Managed AI-agent hosting: your agent, your key, our server
Always-on hosting for personal AI agents (OpenClaw and successors) on a hardened, isolated VPS. Bring your own API key, pay in USDT, BTC or XMR.
Yes, you can pay for AI agent hosting with crypto. MartCC runs your personal AI agent (OpenClaw or a compatible framework) on a hardened, isolated VPS from $9.99 a month, you bring your own model API key, and checkout takes Bitcoin (BTC), Tether (USDT, TRC20 by default), Ethereum (ETH), Litecoin (LTC), USD Coin (USDC), Solana (SOL), Monero (XMR), Dogecoin (DOGE), Algorand (ALGO) and Bitcoin Cash (BCH). An email address is the only thing we ask for; there is no card, no ID and no KYC.
The short version of what you get: an always-on server that is already firewalled, already patched, with the agent's gateway port behind authentication instead of open to the internet, and the agent installed with one click and paired to your Telegram or WhatsApp. The model tokens stay on your own provider account, under your own terms and your own spending limit. We host the agent. We never resell model access.
Plans and prices
All three plans run the same hardened image. They differ in size, in how many agents you can run, and in support.
| Plan | What is included | Price | Billing | Per month |
|---|---|---|---|---|
| Agent Starter | 2 vCPU, 4 GB RAM, 60 GB NVMe, 1 always-on agent, 2 chat channels (Telegram, WhatsApp, Discord or Slack), BYOK, weekly backups, hardened image with automatic updates, email support | $9.99 | monthly | $9.99 |
| Agent Starter, quarterly | Same, 5% off | $28.49 | quarterly | $9.50 |
| Agent Pro | 4 vCPU, 8 GB RAM, 100 GB NVMe, up to 3 agents, browser automation, daily backups with 7-day retention, priority support | $19.99 | monthly | $19.99 |
| Agent Pro, quarterly | Same, 5% off | $56.99 | quarterly | $19.00 |
| Agent Business | 8 vCPU, 16 GB RAM, 200 GB NVMe, unlimited agents, dedicated IPv4, daily backups with 14-day retention, 24/7 priority support, free migration | $39.99 | monthly | $39.99 |
| Agent Business, quarterly | Same, 5% off | $113.99 | quarterly | $38.00 |
| Managed setup or migration (add-on) | A technician installs the agent, loads your API key, pairs your channels and moves an existing instance if you have one | $19 | one-time | n/a |
Starter suits one agent that talks to you on Telegram, reads your mail or watches a few feeds, with no browser automation. Pro is the plan for anyone who wants the agent to open web pages: headless browsing is the thing that pushes memory past 4 GB, and 8 GB is the widely cited comfortable size for it. Business is for teams or power users running several agents side by side, or anyone who needs a dedicated IPv4 address for webhooks and a defined backup window. Every plan clears the $9 minimum payment on every coin, so none of them is wallet-only.
Where the money goes
The chart makes the point we repeat on every page in this section: hosting is the predictable part. The token bill goes to your model provider, not to us, and you control it with the spending limit on your own key.
How to buy AI agent hosting with crypto
- Pick Starter, Pro or Business, monthly or quarterly, and add the $19 setup service if you want us to do the key and channel setup for you. Enter an email address so we can send the login and the backup reports. A promo code can be entered here.
- Choose a coin and network. USDT defaults to TRC20 because the network fee is usually a few cents; ERC20 is there if your funds already sit on Ethereum. The price is shown in the coin at the live rate and locked for 30 minutes, and the network fee is paid by you on top.
- Scan the QR code or copy the address and the exact amount into your wallet. Send once. If the payment is short, what arrived is credited to your MartCC wallet and a "pay the difference" button appears; if it is over, the surplus is credited within minutes. If the 30 minutes expire, one click re-quotes.
- On confirmation the server is provisioned. You get the panel link, a one-time login and the pairing instructions on screen, by email and in the client area. The agent is typically reachable from your chat app within about ten minutes of the last step. Order lookup works without login by order ID.
Monero buyers: XMR is accepted on every plan, and the only identifier we hold is the email address you give us. Bitcoin on-chain is fine for monthly plans; for the $9.99 Starter the network fee can be a noticeable fraction of the price on a busy day, so many customers use USDT TRC20 or LTC for the small tickets and BTC for quarterly. See the fee notes on the pay with crypto page.
How it works technically
Each customer gets a separate virtual machine, not a container on a shared host. The image is the same on every plan, and it is built around the mistakes that got tens of thousands of self-hosted agents compromised in early 2026.
- Firewall on by default. Only SSH (key-only, on a non-standard port) and the authenticated tunnel endpoint are reachable from outside. The agent's gateway listens on loopback, which matches the OpenClaw defaults for a host install (the project's security docs, read 3 Oct 2026, say the gateway binds to loopback on
127.0.0.1:18789, while its container images default to an exposed bind that must be paired with auth). - Authentication on the gateway port. The Control UI is reached through a per-customer authenticated tunnel. The direct path from the public internet to port 18789 is blocked, and
openclaw security auditruns at provisioning so you can see the result in your welcome email. - Isolation per customer. Your VM, your disk, your keys. Support access to the instance is logged and shown to you.
