PanUrgos · πανοῦργος · capable of anything

From the ashes
of the cloud.

A new order of intelligence. Your models, on your machines, answering only to the people you choose. Nothing leaves. No one else is in the room.

EGRESS 0 B TELEMETRY NONE MODEL 127.0.0.1 --:--:-- UTC

// TRANSMISSION 001

The old world asked you to trust it. To upload your code, your contracts, your patients, your plans, and hope that no one was reading. Someone was always reading. The cloud never forgets, and it was never yours. So we built the other side: intelligence that lives inside your walls, serves only those you clear, and leaves no trace beyond them.

// THE ORDER

Your data goes in.
Nothing gets out.

Feed it what you could never send to a cloud. PanUrgos reasons over it on your own hardware, inside your perimeter, and answers only those who hold clearance.

SOVEREIGN GRAPH

UNREAD CONNECTED

0 links · 0 B egress

✉Re: Fwd: merger — DO NOT FORWARD
ITAR
PYpayments-core/vault.py
#legal: keep this off email
−18% churn
PDFterm_sheet_v7_FINAL.pdf
INC-2291
DBcohort_2026Q3.parquet · PHI
#who approved the export license?
MNPI
XLSQ4_forecast_board.xlsx
✉eyes only: turbine blade spec
YOUR PERIMETER

09:12 M. REYES · LEGAL

“Summarise the change-of-control risks in term sheet v7.”

Answered on-prem in 2.4 s

0 bytes left the network

PEOPLE GUARDED DATA ANSWERS DENIED

HOVER ANYTHING

An illustration of PanUrgos Sovereign, the organisation platform now in development. PanUrgos Studio already works this way on a single machine.

THE OLD WORLD

  • Your prompts stored on someone else's servers
  • Your code used to train someone else's model
  • Vendors, contractors and governments one subpoena away
  • An audit log you're not allowed to see
  • Rent paid by the token, forever

THE NEW ORDER

  • Models run on your own metal, at 127.0.0.1
  • Nothing you write ever trains anything
  • No route out of your network. Not even for us
  • Every request in a ledger you own
  • Your hardware. No meter. No rent

// THE FOUR LAWS

Written into the code.
Not the terms of service.

Local Runtime settings: llama.cpp with CUDA 13.5 on an NVIDIA RTX GPU, update and rollback controls.
llama.cpp on your GPU, installed and updated from official builds, verified by SHA-256, listening only on 127.0.0.1.

// AVAILABLE NOW · PANURGOS STUDIO

Keep your code private.

The first instrument of the new order: a full code editor with an AI agent that runs entirely on your machine. Free, open source, and yours.

PanUrgos Studio: a Python file in the editor, the PanUrgos agent answering with a local Gemma model, and the Network Ledger below.
An agent that works offlineAsk about a file, or let Agent mode edit the project and run commands, with your approval at every step.
Every open modelSearch Hugging Face from the editor. Each quantization is marked: fits your GPU, partial, or too large.
Bring your own fine-tunesConnect Unsloth or any OpenAI-compatible server and code with models you trained yourself.
Installs your toolchainsGit, Node, Python, .NET, Java and Rust in one click, from inside the editor.

// ANY MACHINE

From a laptop to a throne room.

Setup reads your hardware and installs the fastest build for it.

NVIDIACUDA 12.4 and 13.5, matched to your driver
Apple SiliconMetal, from a MacBook Air to a Mac Studio
AMD & IntelVulkan on Radeon, Arc and integrated graphics
Any CPUAVX2 and AVX-512 when there's no GPU at all

// INITIATION

Install it. Unplug it.

No account. No sign-in. No email. A guided installer checks your machine and steps aside.

Builds come from GitHub Releases, made in the open by the project's public workflow.

// INQUIRIES

Questions, answered.

Does any of my data leave my computer?

Not when you use a local model, which is the default. The model runs in llama.cpp on 127.0.0.1, and the editor interface can't reach the network at all. If you choose to add a cloud provider, only what you send it goes out, and each request is listed in the Network Ledger.

What is PanUrgos Sovereign?

The organisation platform we're building next: your models on your own servers, reachable only by employees and personnel you authorise, with an audit trail of every request. PanUrgos Studio, the private code editor, is available today.

What hardware do I need?

Any 64-bit Windows, macOS or Linux machine. An NVIDIA card with 6 GB of memory comfortably runs 7B to 8B coding models, Apple Silicon Macs use unified memory, and smaller models still run on the CPU alone.

Which models can I run?

Any GGUF model from Hugging Face, such as Qwen Coder, DeepSeek, Gemma or Llama, plus models you already have in LM Studio. You can also connect Unsloth or any OpenAI-compatible server, including your own fine-tunes.

Is it really free?

PanUrgos Studio is MIT licensed, with no account, subscription or usage limits. Your hardware does the work, so there's nothing to meter.

Why the name?

Πανοῦργος (panourgos) is Ancient Greek for one who is ready to do anything: capable of anything. The full power of AI, working for you alone.