TL;DR — Clawdbot (aka OpenClaw) is an open-source, self-hosted AI assistant by Peter Steinberger that runs fully on your Linux box. No prompts, file paths, or code snippets get shipped to a third party. Install Node.js 22+ via nvm, then:
curl -fsSL https://clawd.bot/install.sh | bash
openclaw onboard --install-daemon
That's it. Below is the full walkthrough — trade-offs, troubleshooting, headless server setup, and how it compares to Ollama/LM Studio/Aider/etc.
Why local-first matters (the 30-second pitch)
Most AI tooling is a remote API call in a trench coat. Every command, file path, and code fragment you feed it becomes a log line somewhere you don't own. For anyone with a proprietary codebase, production access, or just healthy paranoia, that's a non-starter.
OpenClaw flips the architecture: the assistant runs on your machine, talks to your shell, reads your files, and optionally bridges into Discord/Telegram/WhatsApp/Slack. You can back it with a local LLM (Ollama, llama.cpp) for fully air-gapped operation, or route to OpenAI/Anthropic when quality > privacy for a given task. Your call, per query.
What you actually get
| Feature | Description |
|---|---|
| Local execution | All processing stays on the host |
| Shell execution | Runs commands on your behalf (with review) |
| Filesystem access | Reads/writes/indexes local files |
| Chat bridges | Discord, Telegram, WhatsApp, Slack |
| Plugin support | Custom scripts + community plugins |
| Cross-platform | Linux native, macOS, Windows via WSL2 |
| Daemon mode | systemd user service for always-on access |
System requirements
The critical dependency is Node.js 22+, which is newer than what most distro package managers ship. Don't fight apt/dnf — use nvm.
| Requirement | Minimum | Recommended |
|---|---|---|
| OS | Any modern Linux | Ubuntu 22.04 LTS / Debian 12 |
| Node.js | 22.0+ | Latest 22.x LTS |
| Package manager | npm 9+ / pnpm 8+ | pnpm |
| RAM | 2 GB free | 4+ GB |
| Disk | 500 MB | 2+ GB (plugins/cache) |
| Permissions | sudo for daemon install | — |
Confirmed working on: Ubuntu 20.04/22.04/24.04, Debian 11/12, Fedora 38–40, Arch (current), Linux Mint 21+, Pop!_OS 22.04, Raspberry Pi OS (64-bit). Anywhere else: you just need glibc 2.31+ and Node 22.
Step 1: Get Node.js 22 via nvm
# Check what you have
node -v
If it's not v22.x.x, install nvm:
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
# Reload your shell (pick one)
source ~/.bashrc
source ~/.zshrc
# Verify
nvm --version
Then install and pin Node 22:
nvm install 22
nvm use 22
nvm alias default 22
node -v # v22.x.x
npm -v # 9.x+
Don't skip the alias default. Without it, every new terminal session drops back to whatever Node the system had, and the daemon will misbehave.
Why strictly 22+?
OpenClaw leans on ES2023+ features and native fetch. On Node 18/20 the installer may appear to succeed, but the daemon fails to start or behaves unpredictably. Treat node -v showing 22+ as a hard precondition.
Step 2: Install OpenClaw
You have two paths. Pick one.
# Path A — automated installer (recommended)
curl -fsSL https://clawd.bot/install.sh | bash
# Path B — manual, via npm
npm install -g openclaw@latest
Then register the background service:
openclaw onboard --install-daemon
The onboarding wizard will walk you through LLM provider (local Ollama/llama.cpp or cloud), chat integrations, and permission scopes.
What you can actually do with it
These are the workflows that tend to stick once the daemon is running.
1. Natural-language sysadmin
Describe intent, let OpenClaw generate the command, review it, run it.
"Clean up log files in /var/log older than 30 days, keeping the last 5 GB"
"List all systemd services that have failed in the last hour"
"Check disk usage across all mounted volumes; alert if any > 85%"
Community benchmarks in the OpenClaw GitHub discussions report 30–60 minutes/day saved on routine maintenance once scopes are dialed in.
2. Coding assist with real repo access
Unlike Copilot or browser-based ChatGPT, it reads your actual tree. No copy-paste into a web UI, no source leaving the host.
"Summarize what the auth/ directory does"
"Generate a test for processPayment in src/payments.js"
"Find all TODO comments grouped by file"
"What changed in the last 10 commits?"
3. Searchable personal knowledge base
Point it at your notes, PDFs, and markdown. grep stops being your only option.
4. ChatOps from your phone
Wire it into Discord or Telegram, DM the bot, execute ops from anywhere. No VPN, no exposed SSH.
A 2024 State of DevOps report by DORA Research found ChatOps-enabled teams resolve incidents 30% faster on average.
5. Scheduled reports
Cron + OpenClaw + Telegram = daily disk reports, weekly git summaries, CPU alerts pushed to a channel instead of you SSH-ing in to check.
6. Dev environment scaffolding
"Set up a new Python venv for project X with these deps"
"Run the test suite and summarize failures"
"Start the local frontend dev server"
ChatOps: wiring up Discord and Telegram
Discord
- Create a Discord Application at discord.com/developers/applications
- Add a Bot, copy the token
- In onboarding (or
openclaw config), pick Discord and paste the token - Invite the bot to your server with message/command permissions
- Type in the designated channel — OpenClaw replies with output
Security reality check: anyone with write access to that channel can execute commands on your box. Treat it like SSH. Use a private channel with a tight membership list.
