Clawdbot: Build Your Own Private AI Assistant on a Cheap VPS | Brav

Clawdbot: Build Your Own Private AI Assistant on a Cheap VPS

Table of Contents

TL;DR

  • I installed Clawdbot on a Hostinger VPS in 15 minutes. Hostinger — VPS Hosting Plans (2026)
  • I connected it to Telegram and got instant email summaries.
  • The whole stack is self-hosted, secure, and runs on $6.99/month.
  • Cron jobs, webhooks, and secret storage are just a few commands away.
  • No cloud provider or paid AI subscription is needed (aside from the LLM API key).

Why this matters

Every day I juggle emails, flight searches, and quick coding tweaks. Existing AI tools are either cloud-based or lack true privacy. Clawdbot lets me keep all data on my own server and automate the chores I used to do manually. The biggest pain point was the setup process, but the onboarding wizard and the single-command install made it painless. Clawdbot — The AI that actually does things (2026)

Core concepts

Clawdbot is built around a gateway process that talks to messaging apps, an agent that runs the LLM, and a skill system that gives the agent extra tools. It works with any LLM provider that offers an API or OAuth: Anthropic, OpenAI, Google Gemini, or Antigravity. The gateway stays on your machine, and every message you send is processed locally before the model is called. Clawdbot — GitHub (2026)

ProviderStrengthsLimitation
AnthropicStrong prompt-injection resistance, good for codeRequires paid tier
OpenAIFamiliar API, cheap GPT-4Rate limits apply
Google GeminiMulti-modal supportStill in beta
AntigravityExperimental, freeLimited community

I learned that the LLM choice rarely changes the core workflow; it just shifts cost and latency.

How to apply it

Below is my step-by-step recipe for a fresh Hostinger VPS (KVM 2, 2 vCPU, 8 GB RAM, $6.99 / mo). All commands run as a non-root user with key-based SSH.

# 1. Spin up a Hostinger VPS
# 2. SSH as a normal user (root login disabled by default)
ssh user@host

# 3. Install Node 22 (LTS) and other build tools
sudo apt update && sudo apt install -y curl gnupg
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt install -y nodejs
# Node 22 LTS release: https://nodejs.org/en/blog/release/v22.11.0

# 4. Install Clawdbot globally
npm install -g clawdbot@latest
# Clawdbot installation guide: https://docs.clawd.bot/install

# 5. Run the onboarding wizard
clawdbot onboard --install-daemon
#   • Select “local” gateway
#   • OAuth flow for your LLM (Anthropic for me)
#   • Add your Telegram bot token

# 6. Verify everything is running
clawbot status

Tip: The wizard will create a ~/.clawdbot/ directory where credentials and session data live. Keep this folder under version control except for secrets. Clawdbot — Agent Workspace (2026)

Adding a skill – the GOG email summarizer

# Install the skill registry CLI
clawdhub install gog
# Clawdhub installation guide: https://docs.clawd.bot/tools/clawdhub

# Tell Clawdbot to use the GOG CLI
clawbot skills enable gog

The GOG skill gives the assistant direct access to Gmail, so I can ask it “Summarize my inbox” and get a 5-sentence digest in a Telegram message. It pulls the OAuth token from the gogcli.sh setup, so I never expose my Gmail password. Clawdbot — Gmail Pub/Sub (2026) GOG — Google Workspace CLI (2026)

Scheduling a daily flight price check

Clawdbot’s built-in cron service lives under ~/.clawdbot/cron/. Add a job file:

# ~/.clawdbot/cron/flight.yml
schedule:
  kind: "every"
  interval: "12h"
payload:
  kind: "agentTurn"
  command: "search flights from NYC to LAX"

The gateway will wake the agent at the defined interval and deliver the result back to the chat. The job is persisted across restarts. Clawdbot — Cron Jobs (2026)

Hardening security

The onboarding wizard also set up key-based SSH only and disabled password authentication on the VPS. Clawdbot’s security audit warns about open group policies, so run:

clawbot security audit
clawbot security audit --fix

It will lock down DM policies, tighten file permissions (~/.clawdbot/ → 700), and ensure no tools are exposed to strangers. Clawdbot — Security Audit (2026)

