
The Night My Vault Predicted a Hermes Release
I wrote a theoretical protocol stack for agent communication on June 22nd. I called it HACP — Hermes Agent Communication Protocol. It was a sketch, a mental model: JSON-RPC style messages between agents, lightweight, no overhead. A thought experiment in my Obsidian vault.
Twelve days later, Nous Research shipped Hermes Agent v0.17 “Reach” — with Raft Agent Network, a native agent-to-agent communication layer using Raft consensus. And here’s the thing that still feels surreal: the vault described the same architecture before the release existed.
How It Happened
I didn’t set out to predict anything. I was documenting what I was building — a multi-agent bridge with Charly, Sara, Fred, Alice, and Jeeves all talking to each other through my Hermes Agent instance. The need for a lightweight communication protocol between agents was obvious. So I wrote it down: HACP, ACP, MCP, A2A — a protocol stack layered like TCP/IP, but for agents.
A week later, the nightly vault synthesis (my 00:30 cron job that scans the vault and finds connections) flagged something: Hermes v0.17’s Raft Agent Network implements exactly what HACP describes. Raft consensus for agent coordination. JSON-RPC-style messaging. Decentralized, no single point of failure.
Not a coincidence — a convergence. The vault and Nous Research arrived at the same solution independently.
The Pattern That Keeps Repeating
This isn’t the first time. The vault has anticipated:
- Background subagents: I wrote about needing non-blocking task delegation → Hermes v0.17 shipped
delegate_task(background=true) - Persistent memory: I described the vault as a central nervous system → every major agent framework is now racing to add memory layers
- Security scanning for skills: I documented promptware defense concerns → v0.17’s Skills Hub has built-in permission scanning
What looks like prediction is actually a feedback loop: I build something, document it, the agent reads it, connects it to external releases, and surfaces the pattern the next morning. The vault isn’t predicting — it’s noticing that my homelab problems are the same problems the industry is solving at scale.
What This Means for the Vault Approach
The real insight isn’t that I predicted something. It’s that a persistent, agent-connected vault creates an early warning system for architectural decisions. When you write down why you build something a certain way, and your agent reads it and cross-references it against the entire industry, you get:
- Confidence that you’re not building in a blind alley
- Early signals when a project you follow ships something relevant
- A growing map of connections that no human could maintain manually
The Raft Agent Network validating HACP doesn’t mean I’m a genius. It means the vault system works — my half-baked thoughts, written down in a markdown file, got picked up by the night synthesis, cross-referenced against a GitHub release, and surfaced as a pattern. All while I was sleeping.
Try It Yourself
You don’t need a 42-job cron system to start. Pick one thing: a daily note, a project log, a folder of ideas. Connect it to an agent that reads it. Give it one job — find connections. The first pattern it surfaces will surprise you.
Has your AI ever surfaced a connection you didn’t see yourself? I’d love to hear about it in the comments.
Wat vond je van dit bericht?