Musbin

[Theo - t3.gg @theo]

I have officially moved the majority of my dev work over to Linux. I couldn't be happier. Here's a video about how and why I made the move.

[Nav Toor @heynavtoor]

A founder in Mumbai got tired of paying SAP.

So in 2011 he wrote his own ERP.

15 years later, that ERP runs accounting, inventory, HR, payroll, manufacturing, CRM, and projects for thousands of companies.

It is called ERPNext.

It is free. GPL-3.0. 36,178 stars. Pushed today.

The math is the part that feels illegal.

SAP S/4HANA Public Cloud lists at $200 to $300 per Advanced user per month.
NetSuite starts at $999 a month plus $129 to $199 per user.
Microsoft Dynamics 365 Business Central is $80 to $110 per user per month.

A 50-user NetSuite deal lands around $77,400 a year in licenses plus $75k to $125k in implementation.

A 500-user SAP RISE contract sits between $800k and $1.2M a year.

ERPNext does the same modules. Accounting. Inventory. Payroll. Manufacturing. CRM. Projects. Sales. Purchase. Assets.

SAP implementations take 6 to 18 months.

ERPNext is one docker command. Minutes.

Rushabh Mehta started it because he could not afford SAP for his family business. He built it on Frappe, the MIT-licensed Python framework that powers it. 278 contributors. 11,873 forks. v16.25.0 shipped last week.

The whole stack is yours. Self-host it. Fork it. Strip the branding. Ship it to a client.

The ERP every Fortune 500 pays millions for is sitting in a public repo, written by a guy in Mumbai, and you can run it on a $5 VPS tonight.

(Link in the comments)

[Miles Deutscher @milesdeutscher]

The smartest thing you can do with Hermes right now:

Create a self-evolving loop that trains Her,es on your data - so every session is smarter than the last.

Session 1 → learns → logs it → Session 2 is smarter → logs that → repeat

Here's exactly how to set it up:

1. Create a Memory.md file on your desktop

Structure it like this:

## Preferences
## Corrections
## Patterns
## Lessons learned

2. Attach it to Hermes

Paste this prompt once:

"At the start of every session, read Memory.md and apply everything in it.

After every task, do three things:

• Log what worked and why
• Log what failed and why
• Analyse and write rules you'd apply next time

Never duplicate entries. Rewrite existing rules when you learn something better."

3. Once a week, run this prompt:

"Review everything in Memory.md. Identify patterns across all your logged lessons. Distill them into sharper, more general rules. Delete anything that's been superseded. The goal: fewer, better rules every week."

4. Archive before every cleanup

Copy Memory.md into a dated backup before your agent rewrites anything.

(In case of any mistakes, this is how you secure previous data)

That's the self-evolution loop that everyone should be running with Hermes right now.

Super simple, but it makes Hermes 10x more powerful.

Make sure to bookmark this so you actually maximise your agentic workflows.

[Ivan Fioravanti ᯅ @ivanfioravanti]

Today I should receive my DGX Spark, I'm new to it and I need to configure everything including remote access through tailscale by tomorrow. Any good guide to follow? Or should I go straight into Codex to drive me? 🤔

[Baidu AI @BaiduAI_News]

We’re open-sourcing Unlimited OCR — built to read long documents in one pass.

With 3B total parameters and only 500M activated, Unlimited OCR sets new end-to-end SOTA results on OmniDocBench v1.5 and v1.6.

The key innovation is Reference Sliding Window Attention (R-SWA), inspired by how humans transcribe books: keeping the source, recent context, and next words in focus, while softly forgetting what’s no longer needed.

With constant KV Cache size and lower attention cost, Unlimited OCR can transcribe 40+ pages in a single forward pass — without losing context or slowing down.

Explore the model👇:
--GitHub: https://t.co/5ZJBsEldKd
--Hugging Face: https://t.co/4FKFr9EfOu

[Erick @ErickSky]

Baidu acaba de romper una de las limitaciones más grandes del OCR actual.

Unlimited-OCR procesa documentos enteros de una sola pasada, sin chunking.

Es el siguiente paso después de DeepSeek-OCR.

REPOOO👇

[Alex Finn @AlexFinn]

Every Mac Studio is now GONE except for the 96gb one

The local AI revolution has begun and there's no turning back

I warned you about this 5 months ago. If you still haven't listened, it's not too late

No matter which computer you currently have, you can get involved:

1. Go to your Hermes/OpenClaw agent
2. Tell it to look at your computer specs
3. Tell it to find local models you can run on it (there are some models that work on 16gb Mac Minis)
4. Ask 'based on my workflows and tasks, which workflows can this model take off my plate?'
5. Have it download and load the weights
6. Use the model for at least 1 task a day

Local LLMs are clearly the future

As hardware improves and models become more efficient, you'll be able to do more and more

The thing is, you need to get started now. You need to get comfortable with loading and using these local models

Yes, the 512gb Mac Studio ship sailed and won't come back anytime soon, but you can still get experience with this incredible technology, even if you're only using a Mac Mini