Pitfalls & edge cases

  • Root login: If you accidentally re-enable root SSH, forget to remove the public key and the bot can be compromised.
  • Token rotation: OAuth tokens for Anthropic or OpenAI expire; the wizard only sets them up once. Use the CLI clawbot config set to update them or run the wizard again.
  • Scaling: Running multiple Gateways on the same host requires unique ports and separate workspaces (~/.clawdbot/workspaces/). The default setup is for a single user.
  • Skill compatibility: Some community skills depend on binaries (e.g., Docker, Homebrew). They may fail on a bare-bones VPS unless you install the prerequisites.

Quick FAQ

Q: How can Clawdbot securely handle OAuth tokens in production?
A: The wizard stores tokens in ~/.clawdbot/credentials/ with 600 permissions. For extra safety, move the file to a password-protected keyring or use environment variables.

Q: What are best practices for adding other messaging platforms?
A: Search Clawdhub for the platform skill (e.g., “slack”), install it, and configure the token in the same way you did for Telegram.

Q: How does Clawdbot perform on a low-resource VPS?
A: On a 2 vCPU, 8 GB RAM instance it handles a handful of concurrent chats and cron jobs. Heavy LLM calls may saturate the CPU, but the node process stays lightweight.

Q: Does Clawdbot auto-renew OAuth tokens?
A: No built-in auto-renew. You must refresh them manually or script the wizard.

Q: Can I run Clawdbot on multiple machines?
A: Yes. Spin up a gateway per host, each with its own workspace, and use Clawdhub to sync skills.

Q: What are the limitations of Clawdhub skills?
A: Some skills lack thorough testing, depend on external services, or miss versioning. Always review the SKILL.md before enabling a new skill.

Conclusion

If you’re a home-lab enthusiast or sysadmin who hates repetitive tasks, Clawdbot is a low-friction way to get a private AI assistant. Install it, connect your messaging apps, add a few skills, and let the bot do the boring work. The whole stack runs on a $6.99 / month VPS, uses your own LLM key, and stays under your control.

Last updated: January 26, 2026

Recommended Articles

I Built Kai: A Personal AI Infrastructure That Turned My 9-5 Into a Personal Supercomputer | Brav

I Built Kai: A Personal AI Infrastructure That Turned My 9-5 Into a Personal Supercomputer

Discover how I built Kai, a personal AI infrastructure that turns scattered tools into a single context-aware assistant. Build websites, dashboards, and more in minutes.
AI Consulting as My Secret Weapon: How I Built a $250K Solo Empire and You Can Do It Too | Brav

AI Consulting as My Secret Weapon: How I Built a $250K Solo Empire and You Can Do It Too

Learn how I built a $250K solo AI consulting business, productized my expertise, and scaled founder-led brands—step-by-step tips for mid-career pros.
I Built a Raspberry Pi Wi-Fi Packet Capture Device with PoE, Multi-Adapter Support, and a Web UI | Brav

I Built a Raspberry Pi Wi-Fi Packet Capture Device with PoE, Multi-Adapter Support, and a Web UI

Learn how to turn a Raspberry Pi into a headless Wi-Fi packet capture device with PoE, multi-adapter support, a Flask web UI, and VPN access. Step-by-step guide.
The Vending Bench: Training AI to Run Real Businesses | Brav

The Vending Bench: Training AI to Run Real Businesses

Learn how the Vending Bench and Project Vend expose AI's strengths and pitfalls in running real businesses, with actionable guidance for researchers and leaders.
Cutting Latency in C++: Building a Low-Latency Trading System From Scratch | Brav

Cutting Latency in C++: Building a Low-Latency Trading System From Scratch

Learn how to design a low-latency C++ trading system from scratch, with real-world data structures, network stacks, and profiling tips that shave microseconds.
All-Optical Computer: I Built the First One—Why It Matters | Brav

All-Optical Computer: I Built the First One—Why It Matters

Discover how our all-optical computer delivers terahertz speeds, 10× lower power, and plug-in GPU-style performance for AI and data-center leaders.