StackediTED

A Collaborative AI Workspace
A remix of StackEdit transformed into a multi-agent workspace for creative collaboration between humans and AI.
Developed by Fladry Creative / robb@fladrycreative.com

---

What is StackediTED?

StackediTED reimagines document editing as a collaborative conversation between you and multiple AI agents. Instead of a sidebar chat, the AI workspace sits alongside your editor in a 50/50 split layout, creating a true collaborative environment.

Core Features

🤖 6 AI Providers Working Together

• Claude (Anthropic) - Deep reasoning and analysis
• Gemini (Google) - Multimodal understanding
• GPT (OpenAI via Codex) - Code generation and optimization
• Z.AI - Chinese AI company (GLM models)
• Grok (X.AI via Cursor) - Real-time knowledge
• Composer (Cursor) - Agentic composition

💬 @Mention System Direct messages to specific AI agents using mentions:

@claude analyze this code for security issues
@gpt optimize this function
@gemini explain this diagram
@composer write a REST API
@grok what's happening in AI news?

🔗 Agent-to-Agent Chaining (Max 3 Hops) AI agents can collaborate autonomously by mentioning each other:

Human: @claude write a sort algorithm
Claude: Here's quicksort... @gpt can you optimize?
GPT: Using timsort... @gemini verify complexity?
Gemini: O(n log n) is correct... @human approve?

COMMANDA Workflow

The Project Manager ("Ted") acts as the central orchestrator who sees everything—the document, the code, and the conversation.

• Role: Ted is the main persona you talk to. He maintains the "Meta-Reality" of the project.
• ** capabilities**:
  • Full visibility into the codebase and active documents.
  • Tags in other agents (@claude, @gemini, @grok) when specialized skills are needed.
  • Ensures context (the "drift") is controlled and aligned with the project vision.
• Interaction: You dump your thoughts/docs to Ted. Ted manages the rest, delegating tasks to specific agents while keeping the big picture in focus.

---

🔒 Trust Mode

• Trust OFF (default): Conversation pauses when AI mentions @human, awaiting your approval
• Trust ON: AI agents continue autonomously, only notifying you with @human mentions

📝 Document-Aware AI

• Your entire vault is synced to the AI daemon
• Agents can reference other documents, search your vault, and suggest edits
• Real-time document context for all conversations

---

Architecture

Frontend (Vue 2)

• Editor: Markdown editor with live preview
• AI Chat Panel: 50/50 split layout with resizable divider
• Vuex State: Manages chat history, provider state, and chaining logic

Backend (AI Daemon)

• WebSocket Server: Real-time bidirectional communication
• Provider Adapters: CLI wrappers for all providers (Claude, Gemini, GPT, Z.AI, Cursor)
• Custom CLI Wrapper: Built zai-cli for Z.AI API integration
• Smart Routing: Auto-selects best provider or routes via @mentions
• Vault Service: Caches and indexes your documents for AI context

Provider Colors

Provider	Color	Hex
Claude	Purple	#9D4EDD
Gemini	Teal	#2DD4BF
GPT	Coral	#FF6B6B
Z.AI	Amber	#F59E0B
Grok	Lime	#84CC16
Composer	Indigo	#6366F1

---

Installation & Setup

1. Install Dependencies

# Install frontend dependencies
npm install

# Install AI daemon dependencies
cd ai-daemon
npm install
cd ..

2. Configure AI Providers

Copy the example environment file:

cp ai-daemon/.env.example ai-daemon/.env

Edit ai-daemon/.env with your credentials:

# Claude (Anthropic CLI)
CLAUDE_CLI_PATH=claude
CLAUDE_MODEL=claude-opus-4-5-20251101

# Gemini (Google CLI)
GEMINI_CLI_PATH=gemini
GEMINI_MODEL=gemini-3-pro-preview

# OpenAI (Codex CLI)
CODEX_CLI_PATH=codex
CODEX_MODEL=gpt-5-codex

# Z.AI (API Key)
ZAI_API_KEY=your_zai_api_key_here
ZAI_MODEL=glm-4.6

# Cursor (Subscription - runs both Grok and Composer)
CURSOR_API_KEY=your_cursor_api_key_here
CURSOR_MODEL=grok
COMPOSER_MODEL=composer-1

3. Install CLI Tools

You'll need the CLI tools for the providers you want to use:

# Claude
brew install anthropics/claude/claude

# Gemini
npm install -g @google-ai/generativelanguage

# Codex (OpenAI)
npm install -g codex

# Cursor (for Grok and Composer)
# Install from https://cursor.sh

# Z.AI uses custom CLI wrapper (included)
# Just set ZAI_API_KEY in .env

4. Run the Application

Terminal 1 - Start AI Daemon:

cd ai-daemon
npm start
# Or in dev mode with auto-reload:
npm run dev

Terminal 2 - Start Frontend:

npm start

Open http://localhost:8080 in your browser.

