equipo de élite.

Desglose completo + estructura exacta lista para copiar y pegar 👇

Guarda este post antes de que desaparezca.🔖