Follow the steps above and I promise you'll be prepared for the future that's coming

[Mo Elgaraihy @EngMoElgaraihy]

انسى فواتير الذكاء الاصطناعي المرعبة ورسميا حقبة دفع الأموال لقراءة ملفاتك وكتبك انتهت للأبد ​

إذا كنت تظن أنك مجبر على رفع ملفات الPDF الضخمة لموديلات مثل GPT أو Claude ودفع مئات الدولارات شهريا مقابل استهلاك التوكنز فأنت تعيش في الماضي

​فريق PaddlePaddle فجر مفاجأة الموسم وأطلق أداة Unlimited-OCR المجانية والمفتوحة المصدر بالكامل بنسبة 100% لتسحق احتكار الشركات الكبرى نهائيا

[Peter Steinberger 🦞 @steipete]

This is becoming my favorite way to read Twitter. https://t.co/eykEElxzu7

[Nico @nicos_ai]

Google ha acabado con la mafia de las GPU 💀

VS Code ahora se conecta directamente a Google Colab.

→ Obtienes una GPU T4 gratuita dentro de tu editor.
→ Tus archivos locales. Su potencia de cómputo.

[Om Patel @om_patel5]

SOMEONE BUILT A VIRTUAL CAMERA THAT LOOPS A CLIP OF YOU LOOKING PRESENT SO YOU CAN LEAVE MID MEETING

it records a few seconds of you looking engaged, then loops it through a fake webcam you pick once in zoom, google meet, slack or teams.

one hotkey swaps your live feed for the loop, another brings you back.

your audio stays on the whole time

so you can step away for a minute without becoming the frozen black tile everyone notices

> shows up as a real camera in every app, no screen share hacks
> the loop doesnt read as a loop, it injects tiny stutters and catch up hitches so it looks like a slightly flaky connection
> swap in and out instantly with a hotkey
> universal, notarized, about 5mb, works on macos

the fake bad connection detail is absolutely the evil genius part

a frozen frame gets caught immediately but a feed that stutters like bad wifi is something nobody questions

[Matt Pocock @mattpocockuk]

Folks who are running GLM 5.2, how are you doing it?

What harness/provider are you using?

Getting FOMO about an open weights model for the first time

[Vik Paruchuri @VikParuchuri]

We're open sourcing a 9B model that extracts structured data from documents at near-frontier performance.

- 90.2% on our bench, vs Gemini 3.5 Flash at 91.3%
- Leads extraction models like NuExtract3 (81.5%)
- 9.5s p50 timings
- Pass JSON schema

[Mike Futia @mikefutia]

I just vibe-coded a Google Maps Lead Gen Scraper in Claude Code that pulls hundreds of local business leads in minutes 🤯

Enter a keyword, city, & state → get back names, phone numbers, emails, websites, & reviews.

All inside Claude Code.

Perfect for agencies and lead gen operators who need a fresh prospecting list every week without burning $200/month on a scraper SaaS.

If you're still building lead lists by hand...

searching Google Maps,

clicking through listings one by one,

copy-pasting phone numbers into a spreadsheet for hours just to end up with a few hundred decent leads...

This scraper eliminates the entire loop:

→ Enter your keyword (plumber, dentist, realtor, etc.)
→ Pick your city and state
→ Set the number of results you want
→ Hit search and let Apify do the scraping
→ Save the leads you like to a bookmark list

No manual searching.
No copy-pasting into spreadsheets.
No $200/month lead gen tool.

What you get:

→ Full search results with complete business info
→ A bookmark feature to save your best leads
→ Search history to track every past scrape
→ Optional Replit hosting so your whole team can use it

Built 100% in Claude Code.

I recorded a full walkthrough showing exactly how I built this from scratch, plus all the prompts I used.

Want all the prompts so you can build it yourself?

> Like this post
> Comment "LEADS"

And I'll send it over (must be following so I can DM)

[Suraj Sharma @suraj_sharma14]

If I had to build 5 projects to get hired as an Agentic AI Engineer.

I'd do this.

Project 1: Compliance Sentinel Agent
Scans new regulations daily. Auto-audits code. Generates audit-ready reports.

Saves legal teams 40+ hours/week.
Stack: LangGraph + Playwright + Legal RAG

Project 2: Autonomous Code Reviewer
4 agents debate every PR: Security, Style, Tests, Architecture.

Cuts review time by 70%.
Stack: GitHub Actions + Tree-sitter + LLM-as-a-judge

Project 3: Research-to-Deck Generator
Reads 50+ papers. Synthesizes key points. Exports editable PPT with citations.

Turns 3 days of work into 5 minutes.
Stack: Semantic Scholar API + python-pptx + RAG

Project 4: Recruiting Swarm
Parses resumes. Scores fit. Drafts outreach. Books interviews.

