Dhuel FisherProducts / interfaces / systems
Menu
All projects

Personal project / Independent project / March 2026–present

Eleanor AI

A portable continuity hub that keeps AI coding tools and me working from the same, current context across machines, sessions, and projects.

Daily use across work and personal machines; 690+ commits

690+
commits since the hub became a repository
4
AI tools routed through one shared context
2 OS
Windows and macOS with path translation

The problem

AI coding tools forget everything between sessions and know nothing about the other machine or tool I used yesterday. Without a durable shared memory, every session starts with re-explaining the project, and lessons learned the hard way get lost.

My role

Sole designer and developer.

  1. Any machine
    • Windows or macOS
    • USB hub copy
    • Machine registry
  2. Eleanor hub
    • Routing and read order
    • Current state and catch-up
    • Lessons learned
  3. AI tools
    • Claude Code
    • Codex
    • Kiro
    • ChatGPT
  4. Project repos
    • Canonical docs
    • Specs and tasks
    • Verified changes
How a session starts. Every tool reads the same routing and state before touching a project.

What I did

  1. One entry point for every tool

    Each AI tool (Claude Code, Codex, Kiro, ChatGPT) starts from the same routing file, which points it to the active project, its canonical docs, and the read order. Project facts stay in the project repos; the hub only summarizes them and points to them.

  2. Portable and machine-aware

    The hub travels on USB drives between Windows and macOS machines. A machine registry translates paths between them, and a sync policy merges several drive copies so that nothing is lost and the newest information wins.

  3. Durable handoffs and catch-up

    Session catch-ups, handoffs, and a current-state record let a new session, or a different tool, pick up exactly where the last one stopped.

  4. A living lessons-learned base

    Mistakes and fixes become written rules: commit standards, verification steps, and a cross-agent working standard for running Claude and Codex on the same codebase in parallel. Each new session follows rules that earlier sessions earned.

  5. Spec-driven by default

    Requirements, design, and task specs come before implementation, and each task carries its own acceptance criteria and evidence of verification.

Stack

  • Python
  • PowerShell
  • Bash
  • Markdown
  • JSON
  • Git
  • Claude Code
  • Codex
  • Kiro

The repository is private because it holds personal working context.