I WINS. Hermes Helps.
I run Hermes agent on a $5 VPS. It handles my stock research, blog distribution, invoicing, expense tracking, and wiki.
My life runs on four pillars: Wealth, Insights, Network, and Self. Together they spell WINS, a name that captures what the system is meant to produce. I already had frameworks for each of those four pillars. Hermes didn't create those. It accelerated them, and in some cases changed how I do them entirely.
I use Hermes across three of those pillars (Wealth, Insights, and Self). Here's what it does in each one. Every example is live in this site.
The pattern across every use case is the same: use the LLM to build the tool, then use the tool without the LLM. Each use case compounds because they all operate on the same set of markdown files.

# The Setup
Hermes runs on a $5 VPS in a container. I access it through Slack, which requires no public exposure. Slack's Socket Mode is outbound only, so Hermes dials out and holds the connection.
Read the full infrastructure write-up →
Most of what I do runs through skills (reasoning tasks) or Python scripts (repetitive tasks). If I do something once, I let it be a one-off. If I do it again, I ask Hermes to create a skill. If the task is deterministic, Hermes writes a Python program instead. The token budget stays low.
# Wealth
# Stock Research with Actual Discipline
I invest in three kinds of companies. Dividend compounders, mostly Indian PSUs, where the government owns majority stakes and likes dividends. Growth compounders, quality businesses with sustained PAT CAGR above 20 percent, priced reasonably. S-curve growth stocks, a type of growth compounder riding structural shifts, evaluated on a 7-parameter framework. The Indian market doesn't have many of these, but when I find one, they tend to be multi-baggers.
Hermes navigates to Screener.in, pulls financial data, including technical charts, runs a computation script, and populates a structured analysis template. Every position gets entry gates and exit triggers written down in advance. When I batch multiple analyses, Hermes processes them one at a time via cron jobs.
The real value isn't the specific companies. It's the discipline. Hermes surfaces the companies I could invest and ranks them. I apply my judgement and decide if I should invest or not. There is also clear review schedule. Together, it helps me build a wealth compounding machine.
Read the full stock screening write-up →
# Invoice Generation
I don't have many clients; I manage invoicing simply. I posted a Word doc template to Hermes and asked it to generate an HTML template instead. The first pass was wrong. I took the broken HTML to Claude, iterated until it was right, then pasted the corrected template back into Hermes. Now I fill in the fields and Hermes generates PDF invoices. Deterministic, repeatable, no surprises per invoice.
The pattern is the same as everything else: use the LLM to build the tool, then use the tool without the LLM.
# Expense Tracking
My wife and I share expense tracking through Hermes. I type "1000 for lunch paid via ICICI card." Hermes appends it to the expenses file. Blind append, never reads the file back, never wastes context tokens. She posts expenses the same way in our shared Slack channel. Fuel expenses get parsed separately: car model, fuel type, liters, odometer reading. All feeding into hledger for monthly reviews.
Slack on mobile becomes the central point. I coordinate everything from there, whether I'm at my desk or not.
# Insights
Insights is a loop: consume, produce, engage. Hermes keeps my learning loop running smoothly
# Building a Knowledge Base That Compounds
I share a YouTube link and say "extract transcript and add to wiki." Hermes fetches the transcript, spins up sub-agents in the pipeline, and writes interlinked markdown pages into my personal wiki. Cross-references, timestamps, verbatim quotes.
The agent saves transcripts and routes that saved transcript to another sub-agent. On a flat-rate OpenCode Go plan, it is effectively free.
I drop an article URL and it does the same. I drop a PDF on Dropbox and it does the same.
The wiki lives on my server, synced to Dropbox, and I access it through Obsidian on my phone and laptop. I can pull up a video summary while I'm out.
# Writing: From Voice to Published Post
The hardest part of writing was starting. Voice recording fixed that. Hermes fixed the rest.
I record my thoughts with Yap on my Mac or VoiceNotes on mobile. Sometimes I record directly in Slack. Hermes has Whisper installed, so it can transcribe without any LLM cost. If I already transcribed outside, I paste the transcript and Hermes rewrites it with the blog skill.
As they say in writing circles, "it is easier to fix a bad page than a blank page." What used to take an hour of staring at a blank page now starts with a five-minute voice recording. I talk. Hermes structures according to blog_skill. The blog skill enforces specific guidelines: follow ASD-STE100 style, Flesch reading ease of 65 to 70, controlled language, first person conversational style. No AI slop. No "delve into," no "leverage," no "it's worth noting." I trained myself to follow Minto Pyramid style of writing. I brought that to agentic writing too. Every paragraph has a governing thought. The first sentence carries the argument.
My manual editing focuses on one thing: does the argument hold together from top to bottom? Do the paragraphs connect? Does the essay flow as a single coherent piece? I don't worry about word choices or grammar. As a non-native writer, I always had difficulty with those choices. Now, hermes handles that. I focus on whether the point comes through. If the output sounds like me, I'm not worried it was transcribed, structured, and drafted by an AI agent.
# Diagramming and Image Generation
I run two blogs: CTO Field Notes and jjude.com. Both need diagrams. Hermes generates them as self-contained HTML files with inline SVG: dark-themed architecture diagrams, hand-drawn Excalidraw flowcharts, editorial infographics with brand-specific skins for each blog.
I also generate featured images for blog posts. One prompt, one image, a few cents on Gemini Flash. I store the prompt alongside the generated image so I can regenerate or iterate later.
Read the image generation details →
# Distributing Content to Social Media
I've been blogging for years, over 300 posts across jjude.com and CTO Field Notes. Distributing them on social media is the natural next step, but writing platform-specific posts for LinkedIn, X, and Bluesky is tiring work.
The pipeline pulls from my top 100 posts by page views, hermes reads each post and generates three formats in one pass: LinkedIn post, tweet, and thread. I review, tweak, and schedule through Typefully.
Read the full distribution write-up →
# Self
# The Homeschooling Book
I'm working on The Indian Homeschooling Playbook. I already wrote and published the book as series of blog post. Now I piped all the markdown files through hermes to make it Gumroad-ready. Hermes manages the project: chapter drafts, style guide, diagnostic checklists. I dictate, Hermes structures. I review, Hermes refines.
Hopefully, I publish it soon.
# What I Haven't Solved Yet
I want my boys to use Hermes too. So far I haven't found an easy way to create different memory and profiles for everyone in the family. The shared expense channel works for now, but a proper multi-user setup would unlock Hermes as a family tool, not just mine.
# The Interface Is Simple. The Infrastructure Compounds.
When I extract YouTube transcripts into my wiki, that knowledge is available to every future query. When I log an investment thesis with exit triggers, the system reminds me to review it and has the historical context to make that review meaningful. When I dictate a blog post, the transcript becomes a draft, the draft becomes a published post, and the published post becomes social media content.
Hermes isn't a smarter chatbot. It's an agent that writes files, runs scripts, spawns sub-agents, navigates the web, schedules cron jobs, and maintains a growing knowledge base. Each use case builds on the previous ones because they all operate on the same set of markdown files.
The agentic journey has just began.
This is an ongoing project. If you want to follow along with how this setup evolves, subscribe to my RSS feed, or connect with me on LinkedIn and X.
Under: #aieconomy , #hermes , #tools , #tech