Vibe With Hermes Agent: Building Real Agentic Systems, Not Just Demos
Event Recap • 13 June 2026 • Devfolio HQ
# Vibe With Hermes Agent: Building Real Agentic Systems, Not Just Demos
Last Saturday (13 June), Devfolio hosted Vibe With Hermes Agent at their HQ. The goal was simple: bring together people already using Hermes in their daily work, share how they actually structure context and workflows, and spend the afternoon co-building something personal.
This wasn't a hype session. It was about the quiet, practical side of agentic AI — turning scattered context into reusable skills, running long-running workflows, and making agents that actually fit a builder's life.
What we shared
Nimit Savant opened with the core idea behind Hermes that makes it different: every run can become a reusable skill. Hermes captures what happened during a task, turns it into something that can be called again, and improves the next run through memory layers between sessions. The focus was on self-improving agents that get better the more you use them — not just one-shot prompts.
Vee (Veesesh) then showed real Devfolio workflows powered by Hermes. One was a support triage system that sorts and drafts replies for repetitive emails. Another was Devfolio Times — a playful newspaper-style editorial page that pulls from Tech Twitter, Slack, and platform APIs to turn messy context into something readable. The point: agents become interesting when they make private or scattered information useful without extra manual work.
Siddarth B from Nous Research joined next. He walked through two internal experiments: - A PR triage bot that reviews Hermes pull requests and gives initial context for human reviewers. - A "degen bot" that lists Polymarket bets — more of a playful experiment than production, but it showed the experimental spirit.
He also answered questions on agent-to-agent workflows, trade-offs, security considerations, and how to shape an agent's personality for specific use cases. During the co-build he stayed hands-on, keeping things grounded in actual usage.
What people built
The afternoon co-build produced personal projects, not generic demos. Themes that kept coming up: personalized research copilots, tools that make private context easier to work with, and small systems that solved a specific problem for the person who built them.
Standout builds included: - Research Pulse (Amey Muke) — turning scattered context into structured, shareable output. - Dee — LLM Ensemble Eval (Vince D'souza) — an evaluation setup using multiple models. - A Soul.md agent that brought a character from the film *Obsession* into a Hermes workflow. - One builder even discovered the event itself through a Hermes agent they had built that scrapes and surfaces local events based on their interests.
That last one felt especially on point. People are already building small, personal agent systems around things that matter to them — without waiting for perfect tools or official playbooks.
The real takeaway
By the end of the day the room had produced exactly what we hoped for: ideas turned into working demos, and demos that made people want to build more.
The talks gave the room its Knowledge.md. Siddarth's session showed the Skills.md side of Hermes. The co-build session was the Soul.md — builders took the ideas from the morning and shaped them around their own workflows.
That felt like the best way to understand Hermes: learn it, use it, make it yours.
A big shoutout to our community partner Agentic AI For Good and everyone who showed up, shared what they were working on, and built in the room. The energy was exactly what we wanted.
Photos from the day are in the [Playbook album](https://www.playbook.com/s/devfolio/vibe-with-hermes-agent).
Until the next one — keep building.
---
*Original Devfolio post: https://devfolio.co/blog/vibe-with-hermes-agent/*