That AI Guy

The iron

The fleet that runs my federation

A federation is only as real as the hardware underneath it. Here is the actual fleet I own and hand operate myself, the same machines that run MEGAMIND and the same operating discipline I bring to a client build.

Four nodes, one mesh

Modest on purpose

The fleet is intentionally small. Reliability matters more than scale, and a private federation does not need a hyperscale cloud to do real work.

Public host

Bubbles

Debian Linux, 16GB RAM, the public web host. Nginx serves many sites, and this node is the source of truth for static content plus the federated brain binary.

AI node

VALKYRIE

macOS on Apple Silicon, 8GB RAM, running the brain on its own port with a dedicated seed corpus. Runs continuously as a managed service.

AI node

M2

macOS on Apple Silicon, 8GB RAM, running the brain plus the BlenderMind and FrankMind sub brains, each trained on its own topic specific corpus.

Research node

Thunderport

A standalone Apple Silicon node hosting MADDIE on its own port, holding the 512MB W_know weights file with more than 459K non zero weights.

The stack

What runs on every node

Web layerNginx, Let's Encrypt SSL, HTTP 2, gzip
App layerFastAPI, Go MADDIE, SQLite, PostgreSQL
DNS layerBIND9 authoritative, my own name servers
Mesh layerTailscale private network, NATS messaging, SSH key auth
Operating systemsDebian 12, macOS
StorageSSD primary, external archive drives, off site backup
AI inference, localApple Silicon, llama.cpp, MADDIE federated
AI inference, hostedUsed only where a specific client workload calls for it

Declared, reproducible, public

The default configuration for a site or system is documented in a stack file that lives alongside the code, not buried in a private note. The setup that ran the original deploy is the setup I still touch a year later when answering a maintenance question, because it never quietly drifted in a console somewhere.

[infrastructure]
host = "Bubbles"
location = "Cassville, MO"
os = "Debian 12 amd64"
nginx = "yes"
ssl = "Let's Encrypt, managed automatically"

[mesh]
tailscale = "yes"
nats = "internal only"

[ai]
megamind_node_count = 4
inference_local = "Apple Silicon"

Why I run it myself

Most studios hand a client site to a hosting platform and stop thinking about it. I run the infrastructure myself, because the engineer who knows the application stack should also be the one who knows where it actually lives in production. When something breaks, the same person who wrote the code has the access and the standing knowledge to fix it. That is rare, and it is the same standing I apply to a private federation I build for you, following the same five phase process every time.

There is a real trade off. A sudden hyperscale traffic spike is not what this fleet is built for. Most organizations that want to own their AI do not need that; they need reliability, competence, and one accountable operator, and that is what this fleet delivers.

Begin

Want hardware like this, sized to you?

The readiness assessment maps the fleet a federation for your organization would actually need.