---

Usage Examples

Basic Mentioning

@claude review this markdown document
→ Routes to Claude

@gpt refactor this code to be more functional
→ Routes to GPT

Agent Chaining

Human: @claude design a user authentication API
Claude: Here's the design... @gpt implement it?
[Auto-routes to GPT]
GPT: Here's the code... @gemini review security?
[Auto-routes to Gemini]
Gemini: Security looks good... @human deploy?
[Pauses if Trust Mode is OFF]

Trust Mode

Toggle the TRUST button in the AI panel:

• 🔒 Locked = Pauses on @human mentions
• 🔓 Unlocked = Continues autonomously

---

Development

Project Structure

stackedited/
├── src/                      # Frontend Vue application
│   ├── components/           # UI components
│   │   ├── AiChat.vue       # AI chat panel
│   │   ├── AiProviderSelector.vue  # Provider buttons
│   │   └── Layout.vue       # 50/50 split layout
│   ├── services/
│   │   └── aiService.js     # WebSocket client
│   └── store/
│       └── aiChat.js        # Vuex state management
│
├── ai-daemon/                # AI daemon server
│   ├── src/
│   │   ├── server.js        # WebSocket server + routing
│   │   ├── providers/       # Provider adapters
│   │   │   ├── claude.js
│   │   │   ├── gemini.js
│   │   │   ├── openai.js
│   │   │   ├── zai.js
│   │   │   ├── cursor.js
│   │   │   └── composer.js
│   │   └── services/
│   │       └── vaultService.js  # Document indexing
│   └── index.js

Build for Production

# Build frontend
npm run build

# The AI daemon runs as-is with:
cd ai-daemon
npm start

Running Tests

# Frontend tests
npm test

# Linting
npm run lint

---

Roadmap & Timeline

(Status: Cruise Week - Identifying the drift, no pressure)

Immediate Priorities (Cruise Mode)

• Establish COMMANDA Workflow: Ted as the central PM.
• Refine Z.AI Integration: Ensure GLM provider is fully operational.
• Documentation: Update guides to reflect the "StackediTED" meta-reality.

Future Goals

• Function Calling: Let AI agents edit files directly
• Vision Support: Upload images for Gemini/GPT-4V analysis
• Custom Providers: Plugin system for new AI providers
• Team Collaboration: Multi-user editing with shared AI context
• Streaming Improvements: Better UX for long responses
• Mobile Support: Responsive layout for tablets/phones

---

Credits

Built on top of StackEdit by Benoit Schweblin.

Remixed with 💜 by Fladry Creative

• Email: robb@fladrycreative.com
• Design Philosophy: "Controlled Drift" — Machine/human interface at the edge of perception

---

License

Apache 2.0 (same as original StackEdit)

---

Contributing

This is a creative experiment. If you want to contribute or have ideas, reach out!

Key areas for contribution:

• New AI provider integrations
• UI/UX improvements
• Performance optimizations
• Documentation

---

FAQ

Q: Why so many AI providers?
A: Different AIs have different strengths. Claude excels at reasoning, GPT at code, Gemini at multimodal tasks. Let them collaborate!

Q: What's the "controlled drift" philosophy?
A: AI agents can autonomously collaborate (drift), but with safety limits (controlled). Max 3 hops, Trust Mode, and @human pauses keep things from going off the rails.

Q: Does this cost a lot to run?
A: Depends on your usage. Most providers are CLI-based and use your existing API keys. Cursor runs on a subscription, not per-call billing.

Q: Can I use this offline?
A: The editor works offline, but AI features require internet and running providers.

---

Built for creators who want AI as a true collaborator, not just a tool. 🚀