How much does it cost to self-host OpenClaw?

On AIC Cloud, the server itself costs ₹99/month for a 1 GB VPS (sufficient for OpenClaw in API mode). On top of that you'll spend roughly ₹200-500/month on Claude or OpenAI API tokens for typical personal use. Total: ~₹300-600/month, or ₹4,000-7,000/year — significantly cheaper than ChatGPT Plus or Claude Pro at ₹20,000+/year per seat.

Do I need a GPU to run OpenClaw?

Not if you use Claude, OpenAI, or Gemini APIs as the LLM brain — the heavy compute happens at the API provider's datacenter, your VPS just orchestrates. A ₹99/mo 1 GB CPU-only VPS is enough. You only need a GPU if you want to run a local LLM (Llama, Mistral, Qwen) for full privacy and no API costs.

What's the difference between running OpenClaw locally on my laptop vs on a cloud VPS?

Your laptop closes its lid, sleeps, runs out of battery, or gets restarted. OpenClaw can't act on incoming messages or run scheduled tasks during those times. A cloud VPS is always on, has a stable IP, doesn't sleep, and can run automations 24/7. It also means OpenClaw is reachable from anywhere — your phone, work computer, anywhere — through whichever messaging app you connected.

Can I run OpenClaw on a free tier (Google Cloud / AWS / Oracle)?

Yes technically, but they all have caveats: GCP free tier limits CPU to specific zones, AWS free tier expires after 12 months, Oracle's always-free tier requires international credit card and has aggressive eviction policies. A ₹99/mo AIC Cloud VPS is more predictable: month-to-month INR billing via UPI, no expiry, no surprise eviction, and the same price forever.

Which messaging app works best with OpenClaw?

Telegram is the easiest to set up (free bot creation via @BotFather, instant tokens). WhatsApp is the most familiar interface for Indian users but requires a Business Account setup. Discord is great if you want to share your assistant with a team or community. All work — pick whichever you already use most.

Is OpenClaw safe to give shell access to?

It depends on your trust model and your LLM. A well-prompted modern LLM (Claude Sonnet 4.5, GPT-4o) is generally cautious about destructive shell commands, but no LLM is perfectly safe. Best practice: run OpenClaw in a Docker container (so it can only affect its own filesystem), avoid mounting your home directory directly, and don't put production secrets on the same VPS. For sensitive automations, sandbox aggressively or restrict shell access in OpenClaw's config.

Will OpenClaw work behind India's ISP firewalls (Jio Fiber, Airtel Xstream)?

Yes for outbound API calls (Claude/OpenAI work fine). The only thing that might be blocked is inbound SSH on non-standard ports — if your ISP blocks port 20020, you can either change AIC Cloud's SSH port, use the dashboard's web console, or set up Cloudflare Tunnel for SSH access (see our Cloudflare Tunnel guide).

Can I run multiple OpenClaw instances on one VPS?

Yes, especially with Docker — run multiple containers on different ports, each with its own config and API key. The 1 GB VPS handles 2-3 lightweight instances comfortably; for 5+ instances or heavy browser-automation workloads, the 2 GB plan (₹199/mo) is more appropriate.

How to Self-Host OpenClaw on a Cloud VPS (₹99/month, Step-by-Step Guide 2026)

What is OpenClaw and why self-host it?

OpenClaw is an open-source personal AI assistant that runs on your own hardware and automates tasks via natural conversation. It connects to messaging apps (WhatsApp, Telegram, Discord, Slack), browses the web, manages files, controls applications, and integrates with 50+ services (Gmail, GitHub, Spotify, Obsidian, and more).

Why self-host instead of using a SaaS AI assistant?

•Privacy. Your data stays on your server. No third party sees your inbox, files, or conversations.
•Persistent memory. OpenClaw remembers your preferences and context across sessions.
•Full control. Run it 24/7, give it shell access, build custom skills.
•Cost predictability. A small cloud VPS at ₹99/month plus your Claude or OpenAI API tokens, vs $20/month per SaaS assistant that locks you in.

The catch: OpenClaw needs to run somewhere always-on. Your laptop isn't ideal (closes, sleeps, gets stolen). A cloud VPS is — and you don't need anything expensive.

---

What kind of server do you actually need?

OpenClaw in API mode (using Claude / OpenAI / Gemini as its brain) needs surprisingly little hardware. The heavy thinking happens at the LLM provider's datacenter, not on your VPS.

Resource	Minimum	Recommended
RAM	1 GB	2 GB
CPU	1 vCPU	2 vCPU
Storage	10 GB	20 GB
Network	100 Mbps	300 Mbps+
OS	Ubuntu 22.04 / Debian 12	Ubuntu 24.04 LTS
GPU	Not needed	Not needed

This fits comfortably on an AIC Cloud Essential 1 GB VPS at ₹99/month for hobby use, or the Essential 2 GB at ₹199/month if you want headroom for multiple OpenClaw instances or heavy browser automation.

(If you want to run a *local* LLM brain instead of using an API, you'll need a GPU. See our GPU-mode OpenClaw guide for that setup.)

---

Step 1: Provision your VPS

Sign up at aiccloud.in, top up your wallet via UPI (₹100 is enough for the first month), and deploy an Essential 1 GB VPS with Ubuntu 22.04 or 24.04. The provisioning takes about 60 seconds.

You'll receive your server's IP address and root password via email. SSH in:

ssh root@YOUR_VPS_IP

If your local network blocks the default SSH port, AIC Cloud assigns a non-standard port — check your dashboard for the exact command (it'll look like ssh -p 20020 root@...).

