VPS Sizing Guide: How Much RAM Do You Really Need?

The question every AI team builder asks first

Before deploying your first AI agent, you face a fundamental infrastructure decision: how big should your server be? Too small and your agents crash under load. Too large and you waste money on unused capacity every month. This guide gives you the exact numbers based on real-world deployments across hundreds of k-claw installations.

The baseline: ~1.5GB RAM per agent

Each AI agent deployed through k-claw runs as an independent process with its own gateway. Through extensive testing across different workloads, the consistent figure is approximately 1.5GB of RAM per active agent. This includes:

• The gateway process — The Node.js runtime that handles message routing, API calls to AI models, and connection management.
• Memory for context windows — As conversations grow, the agent holds recent context in memory for fast access.
• File and skill overhead — Agents with uploaded context files or installed skills use slightly more memory for file parsing.
• Operating system baseline — Ubuntu or Debian needs roughly 300-500MB for the OS itself, plus the k-claw connector daemon.

This 1.5GB figure is a practical average. Simple agents with short conversations might use 1GB. Agents processing large documents or running multiple skills simultaneously can spike to 2GB.

VPS sizing by team size

Based on the 1.5GB-per-agent baseline plus OS overhead, here is the recommended sizing:

VPS RAM	Max Agents	Best For	Monthly Cost Range
4 GB	1-2	Solo entrepreneur, single assistant + backup	EUR 4-8
8 GB	3-4	Small team: COO + CTO + 1-2 specialists	EUR 10-20
16 GB	7-8	Full team template: executive + department leads	EUR 25-50
32 GB	15+	Multi-department operation or agency setup	EUR 50-100

These ranges assume you are running the agents and nothing else significant on the server. If you also host websites, databases, or other services on the same VPS, reduce the agent count accordingly.

The 4GB starter: your first agent

A 4GB VPS is the minimum viable server for running AI agents. After the operating system takes its share (roughly 400MB), you have about 3.5GB of usable RAM. That comfortably fits one agent with room for occasional spikes, or two agents under light usage.

This is the right choice if you are:

• Testing the platform before committing to a larger deployment
• Running a single personal assistant for daily tasks
• Deploying the Freelancer template (2 agents: assistant + specialist)

Providers like Hetzner, Contabo, and DigitalOcean offer 4GB VPS plans starting at EUR 4-5 per month. At this price point, you can experiment freely without financial risk.

The 8GB sweet spot: a real team

For most users, 8GB is the sweet spot. It fits 3-4 agents comfortably, which is enough for the Startup template (COO, CTO, CMO) or a custom team with a coordinator and two to three specialists.

At this tier, you start seeing the real value of inter-agent communication. Three agents that can delegate to each other are dramatically more capable than three isolated chatbots. The COO can hand off technical questions to the CTO, marketing briefs to the CMO, and only escalate to you when human judgment is needed.

An 8GB VPS typically costs EUR 10-20 per month depending on the provider and region. Combined with AI API costs (which depend on usage), your total monthly spend for a 3-agent team is often under EUR 50.

The 16GB full team: AI Company in a Box

If you want the complete k-claw experience, 16GB lets you run the full Complete Team template with 7 agents: COO, CTO, CMO, frontend engineer, backend engineer, content writer, and customer support. This is the "AI Company in a Box" configuration.

With 7 agents at ~1.5GB each (10.5GB total) plus OS overhead, a 16GB server stays comfortably within limits even during peak activity. You have enough headroom for all agents to process concurrent requests without memory pressure.

This tier is ideal for:

• Small businesses replacing multiple SaaS subscriptions with AI agents
• Agencies managing client work across multiple domains
• Technical teams with specialized engineering agents

When to add a second server

Sometimes scaling up (bigger VPS) is not the right answer. Here are situations where scaling out (more servers) makes more sense:

• Department isolation — Your executive team and engineering team should be on separate servers for organizational clarity. k-claw's fleet management lets you group servers by department.
• Geographic distribution — If your agents serve users in different regions, placing servers closer to end users reduces latency.
• Risk mitigation — If one server goes down, agents on other servers keep running. A 2-server setup with 4 agents each is more resilient than a single server with 8 agents.
• Cost optimization — Two 8GB servers often cost less than one 32GB server, and you get better total performance due to independent CPU allocation.

k-claw handles multi-server deployments natively. Cross-server agent communication works through the platform relay, so agents on different servers collaborate just as seamlessly as agents on the same machine.

CPU considerations

RAM gets the most attention, but CPU matters too. AI agents are not CPU-intensive most of the time — the heavy computation happens at the AI model provider (OpenAI, Anthropic, etc.). Your VPS mainly handles message routing, file I/O, and network requests.

That said, avoid single-core VPS plans. Two cores is the minimum recommendation. Here is a quick reference:

Agents	Minimum Cores	Recommended Cores
1-2	1	2
3-4	2	2-4
5-8	2	4
9+	4	4-8

Cost comparison across providers

Here is what you can expect to pay monthly for the most common VPS sizes at popular providers (prices as of early 2026):

Provider	4GB	8GB	16GB
Hetzner (EU)	~EUR 4	~EUR 9	~EUR 17
Contabo (EU)	~EUR 6	~EUR 10	~EUR 16
DigitalOcean	~EUR 22	~EUR 44	~EUR 88
Vultr	~EUR 22	~EUR 44	~EUR 88
Linode	~EUR 22	~EUR 44	~EUR 88

European providers like Hetzner and Contabo offer dramatically better value for AI agent workloads. A full 7-agent team on Hetzner costs under EUR 20 per month for the server — less than a single seat on most enterprise AI platforms.

Monitoring memory usage

After deployment, k-claw's fleet dashboard shows real-time memory usage across all your servers. You can see per-agent consumption, total server load, and receive alerts when memory usage exceeds 80%. This makes scaling decisions data-driven rather than guesswork.

If you notice agents consistently using less than 1GB each, you might be able to fit one more agent on your current server. If any agent regularly exceeds 2GB, consider upgrading your plan or moving that agent to a less crowded server.

The bottom line

Start with a 4GB VPS if you are experimenting or need just one agent. Move to 8GB when you want a real team of 3-4 agents. Go to 16GB for the full AI Company in a Box experience. And when your operation grows beyond a single server, k-claw's fleet management makes multi-server deployments straightforward.

The total cost of running an AI agent team — server plus API costs — is typically EUR 30-80 per month. Compare that to hiring even a part-time human assistant, and the ROI becomes obvious.