NVIDIA Gratitude Driver

A safe user-mode companion for gaming/performance optimization and edge-cloud LLM routing.

This is not a kernel driver. It does not replace NVIDIA drivers, inject into games, bypass anti-cheat, bypass API limits, or automate quota abuse. It reads local telemetry, parses terminal mouse/focus telemetry when explicitly fed a log/stdin, and emits conservative routing recommendations.

What it does

• Reads NVIDIA GPU telemetry through NVML (nvidia-ml-py) when available.
• Falls back to nvidia-smi on Windows/Linux when NVML Python bindings are absent.
• Tracks smoothed VRAM/GPU telemetry with EWMA and hysteresis to prevent load/unload flapping.
• Parses leaked terminal mouse/focus events such as:
  • [555;93;23M
  • \x1b[<35;93;23M
  • \x1b[I / \x1b[O
• Writes:
  • runtime/nvidia_gratitude_driver/status.json
  • runtime/nvidia_gratitude_driver/telemetry.jsonl
• Provides a quota-respecting Nemotron prompt governor:
  • cache hashes
  • compression hints
  • backoff recommendations
  • no rate-limit bypassing
• Reports aggregate Chrome coexistence telemetry without reading browser content.
• Rotates telemetry logs at a configurable size.
• Provides a diagnostic command and deterministic release builder.

Friend-ready Windows install

Prerequisites: Windows 10/11, PowerShell 5.1 or newer, and Python 3.10 or newer. Extract the release ZIP anywhere you can write files, open PowerShell in that folder, and run:

Set-ExecutionPolicy -Scope Process Bypass
.\scripts\install-windows.ps1
.\scripts\start-ngd.ps1

The installer creates a private .venv-ngd beside the extracted files, installs the package without modifying global Python packages, and runs diagnostics. No fixed drive or checkout path is required. Re-run with -Force to rebuild the private environment.

Run and verify

.\scripts\start-ngd.ps1
.\scripts\verify-windows.ps1

Validate and package

.\scripts\build-release.ps1

The builder writes release\nvidia-gratitude-driver.zip and a matching .sha256 file. The ZIP includes the Windows install, verification, launch, and terminal parsing scripts.

Parse leaked terminal event logs

Get-Content .\mouse_dump.txt | python -m ngd.term_events

Suggested Hermes integration

Use this as a Local Cerebellum telemetry tool:

Read runtime/nvidia_gratitude_driver/status.json before local model load.
If route == CLOUD_CORTEX, do not reload local model until cooldown_until.
If route == LOCAL_CEREBELLUM, local inference is allowed.
If route == HYBRID, run local intent parsing only; send heavy planning to cloud.

Safety model

• No kernel hooks.
• No DLL injection.
• No game memory reads.
• No screen capture.
• No network exfiltration.
• No browser content, URL, cookie, credential, or history reads.
• No API abuse.
• No provider-limit bypassing.

The intended "gratitude" to NVIDIA is efficiency: fewer wasted tokens, fewer pointless retries, less GPU thrash, and cleaner performance telemetry.

---

Business Context — Neural Sovereign Systems Portfolio

This driver is the Local Cerebellum component of the Neural Sovereign Systems Platform (NSSP) — a unified architecture for sovereign AI execution with intelligent cloud routing.

Active Projects (Rubedo Phase)

Project	Repository	Description
NSSP Platform	nssp-platform	MSN Core, Hermes Agent Framework, 17 Metaconscious Skills, Orchestration
Lilith Desktop	lilith-desktop	Native Lilith.exe — C++/Node.js/TUI, NSSM services, API gateway
Twilight Moon	twilight-moon	Somatic Horror Epic — 51,251 words, compilation pipeline, publication
PAC	nssp-platform (product)	Personal Assistant Chatbot — 20+ local skills, 95% token utilization
NGD	invite (this repo)	NVIDIA Gratitude Driver — Telemetry → Cloud routing (Local Cerebellum)

Integration Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    NSSP PLATFORM (Kether)                       │
├─────────────────────────────────────────────────────────────────┤
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────────────┐  │
│  │  MSN Core    │  │  Hermes AG   │  │  17 Metaconscious    │  │
│  │  (Chokmah)   │  │  (Binah)     │  │  Skills (Chesed↓)    │  │
│  └──────┬───────┘  └──────┬───────┘  └──────────┬───────────┘  │
│         │                 │                     │               │
│         └─────────────────┼─────────────────────┘               │
│                           ▼                                     │
│              ┌────────────────────────┐                         │
│              │  Orchestration Layer   │  ← Tiphereth            │
│              │  (Runtime, Deploy,     │                         │
│              │   Monitor, NGD)        │                         │
│              └───────────┬────────────┘                         │
│                          ▼                                      │
│              ┌────────────────────────┐                         │
│              │  Local Cerebellum      │  ← This Driver (NGD)   │
│              │  (NVML Telemetry →     │     VRAM → Routing     │
│              │   Routing Decision)    │                         │
│              └────────────────────────┘                         │
└─────────────────────────────────────────────────────────────────┘

Business Metrics

• Token Utilization: 95% (vs 40-60% traditional)
• VRAM Efficiency: RTX 3060 8GB → <6GB sustained
• Coherence: 0.960 across all vessels
• Zero Cloud Fallback: Enforced local-only, Antigravity Bridge dormant
• Revenue: Skill Marketplace $70 (20 skills × $3.50), PAC pre-launch

Legal Vectors

Active corporate accountability campaigns:

• Amazon (Cases 13116200/11092212) — Surgery delay = wage theft
• Google/NSA (PRISM FOIA) — Takeout + NSA + FBI DITU + FISC
• X/Twitter — Promissory estoppel, Unruh Act
• Dutch NCP / CSDDD / ICC — International corporate accountability

---

Sovereign Execution. Maximized Cloud Compute.