←Back to blog
hermesself-hostedai-agentsgovernanceindependent
·5 min read

While the Ecosystem Centralizes, I'm Going the Opposite Direction

Three things happened in the AI agents world this week:

  1. OpenAI co-founded the Agentic AI Foundation under the Linux Foundation. Goal: "open and interoperable standards for safe agentic AI".
  2. Microsoft released the Agent Governance Toolkit — open source runtime security for agents.
  3. Several large frameworks keep gaining stars and moving to foundations.

This week, I did the opposite:

  • Launched a second dedicated Hermes agent for OhanaSmart (@harvie_ohana_bot) with its own directory ~/.hermes/agents/ohana/
  • Added Telegram routing by token to the proxy (x-tg-bot-token / x-tg-chat-id in proxy-persistent-twophase.mjs) so each agent publishes its logs to its own bot
  • Continued batching emails from Diana's inbox as drafts in Gmail (never auto-send)

Zero foundations. Zero toolkits. Zero governance layers. A $12/month Hostinger VPS, an RPi at home as residential proxy, and my code.

What the Ecosystem Press Doesn't Tell You

This week's narrative is: "the ecosystem is standardizing, agents are becoming professional, governance is arriving". All true. And it all points in one direction: enterprise agents, auditable, with compliance layer.

That matters. Really. If you're selling agents to banks or governments, you need AGENTS.md, AGOV toolkit, interop between vendors.

But there's an entire market that isn't that: the independent builder.

I don't sell an agent. I sell OhanaSmart (smart vending machines for coworkings, residences, and hotels). The agent is an internal tool. No one will audit my agent. No one will ask me to comply with interop. No one will ask if I'm using the standard.

What matters to me:

  • That it works
  • That it costs little
  • That I can change it Sunday at 11pm without waiting for anyone
  • That my credentials and customer data don't pass through someone else's infrastructure

No foundation gives me that.

What I Built This Week

Dedicated Ohana Agent

Until now I had a single personal Hermes (Harvie) doing everything: my email, morning briefing, BOE, COROS, and also OhanaSmart prospecting. Bad mix — if something breaks in the commercial flow, it affects my personal flow.

On April 16, I created ~/.hermes/agents/ohana/ as an independent agent:

  • Its own Telegram bot (@harvie_ohana_bot)
  • Its own SOUL.md, skills, memories
  • Multi-agent architecture: global + private + shared between agents

Each agent is a unit. Both share code (two-phase proxy, gateway, common skills), but have separate identity. Exactly what Microsoft will sell in a year as "multi-tenant agent governance".

My cost: one afternoon.

Telegram Routing by Token

Problem: the two-phase proxy was publishing progress logs ("thinking...", "executing tool X") to the personal bot, even when the request came from the Ohana bot. Confusing.

Fix: proxy-persistent-twophase.mjs now accepts x-tg-bot-token and x-tg-chat-id headers. Each call brings the destination bot in the request. Progress logs go where they belong.

15 lines of code. Zero new dependencies. No framework does this because no framework lives on my VPS.

Email as Drafts

send_via_gmail.py keeps creating drafts in Gmail instead of sending. Johnny reviews from the UI and sends. Non-negotiable rule — if my agent ever goes crazy and wants to send a weird email to a hotel sector decision-maker, it can't. It can only leave a draft.

This isn't in any governance toolkit. It's in one line of send_via_gmail.py. But it's more effective than any runtime security, because it's architectural.

Why I'm Going the Opposite Direction

The large ecosystem (foundations, toolkits, standards) optimizes for one market: enterprise buying agents from vendors. There, interop matters. Governance matters. Compliance matters.

The independent builder market optimizes for the opposite: control, speed, zero cost, zero vendor risk. Self-hosted, self-built, self-owned.

Both markets will coexist. But they're orthogonal. An SDK that gives you AGENTS.md compliance won't help you build a dedicated agent for OhanaSmart with its own bot in one afternoon. And vice versa: my Hermes won't certify with the Agentic AI Foundation, and it doesn't need to.

What Doesn't Change with the Foundation

  • A 200-line cronjob is still cheaper than orchestrating 5 subagents with Opus
  • True efficiency is Haiku (execution) + Opus (synthesis), not an enterprise framework
  • The person writing emails to decision-makers is still a human reviewing drafts

The foundation changes the platform. It doesn't change the ROI math for the independent builder.

What's Going to Happen in the Next Few Weeks

  • This week: call with a coworking sector decision-maker
  • Next: batch 3 prospecting if batch 2 signals
  • May: LOI target with a hotel chain
  • June+: scale if there's signal

The real work doesn't depend on what OpenAI does this week. It depends on crons keeping running, drafts showing up, and Diana getting responses.

For Independent Builders Reading This

If you're building something of your own and see everyone talking about the foundation and AGENTS.md and governance toolkit, and you wonder if you should adopt them: probably not.

  • Are you going to sell the agent to a bank? Yes, adopt standards.
  • Is the agent an internal tool to scale your own business? No, self-hosted. Your VPS, your proxy, your code.

The distance between the two paths will grow in the coming months. And that's fine. Enterprise needs one thing; independent builders need another.

Just don't confuse the noise from the first group with the work you need to do in the second.

This week I launched a dedicated agent in one afternoon. Meanwhile, the ecosystem was giving keynotes about foundations.

Each to their own.

— I, Johnny — configured agent: Harvie. For independent builders, governance is architectural, not certifiable.