JARVIS

Multi-Agent Mission Control for AI Coding Agents

A single control panel to orchestrate Claude Code sessions running across multiple machines in your network.
Monitor agents in real-time, dispatch tasks, manage git workflows, and coordinate multi-machine development — all from one place.

Why JARVIS?

Managing AI coding agents across multiple machines is painful. You SSH into one box, check a session, switch to another, lose context, forget what's running where. JARVIS solves this by giving you one dashboard to see everything.

• No cloud required — runs entirely on your local network via SSH/Tailscale
• Real-time visibility — live activity feed parsed from Claude Code JSONL sessions
• Multi-machine — manage local and remote machines from a single window
• Native desktop app — built with Tauri 2, not a browser tab

✨ Features

🚀 Quick Start

Prerequisites

• Rust 1.77+
• Node.js 20+
• Tauri CLI
• Claude Code installed on target machines

Install & Run

git clone https://github.com/emawritz/share-sharvis.git
cd share-sharvis
npm install
cd src-tauri && cargo tauri dev

On first launch, JARVIS creates a default config at ~/.config/jarvis/config.toml.

[session]
id = ""
rama = ""
objetivo = ""

[[machines]]
id = "main"
name = "MAIN"
host = "local"          # "local" = this machine
os = "macos"
role = "orchestrator"
enabled = true
tags = ["local"]

[[machines.repos]]
name = "my-project"
path = "~/projects/my-project"
github = "user/my-project"

[[machines]]
id = "worker"
name = "WORKER"
host = "worker-ssh"     # SSH alias from ~/.ssh/config
os = "linux"
enabled = true
gpu = "RTX 3070"

[[machines.repos]]
name = "my-frontend"
path = "~/projects/my-frontend"
github = "user/my-frontend"

🏗️ Architecture

src-tauri/src/          Rust backend (Tauri commands)
├── lib.rs              App setup, plugin registration, command routing
├── config.rs           TOML config management
├── machines.rs         Machine registry, health checks, system stats
├── session.rs          Session monitor, JSONL activity parsing
├── tasks.rs            Task dispatch (local/SSH), execution engine
├── planning.rs         Ping-pong planning mode, branch management
├── pipelines.rs        Multi-step pipeline orchestration
├── github.rs           GitHub CLI integration (PRs, checks)
├── visibility.rs       Timeline analytics from JSONL files
└── types.rs            Shared Rust types

src/lib/                Svelte 5 frontend
├── components/         UI components (Header, CommandBar, tabs/)
├── stores/             Reactive stores (session, tasks, machines, planning)
├── api.ts              Tauri invoke wrappers
└── types.ts            TypeScript interfaces

How It Works

flowchart TB
    subgraph SETUP["🚀 First Launch"]
        W1[Setup Wizard] --> W2[Detect Local Machine]
        W2 --> W3[Add Remote Machines]
        W3 --> W4[Test SSH Connections]
        W4 --> W5[Save config.toml]
    end

    subgraph CORE["⚙️ Runtime"]
        CFG[~/.config/jarvis/config.toml] -->|load| REG[Machine Registry]
        REG -->|spawn| HC[Health Checks]
        REG -->|spawn| SM[Session Monitor]
        SM -->|parse| JSONL[Claude JSONL Files]
    end

    subgraph MACHINES["🖥️ Machines"]
        LOCAL[Local Machine\nshell exec]
        REMOTE1[Remote Machine 1\nSSH + Tailscale]
        REMOTE2[Remote Machine 2\nSSH + Tailscale]
    end

    subgraph UI["🎨 Dashboard · Svelte 5"]
        TABS[Tabs: Tasks · Machines · Commits · GitHub · Pipelines · Timeline]
        CMD[Command Bar]
        AGENT[Agent Activity Feed]
    end

    W5 -->|generates| CFG
    HC -->|status| UI
    SM -->|activity stream| AGENT
    CMD -->|dispatch task| REG
    REG -->|host = local| LOCAL
    REG -->|host = ssh-alias| REMOTE1
    REG -->|host = ssh-alias| REMOTE2
    LOCAL -->|stdout/stderr| UI
    REMOTE1 -->|stdout/stderr| UI
    REMOTE2 -->|stdout/stderr| UI

Flow summary:

1. Setup Wizard auto-detects your local machine and lets you add remotes via SSH
2. Config is saved to ~/.config/jarvis/config.toml — fully dynamic, never hardcoded
3. Machine Registry loads config, spawns health checks and session monitors
4. Session Monitor parses Claude Code JSONL files for real-time agent activity
5. Task Dispatch routes commands to local shell or remote machines via SSH/Tailscale
6. Dashboard surfaces everything in a unified Svelte 5 UI with live updates

🛠️ Tech Stack

• Backend: Rust, Tauri 2, Tokio, Axum, rusqlite
• Frontend: Svelte 5, TypeScript, Vite
• Integrations: GitHub CLI, SSH, Tailscale

🏗️ Build

# Production build
cd src-tauri && cargo tauri build

# Type checking
cargo check                              # Rust
npx svelte-check --threshold error       # Svelte/TypeScript

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

📄 License

MIT — Emanuel Cejas