Personal AI Workspace System

A self-documenting workspace built to govern how AI agents operate across a multi-project development environment. The system covers routing, approval levels, zone management, and operational safety — designed to run reliably without constant human supervision.

What it does

The workspace runs multiple active projects simultaneously, each with its own agent context, state, and risk profile. The system provides:

  • Agent pipeline routing — tasks are classified and dispatched to the right model (Claude, Gemini, Ollama) based on complexity, cost, and context requirements
  • Approval governance — six escalation levels (L0–L5) control what agents are allowed to do autonomously versus what requires human sign-off
  • Project zone management — projects move through zones (active, planning, scratch) with structural change protocols that prevent stale references and undocumented state drift
  • Post-structural change protocol — any zone move or rename triggers a mandatory scan of all workspace docs, a review table, and explicit approval before writes

Ops Runbook

The ops runbook documents the full operating model in a single reference: session protocols, approval phrases, agent routing rules, safety review checklist, inbox and handoff procedures, git sync policy, and workspace decision records (ADRs).

It is designed to be readable by a non-engineer — Asif is the product owner and final decision-maker, not the engineering operator.

Holistic System

The holistic system view maps the full architecture: workspace zones, I/O paths, agent roles, approval checkpoints, and integration points between local tools (Claude Code, Ollama, shell scripts) and cloud services (Claude API, Gemini, GitHub, Cloudflare).

It answers the question: how does work actually move through this system, and where does a human need to be in the loop?

Design decisions

  • Flat HTML — no build step, no framework. Dashboards are self-contained files that open directly from disk or a local server.
  • IBM Plex editorial type system — Sans for body, Serif for headings, Mono for code and approval phrases.
  • Offline-first — fonts bundled locally, no CDN dependency.
  • Six-state badge system — ACTIVE, PAUSED, PLANNING, SCRATCH, BLOCKED, DONE — maps directly to project and approval state.

Stack

HTML · CSS · IBM Plex · Cloudflare Pages