Starter
The always-on agent, without browser automation.
$10/mo
- + Up to 1 vCPU
- + 2 GB reserved, up to 3 GB
- + 10 GB persistent storage
- + Browser automation: Pro only
- + Token usage, monthly alerts, and optional pause
OpenClaw cloud // always-on, isolated
Managed OpenClaw VPS hosting, ready for your agent.
Run a personal AI agent in its own managed container from $10/month. Nova runs a tested OpenClaw release on an HTTPS address with persistent storage, daily restore points, a monthly token budget, and updates; you bring the model key.
// 40-SECOND TOUR
OpenClaw Hosting animated tour
// PLANS
Two right-sized tiers, one isolated agent
OpenClaw is open source, but model calls are not included. Bring your own provider key and pay that provider directly for tokens.
Pro
Double the resources plus a bundled Chromium browser for workflows that browse the web.
$20/mo
- + Up to 2 vCPU
- + 3 GB reserved, up to 4 GB
- + 20 GB persistent storage
- + Chromium browser automation included
- + Token usage, monthly alerts, and optional pause
Reserved memory is available to your agent; maximum memory is the hard container limit. Persistent storage survives restarts and approved updates and grows in place when Nova raises a plan size.
// MANAGED, NOT RAW ROOT
VPS-class resources without the server chores
A raw VPS gives you SSH and root, then leaves patching, container hardening, routing, process recovery, and upgrades to you. Nova’s managed OpenClaw hosting gives you the Control UI and focused lifecycle controls instead—there is no SSH or root access.
What Nova operates
- + A tested OpenClaw release, with one-click updates once Nova approves a new version
- + Your own container, persistent storage, fixed resources, and HTTPS address
- + A login in front of the Control UI plus OpenClaw’s own gateway token
- + Dashboard setup for provider keys and Telegram, Discord, or Slack
- + Daily restore points, a safety snapshot before every update, and one-click restore
- + Token usage, a monthly budget with alerts, and an optional pause at the cap
- + Memory and storage headroom, health diagnostics, and one-click repair
// WHAT GOES WRONG, AND WHAT NOVA DOES ABOUT IT
The three ways self-hosted OpenClaw lets people down
Across public issue trackers and hosting reviews, the same three failures come up again and again. Each one is covered here by default, not sold as an add-on.
An update wipes the agent’s memory
OpenClaw ships new releases every few days, and a bad upgrade or a full disk can take months of memory and chat history with it.
Nova takes a restore point every day and another before every approved update. Roll back from the dashboard in one click, or download a portable archive of your whole instance.
A runaway loop burns your API budget overnight
The default half-hourly check-in resends the full conversation every time, and a looping task can turn a $20 month into a four-figure bill before you wake up.
Nova shows token use by provider and model, estimates the cost, and lets you set a monthly budget with alerts at 80% and 100%, or pause the agent at the cap. New instances check in hourly in a lightweight session.
The gateway runs out of memory and only a shell can fix it
On a 2 GB box the gateway degrades silently before it crashes, and the documented fixes are command-line repairs you cannot run on most managed hosts.
Starter now allows up to 3 GB and Pro up to 4 GB, with 10 GB and 20 GB of storage. The dashboard shows memory, storage, and restart health, and a one-click repair runs OpenClaw’s own doctor and cleanup for you. Outgrow Starter and upgrade in place.
// SECURITY BOUNDARY
Isolation with an honest internet boundary
Contained by default
Each agent runs as a non-root user in its own locked-down container with a read-only system disk, fixed resources, and no access to Nova’s internal services or to other customers’ containers.
Public HTTPS remains open
The agent needs outbound HTTPS to reach your model provider, chat channels, and packages. If a prompt injection tricks it, it can still send data to any public website. Isolation limits what the agent can reach inside Nova; it does not make the agent incapable of leaking what you give it.
// SETUP
From checkout to your first channel
- 01
Choose a tier
Sign in and complete checkout. Nova provisions your instance automatically with its own HTTPS address.
- 02
Add your model key
Add your own provider key in the dashboard. Nova does not bundle model tokens.
- 03
Open the Control UI
Open it from your dashboard, sign in with the login Nova generates, then paste your gateway token.
- 04
Connect a channel
Add a Telegram, Discord, or Slack bot token in the dashboard. It shows you when the channel is live.
Why not hardware you already own?