- Automatic updates. Operating system security updates apply unattended. Agent updates are staged: we test a release on our own instances first, then roll it out with a snapshot taken before the upgrade so a bad skill can be rolled back.
- Vetted skills allow-list. Marketplace skills are disabled until you enable them, and the ones enabled by default are from a short list we have read. Around 20% of the skills on the public marketplace were found to be malicious in an early-2026 sweep, and a later compilation reported prompt injections in 36% of listed skills (secondary data, see Sources). You can enable anything you want; you just do it knowingly.
- Outbound port 25 blocked. Agents cannot send raw SMTP from the box. If your agent needs to send mail, it uses an authenticated relay on port 587, such as our SMTP service, which keeps the server's IP reputation clean for everyone.
Installing the agent is one click in the panel: choose OpenClaw (or another compatible Node-based agent), paste your model API key, and pair Telegram or WhatsApp by scanning a QR code from your phone. The OpenClaw hosting page walks through the exact screens, and the bring-your-own-key guide covers OpenAI, Anthropic and Gemini keys, plus crypto-friendly API providers.
If you run a different framework, a Hermes-style assistant, or a custom Python or Node agent, it will usually run on the same image. Open a ticket before ordering with the repo link and we will confirm, or take the $19 setup add-on and let us install it.
Who it is for, and who it is not for
It is for people who want a personal AI agent that is on all the time, reachable from a phone, and not sitting on a laptop that sleeps. It is for developers who could harden a VPS themselves but would rather not spend the evening, and for anyone who wants to pay for the hosting without a card, a bank or an identity check. It is for crypto-native users who already pay for the model tokens through a crypto-accepting API provider and want the rest of the stack to match.
It is not for running local models. These plans have no GPU; the agent talks to a model API over the network, which is how OpenClaw is designed to work. If you want to run open-weight models on your own hardware, a Mac mini or a GPU box is the right tool, and our cost breakdown is honest about when that wins.
It is not for anonymous outbound email, spam, credential stuffing, scraping that ignores robots or rate limits, or any workload that would get the IP range blacklisted. The acceptable-use policy is short and applies to every plan: no abuse of third-party services, no illegal content, no attacks, no mail on port 25. We also do not resell model access, so if your plan was to buy "AI credits" from us, you are on the wrong page: read how to pay for AI APIs with crypto instead.
One honest limit on privacy: a managed host provisions the agent for you, so the API key you paste lives in the agent's config on our machine. It is stored in the agent's own secrets store inside your VM, support access is logged, and you can rotate the key at your provider at any time. If a key on anyone else's hardware is unacceptable to you, a bare VPS or RDP server from the same checkout gives you a plain machine and nothing else.
How this compares with the alternatives
Three facts-only comparisons, because they are the three questions every buyer asks.
Against a bare VPS. A budget 4 GB VPS lists at around $4 a month (prices as listed in a hosting roundup dated 14 Aug 2026). The difference to our $9.99 buys the firewall, the tunnel, the audit, staged updates and backups, plus someone to ask when a skill breaks. Both are BYOK. The VPS route needs a card or a crypto account elsewhere; ours needs an email. Refunds here are MartCC credit within one working day; most VPS providers' crypto top-ups are final.
Against other managed agent hosts. Published managed prices in 2026 run from about $2.99 to $99.99 a month across the providers we have tracked, with the common 4 GB to 8 GB tiers around $9.99 to $29.99 (see the comparison page for the dated table). None of the roundup hosts listed by the largest tutorial site accepted crypto when we checked (14 Aug 2026). Most serious managed hosts, like us, expect you to bring your own key.
Against a Mac mini at home. $899 up front for the current base model (Apple Newsroom, 22 Sep 2026) against $9.99 a month is a seven-and-a-half-year payback on hosting cost alone, and the Mac has no public IP and no redundant power. It wins only if you already own one or intend to run local models.
Refunds and support
Agent hosting is refundable as MartCC credit if you cancel within 7 days of first provisioning and the instance has not breached the acceptable-use policy; credit is issued within one working day and never expires. The $19 setup add-on is refundable until work starts. After the first 7 days, cancellation stops the next renewal; there is no pro-rata refund of the current period. Refunds for any reason are paid as store credit, not back to the chain. Model tokens are between you and your provider and are never refundable by us because we never billed them.
Support is by ticket on all plans, with priority queueing on Pro and 24/7 priority on Business. Live chat and tickets work in ten languages (English, Spanish, Chinese, Arabic, Portuguese, French, Russian, Hindi, Turkish and German). Support can see your instance's health and logs; it cannot see your conversations with the agent unless you share them.
