Active / Experimenting

Using local AI models, coding assistants, workflow automation, and project memory for practical technical work.

A local AI workflow lab focused on LM Studio, Qwen model experiments, OpenCode, Codex/GPT workflows, Git-backed notes, and small reviewable task blocks for coding, documentation, repair notes, and homelab work.

Why Local AI

Local models let me experiment with private workflows, structured prompts, repair documentation, and automation ideas without relying entirely on cloud tools. Cloud models still help where they are stronger; the goal is to use the right worker for the task.

Tools I Use

  • LM Studio
  • Qwen models
  • OpenCode
  • VS Code
  • Continue
  • Codex / GPT workflows
  • Git worktrees
  • Make
  • Cloudflare Tunnel
  • Telegram bots
  • Local prompt files
  • Obsidian / Git memory vault

Main Projects

Automation Workbench

Hermes AI Agent

Project memory, infrastructure notes, repair documentation, coding workflows, and small-block updates backed by Git.

Hermes LabOps

Self-hosted dashboard and future safe control surface for documenting services, roles, status, and operational notes.

Repair Shop Workflow Support

Customer-facing notes, repair summaries, quote/message drafts, and internal process documentation for human review.

Local Model Routing Experiments

Testing how local and cloud model choices affect speed, quality, privacy, and workflow fit.

Git Worktree Agent Flow

Keeping the main branch stable while using small, reviewable task blocks for parallel coding/review work.

Documentation Generation Workflows

Drafting and organizing notes without replacing evidence-based troubleshooting.

Current Focus

  • Improve long-context project workflows.
  • Use Git-backed memory/documentation.
  • Keep main branches stable with small task blocks.
  • Use local models as coding workers/reviewers.
  • Generate safer technical documentation.
  • Support coding and homelab troubleshooting.

Lessons Learned

  • Local AI works best with structured context.
  • Prompt design matters more when workflows are messy.
  • Automation needs validation and guardrails.
  • Sensitive data must be sanitized before public examples.
  • AI should improve documentation and review, not replace technical judgment.