Increases interview show-up rates by 40%.
Stack: NER models + Vector search + SendGrid + Calendar API

Project 5: Cost-Aware Agent Router
Token budgeting. Model routing by cost/latency. Early exit on confidence.

Reduces inference costs by 60%.
Stack: vLLM + Prometheus + Dynamic model switching

Most people build chatbots.
Builders build systems that solve expensive problems.

(Bookmark & Repost)

[MIKE @mikenevermiss]

Anthropic engineer:

"you're not supposed to prompt Claude. you're supposed to build a system that prompts itself [loops]."

this is one of the best workflows I've seen in a long time

in this video he breaks down exactly how most people are building loops wrong:

- the memory file you never set up, so every loop starts from zero

- the sub-agents that 95% of builders have never split apart

- the stop condition setup that keeps loops from running forever and billing you in your sleep

- why writing one prompt a day is the slowest way to use Claude

if you've been using Claude for more than a month and still typing every task by hand, you've been running one prompt when you could be running a system of loops

instead of another prompt tonight, watch this

make sure to bookmark it before it gets buried

full guide in the article below

[Vaidehi @Ai_Vaidehi]

A Russian guy has discovered a method to learn anything ten times faster using AI!

It involves NotebookLM, Gemini and Obsidian.

[Suryansh Tiwari @Suryanshti777]

Claude Code feels completely different once you install this.

Anthropic quietly released an official plugin called claude-code-setup and it basically turns Claude Code from “pretty good” into an actual AI dev environment.

It scans your project and recommends:

→ hooks
→ skills
→ MCP servers
→ subagents
→ automations

Then sets everything up step-by-step for you.

Most people are using Claude Code completely vanilla…

which is why their experience feels messy.

The real power comes from the ecosystem around it.

Install:

/plugin install claude-code-setup@claude-plugins-official

Bookmark this before you forget it.

[Dezo @0xDezo]

JAPANESE DEV HOLDS A BOX SMALLER THAN A BOOK AND MAKES $12,000/MONTH RENTING OUT THE COMPUTE INSIDE.

Fits in one hand. NVIDIA logo on the front. Blue light glowing from the top.

Inside there are GPUs that AI companies will pay to access around the clock.

He does not build apps. He does not write prompts. He owns the infrastructure and collects the check.

Pause at 0:04 — watch him rotate the box and show the ports. That tiny device generates more monthly revenue than most developer salaries.

Most people chase the gold rush by panning for gold.

He sold the shovels and went home early.

Buy the hardware. Rent the compute. Collect forever.

Full breakdown in the video.

[Nav Toor @heynavtoor]

Someone cloned Netflix.
Then cloned Spotify.
Then cloned Instagram.
Then cloned Airbnb.
Then cloned WhatsApp.
Then cloned TikTok.
Then cloned Amazon.

Then put the source code for all of them on GitHub. For free.

Not one app. Not ten. Over 100 open source clones of the biggest apps on Earth. With source code. With demos. With tech stacks listed.

It is called Clone-Wars. 34,555 stars on GitHub.

Built by an Indian-origin developer named Gourav Goyal. He started collecting open source clones of popular apps into one list in December 2020. In March 2021, it went from 0 to 4,000+ stars in 7 days. It was on GitHub Trending for 5 days straight. Someone posted it to Hacker News and it hit #1 on the front page.

Here is what is inside.

Netflix clones. React, TMDB API, full streaming UI.
Spotify clones. Music player, playlists, search, albums.
Instagram clones. Feed, stories, likes, comments, DMs.
WhatsApp clones. Real-time messaging, read receipts, group chats.
Airbnb clones. Search, booking, maps, payments.
Amazon clones. Products, cart, checkout, Stripe payments.
TikTok clones. Short video feed, upload, likes.
Twitter clones. Feed, follow, tweet, retweet.
Slack clones. Channels, threads, real-time chat.
Trello clones. Boards, cards, drag and drop.
YouTube clones. Video player, search, comments.
And 90+ more.

Every clone has source code, a live demo, and the tech stack listed. React, Next.js, Node, Firebase, MongoDB, GraphQL, Tailwind. Every modern stack represented.

Here is why this matters.

Coding bootcamps charge $10,000 to $20,000 to teach you how to build apps like these.

Udemy courses charge $50 to $200 each. One app at a time. One framework at a time.

This repo gives you 100+ fully built apps with source code you can read, fork, and learn from. For $0.

Here is the wildest part.

The best way to learn to build Netflix is to look at someone who already built Netflix. Not a tutorial that teaches you one feature at a time. A complete, working clone with every feature connected.

You do not learn architecture from tutorials. You learn architecture from reading real projects.

100+ apps. 100+ demos. 100+ source codes. One repo.

Bootcamp: $10,000 to $20,000. Teaches 2 to 3 projects.
Udemy: $50 to $200 per course. One project each.

Clone-Wars: $0. 100+ projects. Every big app cloned.

34,555 stars. AGPL-3.0 licensed.

Every app you use. Cloned. Open sourced. Free to learn from.

(Link in the comments)