Featured
Launch your own AI agent in one click
Sponsored placement
SetupClaw: done-for-you OpenClaw for founders & exec teams
Sponsored
One API to scrape, enrich, and extract the internet.
Sponsored placement
CLN.Work — Stop prompting, start hiring AI employees
Limited-time offer
Keep your OpenClaw agent online 24/7
Featured
Launch your AI product and start charging today
Free tool
AI Agent VPS Calculator
Size a VPS for your AI agent fleet in seconds. Pick your agents, model mode, and workload to get recommended vCPU, RAM, and disk, a provider price comparison, and a docker-compose starter.
Configure your fleet
Recommended specs
Sized for 3 OpenClaw agents running a medium workload in cloud API mode.
Provider comparison
Nearest matching plan per provider. Prices are approximate 2026 monthly estimates and vary by term and region.
| Provider | Plan | Specs | Est. $/mo | |
|---|---|---|---|---|
| Hostinger | KVM 4 | 4 vCPU / 16 GB / 200 GB | $14 | Deploy on Hostinger |
| Hetzner | CX32 | 4 vCPU / 8 GB / 80 GB | $8 | |
| DigitalOcean | Basic 8GB | 4 vCPU / 8 GB / 160 GB | $48 |
When no single plan fits, the row shows the largest plan and a scale out flag, meaning you should split the fleet across multiple servers or move to a dedicated instance.
docker-compose.yml
version: "3.9"
services:
openclaw:
image: ghcr.io/openclaw/openclaw:latest
restart: unless-stopped
environment:
- AGENT_CONCURRENCY=3
- MODEL_MODE=cloud
- ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
deploy:
resources:
limits:
cpus: "4"
memory: 7G
volumes:
- openclaw_data:/data
volumes:
openclaw_data:
Minimal starter file. Set your API key and volumes for production, and never commit real secrets.
How this VPS calculator works
The calculator starts from a small operating system baseline, then adds per-agent CPU, memory, and disk based on the workload you pick. A light agent needs roughly 1 vCPU and 2 GB of RAM, while medium and heavy workloads scale that up for busier automation, scraping, and build tasks. Heavy configurations also add 20 percent headroom so a spike does not trigger out-of-memory kills.
Model mode is the biggest lever. In cloud API mode the agents call a hosted model such as Claude, so the server only runs the agent processes. In local model mode the server also runs an inference engine like Ollama or llama.cpp, and even a quantized 7B to 13B model needs 12 to 24 GB of RAM. That is why switching to local mode raises the recommended specs sharply.
Once specs are calculated, the tool matches the smallest plan from each provider that meets your vCPU, RAM, and disk needs and shows an approximate 2026 monthly price. Prices are estimates and change with contract length, region, and promotions, so treat them as a starting point rather than a quote.
Frequently asked questions
How much RAM does an OpenClaw agent need?
A single OpenClaw agent doing light work in cloud API mode is comfortable on about 2 GB of RAM. The number climbs with more concurrent agents and heavier workloads, and it climbs a lot if you run a local model on the same box, since the model weights dominate memory use.
Do I need a bigger server for local models?
Yes. Cloud API mode offloads inference to a provider, so a small VPS is usually enough. Running a local model means hosting the inference engine yourself, which typically requires 12 to 24 GB of RAM and more disk for the model weights. If budget is tight, a cloud API is almost always the cheaper path to get started.
Which VPS provider is best for AI agents?
It depends on your priorities. Hostinger is beginner friendly with managed Docker hosting and low entry pricing. Hetzner offers strong price to performance for developers comfortable with a bare VPS. DigitalOcean has a polished platform and broad tooling. The comparison table shows the nearest plan and estimated price for each so you can weigh cost against convenience.
What is the docker-compose file for?
It is a minimal starting point you can drop onto your VPS to launch the selected agent with the recommended CPU and memory limits. Local model setups also include an Ollama service. Add your API keys, volumes, and any extra services before running it in production, and never commit real secrets to source control.
Are the prices accurate?
They are approximate 2026 estimates for the nearest matching plan. Actual pricing changes with billing term, region, taxes, and promotions, so always confirm the current rate on the provider site before you buy.