---

Step 2: Install OpenClaw

Update the system first:

apt update && apt upgrade -y
apt install -y git curl wget python3 python3-pip nodejs npm

Clone the OpenClaw repository (check the official OpenClaw site for the latest install command — they may publish an install script you can run directly):

git clone https://github.com/openclaw/openclaw.git
cd openclaw
./install.sh

If they ship a Docker image (most modern AI agent tools do), prefer that:

curl -fsSL https://get.docker.com | sh
docker pull openclaw/openclaw:latest
docker run -d --name openclaw --restart unless-stopped \
  -v ~/.openclaw:/root/.openclaw \
  -p 8080:8080 \
  openclaw/openclaw:latest

Docker is the cleanest path — auto-restart, easy upgrades, clean uninstall.

---

Step 3: Connect a brain — Claude or OpenAI API

OpenClaw needs an LLM to think with. Two paths:

Path A — Claude API (recommended for cost + quality)

1. Sign up at console.anthropic.com and create an API key

2. Add credit to your Anthropic account ($5 minimum, lasts a long time for personal use)

3. In your OpenClaw config (~/.openclaw/config.yaml or similar):

llm:
  provider: anthropic
  model: claude-sonnet-4-5
  api_key: sk-ant-api03-YOUR_KEY_HERE

Path B — OpenAI API

1. Sign up at platform.openai.com and create an API key

2. Add billing credit ($5 minimum)

3. Config:

llm:
  provider: openai
  model: gpt-4o
  api_key: sk-YOUR_KEY_HERE

For most personal-assistant workloads, Claude Sonnet 4.5 gives the best quality-to-cost ratio. GPT-4o is comparable. Cheaper models (Haiku, GPT-4o mini) work for simple tasks if you want to save 80% on token costs.

Restart OpenClaw to pick up the new config:

docker restart openclaw

---

Step 4: Connect a messaging app (your interface)

OpenClaw's killer feature is that you chat with it through apps you already use. Pick one:

•Telegram — easiest setup, create a bot via @BotFather, paste the token into config
•WhatsApp — requires a WhatsApp Business account or a meta-business setup
•Discord — create a bot in Discord's developer portal, invite to your server
•Signal — requires linking via signal-cli (slightly more setup)

Check OpenClaw's docs for the exact connector setup per app — they're all in the integrations/ folder of the repo.

After connecting, send your bot a test message: *"What can you do?"* — it should reply with a list of available skills.

---

Step 5: Lock it down

A few security hardening steps for any always-on server:

# Update SSH to a non-default port (already done by AIC Cloud)
# Disable password login, use key-based auth only
nano /etc/ssh/sshd_config
# Set: PasswordAuthentication no
systemctl restart sshd

# Install fail2ban
apt install -y fail2ban
systemctl enable --now fail2ban

# Firewall — only allow SSH + your OpenClaw API port if needed
ufw allow 20020/tcp   # your SSH port
ufw allow 8080/tcp    # OpenClaw, only if you need external access
ufw --force enable

If you only access OpenClaw via Telegram/Discord (not its web UI), don't expose port 8080 publicly at all.

---

Step 6: Set up auto-restart on reboot

Make sure OpenClaw comes back after server reboots (planned or otherwise):

# Docker handles this with --restart unless-stopped (already set above)
# Confirm:
docker update --restart unless-stopped openclaw

If you installed without Docker, create a systemd service. The OpenClaw docs typically include a template.

---

Cost breakdown for a year

Item	Cost
AIC Cloud Essential 1 GB VPS	₹99/month × 12 = ₹1,188/year
Claude API (light personal use, ~$3/month)	~₹250/month × 12 = ₹3,000/year
Domain (optional, if you want a custom URL)	₹500/year
Total	~₹4,700/year

Compare that to ChatGPT Plus ($20/month = ₹20,160/year) or Claude Pro ($20/month = ₹20,160/year) — and you get persistent memory + custom skills + full data ownership as a bonus.

---

Troubleshooting

OpenClaw won't start — check Docker logs: docker logs openclaw. Common issues: missing API key, malformed config YAML, port already in use.

LLM responses are slow — switch to a smaller/faster model (Claude Haiku, GPT-4o mini) for routine tasks. Reserve Sonnet/4o for complex reasoning.

Out of memory — Upgrade to the 2 GB plan (₹199/mo) if you're running browser automation. Chromium headless eats RAM.

Locked out of SSH — Use AIC Cloud's web console (VPS dashboard → "Console") to log in and fix sshd config.

---

Next steps

Once OpenClaw is running, the fun starts:

•Build your own skills (Python plugins that extend what the agent can do)
•Connect more integrations (Gmail, GitHub, Notion, Obsidian — all 50+)
•Set up scheduled tasks ("every morning, summarize my unread emails")
•Multi-agent setups across multiple VPSes for advanced use cases

If you want to run a local LLM (no API costs, full privacy) instead of using Claude/OpenAI, you'll need a GPU server — see our companion guide on running OpenClaw with a local Llama model on a cloud GPU.

---

Deploy an AIC Cloud VPS for OpenClaw from ₹99/month →

Tags:OpenClawAI AgentsVPSSelf-HostingTutorial

READY TO GET STARTED?

Deploy your first VPS for ₹99/mo

India-based servers, INR billing, no lock-in contracts. Get started in minutes.

View VPS Plans →