How to Install and Secure Nanoclaw on a Self-Managed VPS and VDS via SSH

This guide installs NanoClaw manually on a blank Self-Managed VPS or VDS over SSH. NanoClaw is an AI agent that connects to messaging platforms (Telegram, Discord, WhatsApp, Slack, and others) and runs each conversation's agent inside its own disposable Docker container — nothing an agent does touches the host filesystem unless you explicitly allow it.

System Requirements

These specs apply whether you're running on Self-Managed VPS or VDS.

Prerequisites

1. A Self-Managed VPS or VDS meeting the specs above, freshly provisioned with no Application pre-selected.
2. Root SSH access: the server's IP address and the initial root password (or key) provided at provisioning.
3. A credential for your AI provider: by default this is Claude — a Pro/Max subscription, an OAuth token, or an Anthropic API key. You don't need this in hand before you start; the setup wizard asks for it partway through and you can also skip it and connect later.

Connect to Your VPS or VDS via SSH

Example Output:

1. Baseline Linux Host Hardening

Do this before installing NanoClaw, not after — a fresh VPS or VDS is exposed to automated scanning the moment it's on the internet.

Create an Unprivileged Deployment User

The prompts after this (Full Name, Room Number, Phone, etc.) are just optional Linux account metadata — press Enter through all of them.

Install a Text Editor (If Not Present)

Set Up Key-Based Login — Do This Before Disabling Passwords

If you don't already have SSH key-based access set up for this server, generate a key pair on your own computer and copy it over:

Already have a working key for this server? Skip straight to confirming it below.

In a new, separate terminal window — keeping your current root session open — confirm it works without a password prompt:

You should land at a prompt with no password asked for SSH, and sudo whoami should print root (it will ask for nanoclawadmin's own password — that's normal and separate from SSH login). Only continue once this works.

Disable Password and Root SSH Login

Ubuntu's default file already has these three lines, commented out, in the "Authentication:" section — find and uncomment each, setting the values below:

Immediately test a fresh login in yet another new terminal before closing anything:

Configure the Firewall

ufw isn't installed by default on a minimal Ubuntu image:

Allow SSH before enabling the firewall, or you'll lock yourself out the same way as the password change above:

Example Output:

Deploy Fail2Ban

From here on, do everything as your unprivileged user:

2. Install NanoClaw

NanoClaw's installer checks for Node.js, pnpm, and Docker, and installs whatever's missing on its own — you don't need to install Docker yourself first. Make sure git is available, then run the installer:

This launches an interactive setup wizard — it takes roughly 15 minutes on a fresh machine (most of that is the first container image build). It will ask you, in order: Standard or Advanced setup, whether to reuse or install a OneCLI credential vault, how you want to connect your AI provider (subscription sign-in, OAuth token, API key, or skip), a display name for your assistant, your timezone, and whether to connect a messaging channel now or later.

3. Verify the Install

Confirm the background service is actually running:

Confirm the socket the CLI and chat script talk to actually exists:

Then test it directly from the terminal:

If you skipped provider sign-in during setup, this will fail with something like 401 No credentials configured... in OneCLI — that's expected, not a bug; finish sign-in with bash nanoclaw.sh or inside a claude session to resolve it.

4. Agent Groups, Not "Profiles"

Each isolated agent identity in NanoClaw is called a group (found under groups/<folder>/ in your checkout) — not a "profile." The underlying advice still holds: keep unrelated jobs in separate groups rather than combining them, so a compromised or misbehaving agent in one group can't reach another group's files, memory, or mounts. Groups are created and managed with the ncl CLI, covered in a separate article on managing your install day to day.

5. How You Actually Interact With It

Three ways to work with your install once it's running:

• Chat directly from the terminal: pnpm run chat hi
• Open a full Claude Code session (for setup skills, debugging, adding channels): claude
• Query or modify state directly (groups, channels, approvals) with the ncl CLI: pnpm ncl help
• Talk to it for real once you've connected a channel — Telegram, Discord, WhatsApp, Slack, and others install on demand via /add-<channel> skills inside a claude session.

Summary

The security fundamentals for a NanoClaw Self-Managed VPS or VDS are the same as for any server: key-based SSH only, a firewall that only allows what you actually need open, and Fail2Ban watching for brute-force attempts. NanoClaw's own security model — container isolation per agent, mount allowlists, and an approval workflow for sensitive actions — is handled by the application itself once it's running; it doesn't require a manually maintained Docker Compose stack or an exposed web dashboard.