A Mac Mini, a Raspberry Pi, or a Windows PC can run OpenClaw, but the agent only works while that machine stays awake, connected, patched, and reachable from the internet. Hosted OpenClaw removes that chore: the agent keeps running when your laptop closes, and nothing on your home network is exposed. The FAQ below covers each option.
// FAQ
OpenClaw hosting questions
What are the OpenClaw system requirements?
OpenClaw runs an always-on Node gateway, so it needs a machine that never sleeps, persistent storage for its configuration, memory, and workspace, and outbound HTTPS to your model provider and chat channels. In Nova’s testing the gateway uses about 1 GB of RAM at idle and peaks higher, so plan on 2 GB; the browser build needs more. Nova Starter provides up to 1 vCPU, 2 GB reserved / 3 GB maximum memory, and 10 GB of persistent storage. Pro provides up to 2 vCPU, 3 GB reserved / 4 GB maximum memory, 20 GB of storage, and a bundled Chromium browser.
How much RAM does OpenClaw need?
Plan on 2 GB. Starter reserves 2 GB and can use up to 3 GB. Pro reserves 3 GB and can use up to 4 GB, which the Chromium browser build needs.
How much does OpenClaw cost, and is OpenClaw free?
OpenClaw itself is open source. Nova managed hosting costs $10/month for Starter or $20/month for Pro. Model usage is separate: you bring your own supported provider key and pay that provider directly.
Are model tokens included?
Tokens are not included. You bring a supported provider key and your provider bills you directly. Nova shows gateway-reported token usage and OpenClaw’s estimated USD cost, and lets you set an optional monthly token budget with 80%/100% alerts or pause the agent at 100%. Sampling means this is not a provider-side hard cap. New Nova instances default heartbeat to once per hour in an isolated session. You can turn it off, change the interval, or let OpenClaw manage it.
How do I connect Claude or another model provider?
After checkout, add your own Anthropic, OpenAI, xAI, Gemini, Groq, DeepSeek, or OpenRouter key in the Nova dashboard. Nova stores it as a secret that belongs to your instance alone and restarts the agent to apply it.
Which chat channels can I connect?
The Nova dashboard configures Telegram, Discord, and Slack from a bot token and shows you when each channel is live. WhatsApp QR pairing is not available in this version.
Do I get SSH or root access?
No. This is a managed, isolated container rather than a raw root VPS. You use OpenClaw’s Control UI plus Nova controls for credentials, connections, restart, approved updates, logs, and usage.
Does Nova back up my OpenClaw state?
Yes. Nova briefly pauses your agent for a consistent daily storage snapshot, keeps 7 daily, 3 before-update, and 3 manual recovery points, and takes safety snapshots before approved updates and restores. You can restore from the dashboard or download a portable archive. Restores may replay scheduled work or require channel re-pairing; downloads contain plaintext credentials and must be protected.
Can I move from Starter to Pro?
Yes. Upgrade the same instance from its dashboard. Nova shows the prorated amount before charging, briefly stops the agent, takes a restore point when snapshots are available, expands the persistent disk, and restarts the same agent with Pro resources and Chromium. Your URL and state remain in place. Upgrades are one-way because persistent disks cannot shrink.
What happens if OpenClaw runs low on memory or storage?
The dashboard shows sampled memory, storage, restart, readiness, and scheduler diagnostics. You can restart the instance or run a repair that briefly takes it offline, runs OpenClaw Doctor, safely compacts eligible databases when space permits, and removes only known stale or expired artifacts. Repair cannot recover data that has already been deleted or replace a backup.
Can I cancel OpenClaw hosting?
Yes. You can schedule cancellation from your account. When the subscription ends, the instance enters a seven-day retention period before its persistent data is permanently purged.
How does hosted OpenClaw compare with a Mac Mini?
A Mac Mini can be a capable home host if you already own one and keep it awake, powered, patched, reachable, and secured. Nova is the simpler always-on option when you do not want to operate home hardware or expose it to the internet.
Can OpenClaw run on a Raspberry Pi?
It can run on compatible Raspberry Pi hardware, but memory, storage durability, cooling, and architecture support require hands-on care. Nova provides fixed cloud resources and avoids maintaining a home device.
Can OpenClaw run on Windows?
You can operate OpenClaw from Windows with an appropriate local container or Linux environment. Hosted OpenClaw moves the Linux runtime off your PC, so it keeps running when the PC sleeps or disconnects.