Felipe Postigo

Al agent systems for life and business. Building in public non technical. Michigan Ross MBA

[Felipe Postigo @fpostigo]

@alexocheema @TheAhmadOsman @AlicanKiraz0 @spark_arena @NVIDIAAI @NVIDIARTXSpark @SpaceTimeViking I don’t use docked but I’m pretty sure you can. Here’s the repo https://t.co/W7gsDDXCSE everything’s done specifically for the spark, hats off to @SpaceTimeViking as it makes your life so much easier to just start working with it

[Felipe Postigo @fpostigo]

@Teknium

[Felipe Postigo @fpostigo]

@Teknium I created a personal finances app to keep track of my budget, an android ADP automation that adds groceries to my walmart cart for me, news summarizer, a forward looking version of gbrain to help me in recruiting for AI roles, created my website https://t.co/8dvWmqS0TG

[Felipe Postigo @fpostigo]

"I don't have a powerful computer or GPU, I can't use agents!"

No one said you needed one.

My Hermes agent by @NousResearch runs on an old @DellTech @DellXPS laptop with a busted GPU. I bought it 5 years ago just for Zoom calls and Spotify. It still works. My agents still run on it every single day. They run on Mac, Windows, Linux, they’re just waiting for you!

Here's how to think about it: your computer just runs your day to day. An agent is just a robot version of you doing the same stuff you already do.

The difference is the agent gets a boosted brain with AI. It thinks for you. It plans for you. It executes for you, just like when you talk to your favorite AI but one step further:
· You use it to improve an email? It will write it and send it.
· You use it to analyze data? It will create the dashboard or presentation. No more copying and pasting.
· You use it to help you plan trips? It will book the reservations and schedule it on your calendar.
· You use it to help you figure out how to get your kid calm? Well, it can’t yet, unless you build it a little robot for them to interact with, which you absolutely could!

Think of it like Netflix. You don’t need to store all their movies on your tv to watch them. You stream the pieces you need, and they play on your screen.

Agents are the same. You task them with something, and they use computers in the cloud (@OpenAI , @AnthropicAI , @NousResearch , @FireworksAI_HQ ) whoever you’re using) to do the thinking. Then they take those results and act on your own computer. Your computer is just the interface. The heavy lifting happens somewhere else.

So don’t let anyone push you into buying a Mac Mini or any other "AI computer" use what you already have. If your computer turns on, you have enough.

Yes, you can run local models. They have pros and cons, we’ll go over those in another post. But even then, a Mac Mini is probably not the answer. And you definitely don’t need one to get started.

Still wondering how to do all of this? Reach out! I’m happy to help!
https://t.co/OQOGUd7p2p
https://t.co/2tEEB4oEG1

#AI #ArtificialIntelligence #AIAgents #Automation #Tech #Productivity #FutureOfWork #Innovation

[Felipe Postigo @fpostigo]

Getting started with agents doesn't have to be hard or scary

@NousResearch and @Teknium made our life very, very easy. Just download Hermes, point it to your AI, and you're done....ok, that really doesn't help you, does it?

Let's take it step by step. You don't have to be a coder. This will be as easy as installing Spotify.

1. Step 1: Download Hermes Desktop App at https://t.co/1mqIw48nG0
a. No more going into coding apps. Just click next, next, next like every other app.
2. Step 2: Use what you already pay for: @OpenAI ChatGPT, @AnthropicAI Claude, or free Nous models. If you have ChatGPT, just use your email and password. The app does all the heavy lifting. No API keys. No configuration. Just like logging into your account.
3. Step 3: You're done. Start working with agents that have access to your computer, files, email, calendar, your whole digital life.

The question I get most: What's different from just regular ChatGPT?

With ChatGPT, you prompt, you get an answer. That's it.

With an agent, it can:
• Create files on your computer
• Run scheduled tasks overnight while you sleep
• Keep track of your meetings
• Build your website
• Automate your grocery shopping
• Whatever you need, whenever you need it

So what do you actually DO with it?

