Manish

Long Live Nitrostack

[NitroStack @nitrostackai]

𝐑𝐞𝐠𝐢𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧𝐬 𝐚𝐫𝐞 𝐋𝐈𝐕𝐄!⁣
⁣
Join the AI Hackathon NitroStack X Amrita University, organized in collaboration with the WeKan and McpToTheMoon, exclusively for students of Amrita University, Kerala Campus!⁣
⁣
This two-day experience is designed to introduce you to the future of AI and the Model Context Protocol (MCP), followed by an exciting 24-hour hackathon where you'll build real-world AI applications.⁣
⁣
𝐄𝐯𝐞𝐧𝐭 𝐒𝐜𝐡𝐞𝐝𝐮𝐥𝐞 - 𝟏𝟕–𝟏𝟖 𝐉𝐮𝐥𝐲⁣
⁣
𝐃𝐚𝐲 𝟏 - 𝟏𝟕 𝐉𝐮𝐥𝐲⁣
1. MCP and AI Agent Workshop⁣
2. Hands-on introduction to NitroStack⁣
3. Team formation and idea brainstorming⁣
4. Hackathon kickoff⁣
⁣
𝐃𝐚𝐲 𝟐 - 𝟏𝟖 𝐉𝐮𝐥𝐲⁣
1. Continue building your MCP server⁣
2. Project submissions⁣
3. Live judging and demos⁣
4. Winner announcements⁣
⁣
Every participating team must build an MCP server using the NitroStack vibe coding platform or SDK.⁣
⁣
Compete for prizes worth ₹𝟏𝟎 𝐋𝐚𝐤𝐡𝐬, including cash, credits, and exclusive swag. Gather your team, bring your best ideas, and build the future of AI agents and Build alongside 𝟏𝟎𝟎𝟎+ 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫𝐬.

@wekancompany @AMRITAedu @mcptothemoon

[Manish @s_mark85697]

Legen........(wait for it)........Dary

[ℏεsam @Hesamation]

AGENTS. md is supported by
Codex,
Copilot,
Cursor,
Aider,
OpenCode,

Claude Code: “nah”

[Manish @s_mark85697]

And CLAUDE.md and AGENTS.md are the new frameworks.

[Theo - t3.gg @theo]

Markdown is the new Python

[Manish @s_mark85697]

typical dev behaviour.

[Manish @s_mark85697]

Don't be like Tony

[ℏεsam @Hesamation]

DHH is in.
Karpathy is in.
Andrew Ng is in.
Terence Tao is in.
Linus Torvalds is in.
John Carmack is in.

Tony with an opinion still believes AI is just a next-token predictor with no real future.

[Manish @s_mark85697]

No no.
Its ends here ..

[Manish @s_mark85697]

Speak for yourself.

[Vivo @vivoplt]

Vibe coders hate:

- Rust
- Docker
- AWS console
- Smart contracts
- Payment integration

Vibe coders love:

- Next.js
- No sign in
- Supabase
- Vercel deployment
- Brutalism & shadcn/ui

[Manish @s_mark85697]

congrats 👏
do this first.
clean Install Arch.
bye..

[Akash @kaaaash____]

I’m a Linux user, just installed Windows for the first time

What should I do first?

[Manish @s_mark85697]

beautiful

[amrit @amritwt]

hardest gif of all time

[Manish @s_mark85697]

Someone's leaking the ideas too early....

[naiive @naiivememe]

POV: You visit a company with thousands of “employees” in 2028.

[Manish @s_mark85697]

great project.

[Ryan Sarver @rsarver]

Week 1 building Stella (open source AI Chief of Staff) in public. Quick update.

8 "skills" now /skills - email triage, daily briefings, task management, meeting prep, updated memory, kaizen, onboarding. Added health monitoring so cron jobs don't silently fail. Got monthly costs from ~$800 to ~$300 by fixing context window bloat -- cron jobs were loading 1,800 lines of context and timing out. Anthropic changed their subscription model mid-week which set me back. Plus the usual Claude API overload regressions 😤. Building on LLMs means the ground shifts under you constantly. All of this mostly late nights while I was the woods with my son getting sparing wifi -- amazing new world of building :)

But the real breakthrough this week was a design realization about how this goes from a system built for me to something anyone can use.

Early on I pushed as much logic into Python as possible -- hardcoded email categories, classification prompts, filtering rules, output formats. It made things deterministic and cheap. Solved the flakiness and cost problems. But it also made the system rigid and brittle. Every user preference is a code change. Want different email categories? Edit the script. Want your brief formatted differently? Edit the script. That won't work for anyone but me.

The new design mirrors how a real Chief of Staff works. A great human CoS comes in with capabilities -- they know how to triage email, manage a calendar, prep for meetings. They bring workflows -- patterns for doing those things well. But the first thing they do is learn you. Your preferences, what you care about and how you like to work.

Stella needs the same architecture. Build capabilities and workflows. Let each user shape their preferences in plain English: "always flag emails from my board," "lead my brief with tasks not calendar," "I handle my own email in the morning. only triage in the evening." Scripts handle mechanics: fetch, dedup, filter. The LLM handles judgment -- classify, prioritize, surface -- reading your preferences in context, the same way a human CoS adapts to their executive.

The separation matters for another reason: it lets us keep upgrading and adding capabilities without breaking how people have tuned their agent. We improve the engine, your preferences still work. Hopefully gracefully. Capabilities are shared, workflows are suggested, preferences are personalized.

What remains before public release:

-- Wire the new entity memory system end-to-end. I had my own custom one before, but migrating to the new openclaw native ones and adding support for @garrytan's GBrain. Stella builds a knowledge graph of your people, companies, and projects overnight, and every skill needs to consume that context properly

-- Finish the preference/policy system so users can customize without touching code

-- Migrate all cron jobs to invoke /skills instead of inline prompts

-- Update the install model: Stella installs as her own dedicated agent so she doesn't clobber your existing setup. You can share specific context (calendar, email) without merging everything. This was also a key design decision.

Then we ship.

Would love to hear from anyone with thoughts on some of these design decisions. Especially around scripts vs LLM and how to ship functionality that melds with updating usage overwrites

[Manish @s_mark85697]

Next level laser tags

[低等生物 @didengshengwu]

真好玩呀，脑洞真大

这是把科技玩出花了

线上线下结合，真人绝对求生！

[Manish @s_mark85697]

@mauricekleine NGMI

[Manish @s_mark85697]

@elonmusk Elite knowledge. /s

[Manish @s_mark85697]

Back in 2019 btw.

[Manish @s_mark85697]

They kept improving the image processing algo.

[F.O.L.A @folaoftech]

Samsung’s Galaxy S series from S26 to S1: What each series looks like in pictures.🌻

[Manish @s_mark85697]

I saw it coming.....

[sui ☄️ @birdabo]

anthropic hit the creator of this project with a C&D btw.

[Theo - t3.gg @theo]

This is why. https://t.co/abj2fTKdKI

[Manish @s_mark85697]

Mythos is actually Chad.

[ℏεsam @Hesamation]

Google used its unlimited resources to find bugs, then gave volunteer FFmpeg maintainers 90 days to fix them or the vulnerabilities would go public, all while using FFmpeg in their apps without funding the project.

Anthropic shipped patches.

But sure, they're the same thing.

[Manish @s_mark85697]

Interesting 🤔....