NativeClaw
Open Source · MIT Licensed

Deploy a team of
always-on AI agents

An opinionated framework for running multiple Claude Code agents with structured roles, shared config, and Telegram integration. One command to install. One wizard to set up. Always running.

terminal

$ curl -fsSL https://nativeclaw.org/install.sh | bash

Claude Code CLI installed

tmux installed

bun installed

Launching setup wizard...

Welcome to NativeClaw! I'm the setup wizard.

Let's set up your agent fleet in just a few minutes.

Everything you need to run an agent fleet

NativeClaw is built native to Claude Code — no wrappers, no abstractions, just structured configuration and scripts.

🤖

Multi-Agent Fleet

Run multiple specialized agents on a single machine. Each agent has its own role, scope, and instructions.

💬

Telegram Integration

Every agent gets its own Telegram bot. Message your agents from anywhere with per-agent access control.

Setup Wizard

Go from zero to running agent in minutes. The wizard installs prerequisites and walks you through everything.

📃

Structured CLAUDE.md

Shared root config + per-agent instructions. Claude Code's directory traversal gives each agent the right context.

🧠

Persistent Memory

File-based memory system with typed entries — user preferences, feedback, project context, references.

📝

Task Tracking

Each agent maintains a markdown task list. Simple, transparent, and easy to review.

👥

Team Access

Control who can message which agent. Allowlists, group chat policies, and Telegram pairing — per agent.

💻

Run Anywhere

EC2, a VPS, your laptop — anywhere Claude Code runs. Agents persist in tmux sessions across reboots.

How it works

Each agent is a Claude Code session running in tmux, with its own identity and Telegram connection.

~/nativeclaw/
├── CLAUDE.md                   # Shared config ─── inherited by all agents
├── wizard.sh                   # One-command setup
├── create-agent.sh             # Scaffold new agents
├── start-agent.sh              # Launch agent in tmux
├── start-all.sh                # Launch the fleet
│
├── agents/
│   ├── wizard/                 # Setup wizard (runs once)
│   │   └── CLAUDE.md
│   │
│   ├── chief-of-staff/         # Your first agent
│   │   ├── CLAUDE.md           # Role: coordination, task mgmt
│   │   ├── tasks.md            # What it's working on
│   │   └── memory/             # Persistent context
│   │
│   ├── dev-agent/              # Code & deploy
│   │   ├── CLAUDE.md           # Role: engineering
│   │   ├── tasks.md
│   │   └── memory/
│   │
│   └── repos/                  # Shared repositories
│
└── docs/                       # Shared knowledge base
1

Shared Config

The root CLAUDE.md defines your identity, team, and shared resources. Every agent inherits it automatically.

2

Agent Roles

Each agent's own CLAUDE.md layers on role-specific instructions — scope, responsibilities, security boundaries.

3

Always On

Agents run in tmux sessions with Telegram connected. Message them anytime — they're always listening.

Get started in one command

The installer sets up everything — Claude Code, tmux, bun — then launches a wizard that creates your first agent.

terminal
$ curl -fsSL https://nativeclaw.org/install.sh | bash

Or clone manually: git clone https://github.com/VantaSoft/nativeclaw.git ~/nativeclaw && cd ~/nativeclaw && ./wizard.sh

1Installs prerequisites (Claude Code, tmux, bun)
2Sets up your profile — who you are, your team
3Creates your first agent — an AI chief of staff
4Connects it to Telegram via @BotFather
5Launches it in a background tmux session

Frequently asked questions