GGG Sovereign Edge Daemon v1.4.5

πPi-Engine Metabolic Pacer

Admission control where time is pi: work is admitted only while the pressure integral over the pacing window stays strictly inside the circle. The irrational bound never resonates with periodic load, so the node converges to its sustainable rhythm instead of oscillating into overload.

⇄Non-Zero-Sum Vault Balancer

When infinite negative data weight (tombstone debt) meets positive stored weight, classical accounting collapses to NaN. The balancer binds opposing infinities into preserved dipoles held in equilibrium tension — balance is maintained, never annihilated.

⩘π-Gate AI Admission

Every request bound for the local Ollama instance passes the pi gate first — payload weight is normalized asymptotically, so even unbounded data is scheduled without breaching the window. Loopback allowlist enforced before any socket opens.

⌁VS Code — Continue Extension

Paste this model block into ~/.continue/config.json. Continue then talks to the daemon instead of any cloud — completions never leave the machine.

{
  "models": [{
    "title": "GGG Local (π-gated)",
    "provider": "openai",
    "model": "qwen2.5-coder:latest",
    "apiBase": "http://127.0.0.1:11834/v1",
    "apiKey": "ggg-local"
  }]
}

Any OpenAI-compatible provider entry works — the daemon enforces the loopback allowlist regardless of what the client requests.

◈Custom Web UIs & Scripts

Open WebUI, LibreChat, or your own frontend: set the OpenAI base URL and any placeholder key. With the official openai Python SDK it is three lines:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:11834/v1",
    api_key="ggg-local",  # loopback only
)
resp = client.chat.completions.create(
    model="qwen2.5-coder:latest",
    messages=[{"role": "user", "content": "ping"}],
)

Raw REST works too: POST http://127.0.0.1:11834/chat with {"model": …, "messages": […]}.

⩘Why It Stays Sovereign

Swapping tools never swaps your trust model: the π-gate schedules every payload, the router refuses non-loopback upstreams, and each response carries the cryptographic brand — no matter which UI sits in front.

Start the proxy with ggg-daemon.exe --serve (or let install-startup.bat register it in shell:startup so it is always on).

1Install a Friendly AI App

Desktop apps like LM Studio or Jan give you a chat window that feels like any messaging app — no programming required. Download one, pick a model from its built-in catalog, and you're 90% done.

These apps handle the AI model for you. The daemon adds the security layer around it.

2Point It at the Daemon

Every such app has a settings field called something like "server address", "API base", or "custom endpoint". Type in http://127.0.0.1:11834 — that's it. Your app now talks to the daemon instead of the internet.

The address never leaves your computer. 127.0.0.1 literally means "this machine".

3Let an AI Walk You Through It

Stuck on a step? Hand this prompt to any AI assistant and it will explain, in plain English, exactly where to click in your specific app:

I am not a developer. I use a desktop AI
chat app (LM Studio / Jan / similar), and
I have the GGG daemon on my computer.

Please give me simple, numbered,
click-by-click instructions — no code, no
technical jargon — for setting my app's
"server address" or "API base URL" field
to this local address:

  http://127.0.0.1:11834

Then tell me how to check that it worked,
and what to do if my app shows an error.

Works with any AI assistant — including the one inside LM Studio or Jan itself once you're connected.