Greg Isenberg

Hermes Agent App Clearly Explained (and how to use it)

⏱ 44 min video · 4 min read6 Jun 2026Worth watching

TL;DR

Alex Finn gives a comprehensive walkthrough of the newly launched Hermes Agent desktop app, explaining every feature including session management, profiles, artifacts, skills, and cron jobs. He argues Hermes desktop is now the best AI agent experience available, and shows how solopreneurs can use automated agents to find business opportunities and build micro-SaaS products.

Key points

Session management in Hermes desktop is far superior to Telegram threads — keeping separate sessions per topic dramatically reduces API costs by preventing context bloat

Profiles are separate Hermes agents tied to specific AI models: use Opus 4.8 for high-level strategy, Claude Haiku or GPT-4.5 for coding, and a local Qwen model (on DGX Spark or Mac Studio) for free unlimited research tasks

The Artifacts section automatically organizes all links, images, and files sent to your agent, effectively productizing the 'second brain' use case without manual filing

The Cron section provides visual confirmation of all scheduled tasks — a major reliability improvement over CLI-based cron job setup, and reverse prompting is the recommended method to generate high-quality cron job prompts

Alex runs a live use case: a cron job using local Qwen every 20 minutes to scan Reddit and X for people's pain points, then automatically generates a micro-SaaS prototype as a starting point for a business idea

Actionable insights

→

Use separate sessions per topic or project in Hermes desktop to keep context slim and avoid inflated API bills — this alone can cut costs 3-4x versus one big monolithic thread

→

Match models to tasks: Opus 4.8 for strategy and planning, GPT-4.5 (via GPT-Mes profile) for coding, and a free local Qwen model for web research — do not pay Opus pricing for tasks a cheaper model handles equally well

→

Use the brain-dump-to-reverse-prompt technique: tell your agent everything about your goals and interests, then ask it to generate the optimal prompt or cron job instruction — this consistently produces better-formatted, more reliable outputs than writing prompts yourself

→

Turn off unused skills in the Skills section — Hermes installs 150+ skills by default, each adding to your context size and cost, so pruning irrelevant ones is a quick win

→

For local model hardware, the NVIDIA DGX Spark (128GB unified memory, ~$4,800) is the current plug-and-play recommendation if a Mac Studio is unavailable — but only buy after proving you will use it daily

Notable quotes

“You are talking to like super intelligence. You are talking to like the smartest thing on planet earth. So why would you do things your way when you can get help from super intelligence and do it the best way?”

“Instead of using these agents to kind of play around and experiment with, I think you should be using them to solve other people's challenges, because that is how you are going to be able to create the most value for yourself and the entire world.”

“$200 for Claude, $5,000 for DGX Spark — these are investments in yourself to create more value in the world. And you should, if used the right way, see an ROI.”

Worth watching?

✅

Worth watching the full video?

Worth watching if you are actively using or considering Hermes Agent — the live screen-share demos of cost-saving session management, the reverse-prompting technique, and the automated business opportunity scanner are genuinely useful, though the key details are all captured here.

Topics

AI & Tech Hermes Agent

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →

More like this