Garza Global Graviton LLC released version 1.4.0 of their sovereign edge computing platform, a system that runs entirely on local hardware with no network dependency, no cloud telemetry, and no vendor access to data. The update makes the local web interface fully responsive on mobile devices and adds a cryptographic attestation framework that lets buyers verify their workloads ran in a genuinely air-gapped environment.
What a Synthetic Data Center Actually Looks Like
The core concept is what the company calls a Synthetic Data Center. Each node is a single piece of local hardware organized like a biological organism. Four cooperating daemons manage the system, none of which ever touch the network.
The Metabolic Heart paces admission and pulses the node's cadence at microsecond precision. The Liver scrubs stale memory in place using vectorized zeroization, leaving no residue. The Lungs move data through a zero-copy ring buffer, shuttling bytes without duplicating them. The Immune System acts as a zero-trust scanner that quarantines any byte pattern it does not recognize.
Every buffer is a single preallocated memoryview over a numpy array. Data moves through the organism without a single extra copy. Because the node is air-gapped, every daemon runs at local hardware speed instead of network speed.
The Ollama Bridge Stays Local by Construction
The platform includes a local Ollama bridge for running inference through models like Llama 3.2 3B. The bridge only accepts localhost, 127.0.0.1, and [::1] endpoints. Remote hosts are rejected before any request is ever sent, keeping the bridge 100% offline by construction.
Developers can point VS Code extensions at the same Ollama daemon on port 11434 for fully offline AI assistance. The repository includes ready-made configurations for Continue.dev and Ollama extensions. GitHub Copilot does not support custom local model providers, so the project recommends Continue or an Ollama extension for air-gapped coding workflows.
Cryptographic Proof of Air-Gapped Execution
The most technically dense addition is the airgap attestation system under the airgap_attestation/ directory. It gives enterprise buyers cryptographic proof that a sensitive manufacturing or data pipeline ran in a strictly air-gapped, tamper-evident environment, verifiable entirely offline without ever seeing the vendor's proprietary source code.
The protocol works through a commit-reveal scheme. The client submits a blinded hash of their sample with a salt. The air-gapped enclave responds with a commitment ID and nonce. The client then delivers the encrypted sample out of band through a separate channel. The enclave runs the job, building an AuditManifest that records process start and end events, network interface snapshots, and syscall counts.
These events get sealed into a domain-separated Merkle tree that avoids the classic odd-node duplication forgery bug. The TPM or enclave produces a QUOTE over the nonce and Merkle root. The platform signs everything with Ed25519, producing an AttestationBundle that the buyer can verify offline using the provided CLI tool.
The verification checks the Merkle root, TPM quote, platform signature, nonce freshness, and the zero-network invariant. Exit code zero means passed. Exit code one means a real check failed and the buyer should not proceed with purchase.
Container Hardening Goes Beyond Network Namespaces
The Docker deployment applies multiple independent layers of network isolation. The docker-compose.yml sets network_mode: none, which removes the network namespace peer entirely. The Firecracker configuration goes further by specifying no NIC device exists at the guest level, a hypervisor-level guarantee stronger than network namespace isolation.
A custom seccomp profile denies every network syscall at the kernel level, providing defense in depth under the network none configuration. The container uses a distroless nonroot base image with no shell, no package manager, and uid 65532. The rootfs is read-only and all capabilities are dropped.
Any single layer being misconfigured does not compromise the others. The architecture document spells this out explicitly: the controls are designed so that a failure in one does not cascade.
Windows Service Integration and Daemon Management
For Windows deployments, machine_boost_service.py wraps the supervisor as a native Windows service using pywin32. It starts at boot rather than at user logon. The supervisor runs five daemons on different intervals: watchdog and system watchdog every five minutes, snapshot every hour, mesh ping every minute, and health monitor as a persistent self-looping process that auto-restarts if it exits.
The environment hardening for child processes includes setting PYTHONIOENCODING to UTF-8, PYTHONUTF8 to 1, and PYTHONUNBUFFERED to 1. Without these, daemons crash with UnicodeEncodeError when running under a Windows service context where no real console exists. The supervisor also passes encoding="utf-8" explicitly to subprocess calls, because text=True alone still uses the OS locale for decoding.
The repository deliberately excludes traffic_daemon.py and live_cluster_daemon.py from the supervisor. Both generate synthetic client orders tagged with real defense contractor names against a local API with a hardcoded key. The project treats this as something that should not run unattended at every machine startup fabricating usage data attributed to real companies.
What This Means for High-Security Computing
Garza Global Graviton's platform occupies a specific niche: organizations that need local AI inference and data processing without any network exposure. The airgap attestation system addresses a real problem in defense and manufacturing procurement, where buyers need evidence that workloads ran in isolation without trusting the vendor's word.
The biological daemon metaphor is unusual but the underlying engineering is straightforward. Zero-copy memory management, hardware-level network isolation, and cryptographic verification are well-understood techniques applied here in combination. The 60-test suite with 84 percent coverage covers most of the attestation package, with the gap being the real TPM 2.0 code path that requires physical hardware to test.
For developers building systems that need to prove isolation, the commit-reveal attestation protocol and offline verification CLI offer a concrete starting point. The project is open source with binaries available for Windows, macOS, and Linux.