Running OpenClaw or Hermes Agent: hosting and monthly cost

Last checked Oct 23, 2026. Requirements come from the official OpenClaw and Hermes Agent docs. Prices come from the vendors’ own pricing pages, linked inline. AideDex did not run either agent and has no usage data of its own, so this guide gives you a formula and labelled examples, not a typical bill.

Both projects are free software. OpenClaw is MIT licensed, with “no enterprise edition and no paid version” (openclaw.ai). Hermes Agent is MIT licensed, and “model providers and optional hosted services have their own pricing” (Hermes Agent). Your monthly cost is the machine it runs on plus the model it calls. Plan prices for hosted agents are on our cost page.

What each one needs

OpenClaw

• Gateway host: the docs’ absolute minimum is 1 vCPU, 1 GB RAM and about 500 MB disk. They recommend 1–2 vCPU and 2 GB+ RAM, since browser automation and Node tools can be resource hungry. Ubuntu LTS is the best-tested OS (OpenClaw FAQ).
• Operating system: macOS or Linux, with Windows through WSL2 or Windows Hub. A Mac mini is popular but not required; you only need a Mac for macOS-only tools like iMessage (OpenClaw FAQ).
• Raspberry Pi: a Pi 4 with 4 GB is the docs’ “sweet spot”. A Pi 4 with 2 GB works with swap (Raspberry Pi docs).

Hermes Agent

• Desktop app: macOS 12+ and Windows 10/11; Linux uses the terminal install (Hermes Agent). The macOS installer is Apple Silicon only (Installation docs).
• Runtime: current installs run on Python 3.14, which the installer manages. Docker, Nix and an Android (Termux) package are also documented (Installation docs).
• The Hermes docs we checked don’t publish a minimum RAM or CPU figure for the agent itself when it uses a cloud model.

Where to run it

Option	OpenClaw	Hermes Agent
Your own computer	Supported; no server cost, but sleep and reboots disconnect it (OpenClaw FAQ)	Desktop app or terminal (Installation docs)
Your own VPS	Guides for Azure, DigitalOcean, GCP, Hetzner, Oracle, Railway, Render and more (Linux server); Hostinger OpenClaw on VPS, which you administer yourself (Hostinger docs)	Terminal install or Docker image on a server (Installation docs)
Managed hosting	Hostinger 1-Click (managed) (Hostinger docs)	Hermes Cloud from Nous Research (Hermes Cloud)

The OpenClaw docs’ rule of thumb: use a VPS for 24/7 reliability, and run it locally if you accept sleep and restarts (OpenClaw FAQ). Setup steps are in our OpenClaw and Hermes Agent guides.

Published server prices

• DigitalOcean: the OpenClaw docs use the 1 GB Basic Droplet. DigitalOcean lists it at $6/mo, and the 2 GiB, 1 vCPU Droplet at $12/mo (DigitalOcean). The docs say local model inference does not fit in 1 GB (DigitalOcean docs).
• Oracle Cloud Always Free: the OpenClaw docs describe running on Oracle’s free ARM tier (up to 4 OCPU, 24 GB RAM) “at no cost”, with the caveat that free capacity is limited and signup can fail with “Out of capacity” (Oracle docs).
• Hostinger managed OpenClaw: on Oct 23, 2026 the US page showed a time-limited Black Friday offer of $4.99/mo (regular price $21.99/mo), renewing at $11.99/mo for 2 years (Hostinger). Plans are paid upfront; the monthly rate is the total divided by the months in the plan (Hostinger). Plans include “pre-configured AI credits”, or you can add your own Anthropic, OpenAI, Gemini or xAI key (Hostinger docs). The page doesn’t say how many credits are included.
• Hermes Cloud: billed daily from your Nous credit. A running instance pays for compute (metered by the second) plus storage; a stopped one pays storage only. Inference and tool usage are billed on top. Instance prices are shown in Nous Portal, not on the public page. You need a $2 minimum credit or an active subscription to deploy (Hermes Cloud).

Hosting from India

• Hostinger’s India page lists managed OpenClaw from ₹429/mo as a sale price, renewing at ₹999/mo for 2 years, paid upfront (Hostinger India).
• Hermes Cloud lists Mumbai among its server locations. Nous picks the closest location that has availability, so you can’t be sure of getting Mumbai (Hermes Cloud).