Telegram
# After creating a bot via @BotFather and copying the token:
openclaw config --chat telegram --token YOUR_TOKEN
Telegram is the fastest integration because BotFather handles registration inside the chat. Personal setup: under 5 minutes to first command.
Access control
openclaw config --allowlist # restrict to specific user IDs
Audit allowlists regularly on shared bots.
| Integration | Setup complexity | Best for |
|---|---|---|
| Discord | Medium (dev portal) | Team/DevOps channels |
| Telegram | Low (BotFather) | Personal, mobile control |
| High (Meta Business API) | Enterprise | |
| Slack | Medium (App directory) | Enterprise dev teams |
How it compares to other local AI tools
| Tool | Chat integration | Terminal exec | File indexing | Setup |
|---|---|---|---|---|
| OpenClaw (Clawdbot) | Discord, Telegram, WhatsApp | Full | Yes | Low–Medium |
| Open WebUI + Ollama | None (web UI) | No | Limited | Medium |
| LM Studio | None | No | No | Low |
| Jan.ai | None | No | Limited | Low |
| Aider | None | Yes (git-focused) | Yes (git repos) | Low |
| Continue.dev | None (IDE plugin) | Via IDE | Yes (project) | Low |
Trade-off summary: if you want terminal exec + chat bridges in one package, OpenClaw is the only game in town as of 2026. If you just want a conversational UI over a local model and nothing touches your shell, Open WebUI + Ollama or Jan.ai is a smaller attack surface.
Troubleshooting
Most install failures land in one of: PATH, Node version, daemon permissions, port conflicts, incomplete onboarding.
openclaw: command not found
npm's global bin isn't in PATH.
npm config get prefix
# e.g. /home/you/.local/share/nvm/versions/node/v22.x.x
echo 'export PATH="$(npm config get prefix)/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc
openclaw --version
node: version not supported
nvm install 22
nvm use 22
openclaw onboard --install-daemon
Permission denied during daemon install
The daemon registers as a systemd user service, not system-wide. You need lingering enabled:
loginctl enable-linger $USER
openclaw onboard --install-daemon
Daemon status: failed
journalctl --user -u openclaw -n 50
Usual suspects: port 3147 in use (lsof -i :3147), Node version mismatch, missing env vars from a half-finished onboarding.
Quick reference
| Error | Cause | Fix |
|---|---|---|
command not found |
PATH missing | Add npm bin to PATH |
version not supported |
Node < 22 | nvm install 22 |
EACCES: permission denied |
Global install perms | Use nvm (no sudo) |
Port 3147 in use |
Conflict | Kill process or reconfigure port |
Daemon failed to start |
Missing env vars | Re-run openclaw onboard
|
Running OpenClaw headless (VPS, home server, Pi)
The daemon architecture was built for this. Install identically via SSH; the only difference is how you interact afterward.
Three interaction paths:
# 1. SSH into the interactive CLI
ssh user@server openclaw
# 2. Localhost API
curl http://localhost:3147/api/query -d '{"query":"disk usage"}'
# 3. Chat apps (Discord/Telegram) — the primary recommended path
Hardening checklist
| Measure | Implementation |
|---|---|
| API localhost-only | Default; never expose 3147 externally |
| Chat allowlist |
openclaw config user ID allowlist |
| Read-only sensitive dirs | Filesystem perms + OpenClaw scope |
| Audit logs | openclaw config --audit-log |
| Firewall | Block 3147 externally at the firewall |
The non-negotiable one: port 3147 must stay on localhost. It's an unauthenticated command executor — internet-exposing it is equivalent to handing out root shells.
Updating and a few notes on naming
# Via npm
npm install -g openclaw@latest
# Or built-in
openclaw update
# Restart the daemon (npm path requires this manually)
systemctl --user restart openclaw
openclaw --version
"Clawdbot" vs "OpenClaw": same software. The project originally shipped as Clawdbot and was rebranded to OpenClaw as it outgrew its chatbot origins. npm package: openclaw. Binary: openclaw. Site: clawd.bot. Older tutorials using "Clawdbot" refer to the same thing you install with npm install -g openclaw@latest.
Multi-user: per-user installation, user-space daemon. Each user gets an isolated instance, config, scopes, and chat integrations. No cross-contamination.
Pointing it at a local LLM: pick "Local LLM" in onboarding, choose Ollama or llama.cpp. Ollama is the path of least resistance — ollama pull llama3 and OpenClaw auto-detects the running server.
Summary
- Local-first AI assistant with terminal + filesystem + chat app integration in one package
- Node.js 22+ via nvm is the only gotcha
-
curl -fsSL https://clawd.bot/install.sh | bash→openclaw onboard --install-daemon - Headless works great; keep 3147 on localhost, use chat apps as the front door
- When things break: check
node -v, check PATH, checkloginctl enable-linger, checkjournalctl --user -u openclaw
Reference: https://videodubber.ai/blogs/how-to-install-clawdbot-linux/.
Top comments (0)