AI Agent Workflow Orchestration Platform for Claude Code
Create agents, organize them into teams, plan tasks with AI, and monitor execution in real-time. All powered by Claude Code.
Named after Claude Shannon, the father of information theory.
Note: This is a hobby project. Built for personal use, shared in case others find it useful.
Real-time metrics, agent performance, task analytics, and activity feeds — all at a glance.
- Stat Cards — Project, Agent, Team, and Session counts at a glance
- Live Running Tasks — See what's actively executing right now
- Charts — Task status distribution, session overview, success rate gauge, code review stats
- Completion Trend — Track task completion velocity over time
- Agent Leaderboard — Performance metrics and model distribution
- Recent Activity — Latest sessions and active tasks feed
Drag and drop agents onto the canvas, connect them to define execution flow. Choose from parallel, sequential, or fully custom strategies.
- Drag & Drop Canvas — Powered by ReactFlow for smooth interactions
- 3 Strategy Presets — Parallel (all at once), Sequential (chained), or Custom (draw your own DAG)
- Multi-Model Teams — Combine Opus, Sonnet, and Haiku agents in a single workflow
- Visual Edge Editing — Click to connect, click to remove
A purpose-built Monaco code editor for crafting Claude system prompts. Semantic syntax highlighting, intelligent autocomplete, and one-click AI improvement.
- Semantic Tag Coloring — 6 categories with distinct colors: Role (emerald), Structure (cyan), Context (sky), Tool (amber), Thinking (violet), Artifact (pink)
- Smart Autocomplete — XML tag snippets,
{{variable}}suggestions, auto-closing tag detection with stack-based matching - Hover Documentation — Hover over any tag name for category and description
- AI Improve — One-click prompt enhancement via Claude with explanation, Revert/Keep workflow
- Full Monarch Tokenizer — Markdown headings, bold/italic, lists, code blocks,
IMPORTANT:keywords, template variables — all properly highlighted - Custom Theme —
claude-zinc-darkwith bracket pair colorization and emerald cursor
Watch your task dependency graph execute in real-time. Chat with agents, review code changes, and manage workflow — all in one split view.
- DAG View — Interactive task dependency graph with live status updates (completed, running, pending)
- Live Chat — Stream agent output as it works, see tool usage in real-time
- Changes Panel — Review code diffs with file-level +/-/~ indicators
- Real-Time Events — Wails event streaming with 15s polling fallback
- Session Controls — Start, Stop, Complete from the workspace header
Describe your goal in natural language. The AI analyzes your workspace structure and proposes an optimal task breakdown with dependencies.
- Workspace-Aware — Analyzes your codebase structure before planning
- Auto-Dependencies — Automatically resolves task ordering
- Smart Assignment — Maps tasks to the most suitable agents or teams
- Accept & Iterate — Review proposed tasks, re-plan if needed, then accept
| Feature | Description |
|---|---|
| Custom Agents | Create agents with specific models (Opus, Sonnet, Haiku), system prompts, tool permissions, and MCP server connections |
| Team Workflows | Organize agents into teams with visual DAG editor — parallel, sequential, or custom execution |
| MCP Servers | Connect to Model Context Protocol servers with JSON editor, health checks, and import from Claude |
| Sessions & Tasks | Task queues with dependency graphs, retry policies, and automatic test execution |
| Project Workspaces | Link repos, configure build/test commands, and manage CLAUDE.md context files |
| Encrypted Vault | AES-256 encrypted storage for API keys and environment variables — injected at runtime, never exposed |
| Code Review | Hunk-level and file-level diff review with feedback loop and inline Monaco editing |
| Prompt Improver | AI-powered system prompt enhancement with explanation and revert capability |
1. Add Workspace → Link your project, configure build/test commands
2. Create Agents → Specialized agents with models, prompts, and tools
3. Build Teams → Organize agents with visual DAG workflows
4. Launch Session → Plan tasks with AI, execute, monitor in real-time
| Layer | Tech |
|---|---|
| Desktop shell | Wails v2 (Go + WebView) |
| Backend | Go 1.23, SQLite (GORM) |
| Frontend | React 18, TypeScript, Tailwind CSS |
| Code editor | Monaco Editor (custom claude-prompt language) |
| Graph viz | React Flow (XYFlow) + Dagre |
| State | Zustand (granular selectors) |
| Virtualization | TanStack Virtual (1000+ log entries) |
| AI engine | Claude Code CLI |
- Claude Code CLI installed and authenticated
- Wails v2 development environment
- Go 1.23+
- Node.js 18+
# Clone
git clone https://github.com/yourusername/shannon.git
cd shannon
# Install frontend dependencies
cd frontend && npm install && cd ..
# Run in development mode
wails dev
# Build for production
wails buildshannon/
├── backend/
│ ├── claude/ # Claude Code CLI integration
│ ├── config/ # App configuration & encrypted vault
│ ├── models/ # Data models (Agent, Team, Session, Task...)
│ ├── services/ # Business logic (TaskEngine, Planner, DiffTracker, PromptImprover...)
│ └── store/ # SQLite database operations
├── frontend/
│ ├── src/
│ │ ├── components/
│ │ │ ├── agents/ # Agent editor, ClaudePromptEditor, steps
│ │ │ ├── chat/ # Chat components
│ │ │ ├── dashboard/# Stat cards, charts, leaderboard
│ │ │ ├── layout/ # AppShell, Sidebar
│ │ │ ├── mcp/ # MCP server management
│ │ │ ├── session/ # DAGView, ChatPanel, ChangesPanel
│ │ │ └── team/ # TeamAgentNode (ReactFlow)
│ │ ├── pages/ # Dashboard, Agents, Teams, Sessions, Workspace, Monitor...
│ │ ├── stores/ # Zustand state (session, agent, team, project)
│ │ └── types/ # TypeScript type definitions
│ └── index.html
├── build/ # App icons and build assets
├── app.go # Wails bindings (all exposed methods)
└── main.go # Entry point
- Requires Claude Code CLI — this is not a standalone AI tool
- Workspace copies can use significant disk space for large projects
- No remote/cloud mode — it's a local desktop app
- Hobby project — expect rough edges
MIT