NeuroStack v1.1

A self-managing memory dashboard for Claude Code. Built on the memory system architecture described by sammii.

Reads and writes your Claude Code memory files directly from the browser using the File System Access API. No server, no database, no external dependencies — just markdown files and a fast local UI.

---

Features

View	What it does
File Editor	Read and edit any .md memory file with ⌘S to save
Agent Tracker	Kanban board parsed from active-work.md — working / blocked / done
Timeline	Chronological feed of decisions.md + worklog.md with quick-add forms
Search	Fuzzy full-text search across all loaded memory files (powered by Fuse.js)

---

Getting started

1. Install and run

cd neurostack
npm install
npm run dev

Open http://localhost:5173.

2. Set up your memory directory

Copy the templates to your Claude memory directory:

# Find your Claude project hash first (check ~/.claude/projects/)
MEMORY_DIR=~/.claude/projects/<your-project-hash>/memory

mkdir -p $MEMORY_DIR/projects
cp memory-templates/MEMORY.md       $MEMORY_DIR/
cp memory-templates/active-work.md  $MEMORY_DIR/
cp memory-templates/decisions.md    $MEMORY_DIR/
cp memory-templates/gotchas.md      $MEMORY_DIR/
cp memory-templates/worklog.md      $MEMORY_DIR/
cp memory-templates/stack.md        $MEMORY_DIR/
cp memory-templates/infra.md        $MEMORY_DIR/
cp memory-templates/projects/_template.md $MEMORY_DIR/projects/

3. Wire up each project

Copy claude-templates/CLAUDE.md into each project root and update the memory directory path at the top.

4. Open in the dashboard

Click "Open memory directory" and select your memory/ folder. The dashboard reads all .md files recursively and begins watching for changes every 4 seconds.

---

The memory system

memory/
  MEMORY.md              # Index / routing table (<100 lines)
  active-work.md         # Live multi-agent coordination
  decisions.md           # Architecture decisions with rationale
  gotchas.md             # Known bugs and workarounds
  worklog.md             # Rolling session history (last 30)
  stack.md               # Stack conventions and preferred patterns
  infra.md               # Ports, services, env var locations
  projects/
    usepopcorn-v2.md
    movie-pocket.md
    component-library.md
    antihero-magazine.md
    …

The protocol

Session start: Read MEMORY.md → active-work.md → your project file → gotchas.md if debugging

During work: Update active-work.md when you start/change/finish tasks. Update project files immediately after meaningful changes.

Session end: Remove your active-work.md entry. Append to worklog.md.

Key rule: write continuously, not at session end. If an agent crashes, its entry stays in active-work.md as a signal.

---

Tech stack

• Vite + React 18 + TypeScript
• File System Access API (Chrome/Edge/Brave — no Firefox)
• Fuse.js for fuzzy search
• date-fns for date formatting
• DM Mono + Syne (Google Fonts)
• Zero backend, zero dependencies beyond the above

---

Browser support

The File System Access API requires a Chromium-based browser:

• ✅ Chrome 86+
• ✅ Edge 86+
• ✅ Brave
• ❌ Firefox (not supported)
• ❌ Safari (partial, no write access)

---

Project structure

src/
  components/
    layout/      Sidebar, TopBar
    views/       FileEditor, AgentTracker, Timeline, Search, Landing
  hooks/         useMemoryFS, useFileWatcher, useSearch
  lib/           parseActiveWork, parseDecisions, parseWorklog
  types/         memory.ts
  styles/        global.css
memory-templates/   Copy these to your Claude memory directory
claude-templates/   CLAUDE.md — drop into each project root