I started the same as you. Installed it, gave it access, but didn't know what to do. So I asked it: "What can you do with me?"

It gave me a bazillion options. Most were red-taped by work. My life wasn't that busy so I didn't need help scheduling things. But then I just started using it instead of web ChatGPT and things came up:
• Created my website
• Built Walmart automation so I don't worry about groceries anymore
• Had it research and contact companies for my wife's trip
• Built a personal finance app for our budget
• Whatever else comes to mind

What keeps coming back to me: keep experimenting.

If you don't download the tool, you'll never get ideas and definitely won't learn how to use it.

And if you don't like sitting in front of your computer, connect it to WhatsApp, Discord, or email. Send it a message in traffic, while shopping, on your break. A small prompt and Hermes works as long as needed. No scary Matrix window. Just seamless.

Download it. Connect your AI. Start playing around.

That's it.

If you need help getting started just let me know!

https://t.co/OQOGUd7p2p
https://t.co/dmPk56UHgu

#AI #Agents #Hermes #NousResearch #Automation

[Felipe Postigo @fpostigo]

I built an AI agent that manages my grocery list, and it says something about where AI strategy is heading.

Most AI strategy convos stay at the framework level. I wanted to see what agents actually look like running against real systems, not sandboxes:

• Reasoning across purchase history • Deciding reorder now vs. wait (no human in loop)

• Executing against Walmart's actual app Android ADB + local LLM (Qwen) + purchase frequency analysis.

Running in production for months.

The people defining AI strategy in 2-3 years are running agents, not just evaluating them.

Link in bio.

#AIAgents #AgenticAI #AIStrategy #LLM

https://t.co/ziBjcgFhMu

[Felipe Postigo @fpostigo]

@SpaceTimeViking Do you recommend Gemma 4 vs qwen 3.6 now?

[Felipe Postigo @fpostigo]

@SteveGaudio @Teknium I set it up through WhatsApp

[Felipe Postigo @fpostigo]

@LottoLabs What do you recommend for a laptop 3050 4gb?

[Felipe Postigo @fpostigo]

@NousResearch

[Felipe Postigo @fpostigo]

4GB VRAM. 32k context. 52 tok/s. Full GPU offload.
Qwen2.5-3B on a laptop RTX 3050 beats Qwen3.5-4B for local inference because speed matters more than theoretical smarts when your API is down.
KV cache quantization is the hack that makes it fit.

[Felipe Postigo @fpostigo]

@harsette @Teknium Sorry I can’t find the link to install on the nous website and Hermes saying no such beta exists, where did you download it from? Thanks!

[Felipe Postigo @fpostigo]

@harsette @Teknium Awesome thanks!

[Felipe Postigo @fpostigo]

@aijoey You got better results than using aeon’s build on vllm?

[Felipe Postigo @fpostigo]

Gpt 5.4 mini on hermes working on setting vllm to run qwen 3.6 35b something’s not right
Step 3.5 flash cleaning my computer as it was bloated
Opus 4.6 on my work computer writing code for margin analysis

Could’ve never done this before!

[Felipe Postigo @fpostigo]

@NousResearch @novita_labs still free? how do I access it? I only see step :(

[Felipe Postigo @fpostigo]

Local vs cloud test:
vLLM 27B AEON-7 NVFP4 by @SpaceTimeViking on @ASUS GX10 vs @fireworksai_hq kimi 2.6 turbo firepass

Raw: 81 vs 20 tok/s
Tools: 2.9s vs 4.2s, 100% vs 80% acc

Fireworks = speed+reliability
vLLM = privacy

Running via Hermes of course by @NousResearch @Teknium

[Felipe Postigo @fpostigo]

Damn flux and Lora keep failing. Restarting whole process bringing down from 18k pictures to best 200 using qwen running on vllm. Let’s see how this goes