More on agents in India: India availability.

What drives the monthly bill

1. Model API usage (usually the biggest line)

With an API key, you pay per token:

Monthly model cost = (input tokens per month ÷ 1,000,000 × input price) + (output tokens per month ÷ 1,000,000 × output price)

Prices are per million tokens (MTok). Two examples from Anthropic’s pricing page (Claude pricing):

Model	Input	Output
Claude Sonnet 5.5	$2 / MTok	$10 / MTok
Claude Haiku 5.5 (prompts up to 100,000 tokens)	$0.10 / MTok	$0.50 / MTok

Prompt caching lowers this: cache hits on Sonnet 5.5 cost 0.05x the base input price (Claude pricing). The examples below ignore caching, so real bills with caching would be lower.

2. Subscriptions instead of API keys

OpenClaw can also sign in with a ChatGPT/Codex subscription (OAuth) instead of a pay-per-token API key. Codex OAuth limits are plan-dependent quota windows that can differ from ChatGPT web, so subscription use is capped by those windows rather than billed per token (OpenClaw FAQ). Hermes users can use Nous Portal. It has a Free tier (“free models only”) and paid tiers of $20, $100 and $200 a month with $22, $110 and $220 of monthly credits (Hermes Agent).

3. Local models: no token bill, more hardware

• Hermes with Ollama: 8 GB RAM minimum for 3B models, 32+ GB recommended for 27B+ models, and 4+ CPU cores. A GPU is optional (Hermes Ollama guide). In the guide’s model table, only gemma4:31b (24+ GB RAM) supports tool calling (Hermes Ollama guide).
• OpenClaw: the smallest managed llama.cpp recipe has an 8 GiB host-memory floor, and the docs say these floors “do not guarantee fit or speed” (Local models). Asked whether a local model is OK for casual chats, the FAQ answers “Usually no”, and adds that smaller models raise prompt-injection risk (OpenClaw FAQ).

Electricity and hardware costs depend on your machine, so we don’t estimate them.

Worked examples

Assumptions (ours, not from any vendor): 30 agent requests a day for 30 days (900 requests). Each request averages 20,000 input tokens, since agents resend context and tool results, plus 1,000 output tokens. That is 18 MTok input and 0.9 MTok output a month. No caching, no taxes.

Example A: VPS + Claude Sonnet 5.5

• Model: 18 × $2 + 0.9 × $10 = $36 + $9 = $45
• Server: DigitalOcean 2 GiB Droplet = $12 (DigitalOcean)
• Total under these assumptions: $57/month

Example B: VPS + Claude Haiku 5.5

• Model: 18 × $0.10 + 0.9 × $0.50 = $1.80 + $0.45 = $2.25
• Server: $12
• Total under these assumptions: $14.25/month

Double your request count and the model line doubles. The server line stays the same.

Measure your own usage

Replace our assumptions with your own numbers. In OpenClaw, openclaw gateway usage-cost --days 7 summarizes usage and cost from session logs (Gateway query docs). In Hermes, /usage shows token usage and an estimated cost breakdown, and /insights shows the last 30 days (Slash commands).

Agent pages: OpenClaw, Hermes Agent.

FAQ

Is OpenClaw free?

The software is. It is MIT licensed with “no enterprise edition and no paid version” (openclaw.ai). You still pay for hosting and model usage unless you use your own computer and a local model.

Is Hermes Agent free?

Yes, it’s MIT licensed. Model providers and optional hosted services have their own pricing, and Nous Portal’s Free tier offers free models only (Hermes Agent).

What is the minimum VPS for OpenClaw?

1 vCPU, 1 GB RAM and about 500 MB disk. The docs recommend 2 GB+ RAM (OpenClaw FAQ).

How much does Hostinger’s OpenClaw hosting cost in India?

From ₹429/mo as a sale price, paid upfront, renewing at ₹999/mo for 2 years, per Hostinger’s India page on Oct 23, 2026 (Hostinger India).

How is Hermes Cloud billed?

Daily from your Nous credit: compute by the second while running, storage always, and inference and tools on top (Hermes Cloud).