Isolated environment
Every agent runs in a private container with a unique customer identity — not on a shared host.
AI AGENTS
Private agent runtimes. Not a server to administer.
Choose a managed agent, give it an AI connection and an approved channel, then launch it in an isolated runtime with its own workspace, limits and secrets. You never administer the host.
In development: the catalog, purchase flow and runtime described on this page are being built. No agent instance is sold as finished today.
Everything an agent instance ships with
One managed runtime with the capability set below — and nothing in it that expects you to log into a host.
Persistent workspace
The agent keeps a workspace that survives restarts, so its files and state are still there when it wakes up.
Private secrets
Secrets are stored as references and kept out of the workspace, out of logs and out of anyone else reach.
CPU and RAM limits
Each instance runs under resource caps configured by platform admins — caps, not claims of dedicated cores.
Backups
Agent backups cover the persistent workspace and its configuration, under the same policy as other project types.
Approved channels
Telegram and WhatsApp arrive as approved integrations. The channel list is curated, not open.
AI Router
Point the agent at Hostingsurge AI credits or OpenRouter-backed models through the platform AI router.
Bring your own key
BYOK is a first-class option: use your own provider API key when you prefer your own account and quota.
A curated catalog, not an open marketplace
Two agents are approved today. Further agents are added only after review, so every entry can be described honestly and runs under the same isolation rules.
Hermes Agent
A general-purpose agent for work that runs over time.
- Persistent memory across sessions
- Scheduled and recurring automation
- Browser and tool use inside its own workspace
- Sandboxed execution for whatever it runs
OpenClaw
A self-hosted gateway for agents, tools and channels.
- Channels as the entry point to your workflows
- Agents, tools and automation in one gateway
- Runs as a managed instance, never as your server
- Approved channels such as Telegram and WhatsApp
Nothing is running today: the agent module is still being built. There is no live purchase path and no launched instance yet — this page describes the runtime in development and the guarantees it is designed to hold.
From choice to running agent
Five steps, none of which involves opening a server console.
- Choose an agentPick Hermes Agent or OpenClaw from the approved catalog.
- Choose a planSelect a billing commitment. The total, the renewal terms and the tax status are shown before you pay.
- Choose an AI connectionHostingsurge AI credits, an OpenRouter-backed model, or your own API key through BYOK.
- Connect a channelAttach an approved channel such as Telegram or WhatsApp so the agent can be reached.
- LaunchThe instance starts in isolation with its own workspace, limits, secrets and backup policy — no server administration.
Bounded by design
An agent instance is an isolated managed runtime. These boundaries are written into the platform rather than left to settings you are trusted to get right.
Never granted to a customer agent
Public traffic reaches an agent only through an approved reverse proxy.
- Privileged mode
- The host Docker socket
- A host root mount
- Host networking
- Unrestricted east-west network access
Instance properties
- RuntimePrivate container per customer
- IdentityUnique customer identity
- WorkspacePersistent across restarts
- LimitsAdmin-configured CPU and RAM caps
- SecretsReferenced, never inlined
- Inbound trafficApproved reverse proxy only
- Scale-outDedicated AGENT node role
Node role in development: agents get their own AGENT node role as the fleet grows, and pilot deployments co-locate behind an explicit environment flag. Node-level isolation is designed, not yet operated at scale.
Indicative launch pricing
Every monthly figure below is shown next to what is actually billed, so nothing reads cheaper than it is.
| Plan | Billing commitment | Total billed upfront | Monthly equivalent | Renewal benchmark |
|---|---|---|---|---|
| 24-month plan | 24 months, paid upfront | $186.96 once, for 24 months | $7.79 / month equivalent | $15.59 / month benchmark |
| 12-month plan | 12 months, paid upfront | $124.68 once, for 12 months | $10.39 / month equivalent | $15.59 / month benchmark |
| Monthly | Rolling monthly, no upfront period | $14.29 each month | $14.29 / month | $15.59 / month benchmark |
Agent plans are separate from Hostingsurge Build Credits and from platform plans.
Read this before comparing prices: these are indicative launch prices, and tax is shown at checkout. Long-term plans are billed upfront for the whole period — the monthly equivalent is not a monthly charge. Renewal is benchmarked at $15.59 per month at launch. Resource limits are admin-configured caps, not claims of physically dedicated CPU or RAM.
Questions worth asking
Straight answers about what a managed agent is, and what it is not.
Do I administer a server?
No. You get a managed agent instance — a private container with a gateway, limits and backups. There is no SSH console, no host patching and no proxy to configure.
Can I bring my own AI provider key?
Yes. BYOK is a first-class option alongside Hostingsurge AI credits and OpenRouter-backed models, so you can keep your own provider account and quota.
Which channels can an agent use?
Approved integrations such as Telegram and WhatsApp. Channels are reviewed before they are added, so the list is deliberately shorter than on a general-purpose bot platform.
Is this a VPS?
No. An isolated container is not priced, described or presented as a VPS. A VPS is a virtual machine you administer; a managed agent instance is a runtime you simply use.
What happens when I stop using it?
Instances can be suspended and woken where agent functionality allows it. Agents that are scheduled or expected to be always on are not suspended, because that would break the behaviour you rely on.
Your automations. Your isolation.
One runtime, one workspace, one catalog — with the boundaries held by the platform instead of by your patience with servers.