A public home for my local-first AI, agent, automation, and personal-intelligence experiments.
This repository contains Gumbii Digital's independent work. Official NVIDIA DGX Spark playbooks remain in the NVIDIA repository; my hardware notes, modified workflows, scripts, and measured results live here instead.
- Local-first AI agents
- Voice-first and wearable AI workflows
- Browser/computer automation
- Local inference and GPU experiments
- Personal knowledge systems
- Public-data and OSINT interfaces
- Prefer user-owned devices and local processing
- Keep private data private by default
- Build practical workflows over flashy demos
- Document setup, limits, and privacy behavior clearly
Experiments around personal assistants that run on user-controlled hardware and talk to local LLMs.
- even-g2-local-agent — local-first voice agent for Even Realities G2 glasses using Even Terminal Mode, Android, and Ollama.
Voice-first interfaces and glanceable HUD workflows for lightweight, hands-free computing.
Agent-controlled workflows for research, coding, and daily operations.
GPU and quantization experiments for running stronger models without depending on cloud APIs.
Independent, hardware-tested playbooks and evidence from my DGX Spark fleet. This is community work, not NVIDIA documentation, certification, or support.
- Open the Gumbii DGX Spark Lab
- Read the multi-node vLLM validation package
- Read the GLM-5.2 operations kit
- Review the separation and provenance record
even-g2-local-agent— local-first G2 voice agent templatedgx-spark/— personal DGX Spark operations, validation, and measured resultsspark-nvfp4-lab— local GPU inference / NVFP4 experimentsbrowser-harness— browser automation researchvibevoice— open-source voice AI researchholaos— agent-computer UX referencescookbook— Gemma/local-model examples and notes
Individual projects may be experimental, reference material, or active builds. Product and company names are used descriptively; no affiliation or endorsement is implied.