FAQ
Can I pay for AI agent hosting with crypto? Yes. Every plan on this page is paid through our own self-hosted checkout in BTC, USDT (TRC20 or ERC20), ETH, LTC, USDC, SOL, XMR, DOGE, ALGO or BCH. There is no card option and no KYC; an email address is all we ask for.
Is there a minimum payment? $9 on every coin. All agent-hosting plans are above that, so you can pay for any of them directly. Smaller items are paid from your MartCC wallet, which you top up with $9 or more and which never expires.
Do I need my own API key? Yes, on every plan. We never resell model access, so the key, the account and the spending limit stay yours. OpenAI, Anthropic and Gemini keys all work, as do crypto-accepting API providers; the BYOK guide has the details.
Which agents can I run? OpenClaw is the one-click option and the one we test every release of. Compatible Node-based agents install the same way, and custom agents usually run fine on the image. Ask before ordering if you are unsure, or add the $19 setup service.
How is this "no KYC" if you need my email? We need somewhere to send the login and the backup reports, and an email address is the whole identity we hold. No name, no ID document, no card, no phone number. Use any mailbox you like.
What does "hardened" actually mean here? Firewall with only SSH and the tunnel exposed, the gateway port behind authentication and off the public internet, per-customer VM isolation, unattended security updates, staged agent updates with a snapshot before each one, a vetted skills allow-list, and outbound port 25 blocked. The security checklist lists each control and why it exists.
Can the agent browse the web? On Pro and Business, yes; headless browser automation is installed and sized for. On Starter it is not enabled because 4 GB is tight once a browser opens, and an agent that swaps is worse than one that cannot browse.
Can I run more than one agent? Starter runs one. Pro runs up to three. Business is unlimited within the box's resources.
What happens if I underpay or the price quote expires? A short payment is credited to your wallet and the checkout shows a "pay the difference" button for the remainder. An expired quote is re-quoted with one click at the current rate. Overpayments are credited to your wallet within minutes.
Can you see my API key or my chats? The key is on the machine we manage, in the agent's secrets store, and support access to the VM is logged. We do not read your conversations; they stay in the agent's storage on your instance. Rotate the key at your provider any time.
What is the refund policy? Credit to your MartCC wallet within one working day if you cancel within 7 days of provisioning and the acceptable-use policy was respected. After that, cancellation stops renewal. Credit never expires and spends on any product in the store.
Related products and guides
- Managed OpenClaw hosting: the one-click product page with screenshots of each step
- Best OpenClaw hosting providers compared: dated price and feature table
- OpenClaw VPS requirements: RAM, CPU, browser, and VPS versus Mac mini
- OpenClaw security hardening checklist
- OpenClaw API key setup (BYOK)
- How to pay for AI APIs with crypto
- Windows VPS and RDP paid in crypto: a plain machine if you want to do it yourself
- Private cloud storage: a Nextcloud instance your agent can read and write to
- Pay with USDT and pay with Monero
[IMAGE: Diagram placeholder. Alt text: "Request flow from a phone chat app through an authenticated tunnel into an isolated VPS, where the agent gateway on loopback calls the customer's own model API; a crossed-out arrow shows the public internet blocked from the gateway port." Brief: flat, modern, no text other than port labels, palette as above, wide format.]
Sources
- OpenClaw documentation, Security and gateway FAQ (loopback bind, container default bind, remote access via tunnels,
openclaw security audit), https://docs.openclaw.ai/gateway/security (read 3 Oct 2026) - Hostinger, "Best OpenClaw hosting" (4 GB minimum, 8 GB for browser automation; VPS and managed prices; no listed host accepting crypto), https://www.hostinger.com/tutorials/best-openclaw-hosting/ (14 Aug 2026)
- Agent 37, "Best managed OpenClaw hosting in 2026" (managed tiers $3.99 to $99.99; "most serious managed hosts expect BYOK"), https://www.agent37.com/blog/best-managed-openclaw-hosting-in-2026 (7 Aug 2026)
- Medium, "OpenClaw in 2026: what's changed since the chaos started" (63% of 42,000+ exposed instances without gateway auth; about 824 malicious skills, about 20% of the marketplace; CVE-2026-25253), secondary source, cited from research notes dated 30 Sep 2026
- gradually.ai, "OpenClaw statistics" (155,000+ exposed instances; prompt injection in 36% of ClawHub skills), https://www.gradually.ai/en/openclaw-statistics/ (22 Sep 2026; unverified secondary data)
- Apple Newsroom, "The new Mac mini and Mac Studio are available today" ($899 starting price), https://www.apple.com/newsroom/2026/09/the-new-mac-mini-and-mac-studio-are-available-today/ (22 Sep 2026)
- MartCC pricing sheet, Phase 2 launch plans (3 Oct 2026)
Ready when your wallet is
Pick a product, choose a coin, and have it on screen before the kettle boils. No card, no forms, no waiting on a bank.