What happened
Nous Research's Hermes Cloud page describes deploying a hosted Hermes Agent on a dedicated cloud instance. A user chooses a name, region and model; the company says deployment takes two clicks and about 60 seconds. The instance has its own workspace and dashboard, and the agent can be accessed in a browser or connected to Telegram, Discord, Slack, email and CLI. The page says one cloud agent can use one memory across these surfaces.
The listed capabilities include always-on operation, natural-language scheduling for reports, backups and briefings, persistent memory, and isolated containers. Nous says each agent has its own hardened container and that users can spin up parallel subagents. It also says the service scales to zero when idle. The workspace, files, memory and configuration remain available when an agent is stopped, and users can later restart it. Instance size can be increased; reducing it is conditional on the disk fitting, since disks do not shrink.
Billing is taken daily in arrears from Nous credit, based on usage. While running, the instance incurs compute metered to the second of actual uptime plus storage; while stopped, only storage is charged. The page says charges usually appear within a day or two after the hosting provider finalizes usage. Inference and tool usage are billed separately from instance hosting. Deployment requires at least $2 in credits or an active subscription, while the current instance sizes and their prices are shown in the portal's deployment screen.
The page also lists Nous Portal plans: Free is pay-as-you-go and limited to free models; Plus costs $20 per month, Super $100 and Ultra $200. Paid plans include one cloud agent up to a stated maximum size: 4 GB RAM and 25 GB disk on Plus, 4 GB and 100 GB on Super, and 8 GB and 100 GB on Ultra. The page names possible server locations in Europe, Asia, North America and Australia, with the actual location selected based on proximity and availability.
What the archive adds
New Runtime's earlier coverage described Hermes Agent as a self-hosted, MIT-licensed open-source agent in an October 24 note, and covered the v0.23.0 release with a browser-served workspace and unified gateway in a November 6 note. This page adds hosting and billing details to that known product story; it is not evidence of a newly announced browser UI.
Trend signal
This extends the practical product boundary around an agent: a persistent workspace, dashboard, connected channels and isolated runtime are presented together as a hosted service. That is consistent with agent security moving to runtime boundaries, although the page does not explain the container's security controls in enough detail to assess them. It does not establish a general trend by itself.
Why it matters
For a team evaluating an agent that must run when nobody is at a laptop, hosted deployment removes server setup and offers a persistent place for memory and files. The trade-off is that “always on” does not mean one simple all-inclusive price: compute depends on uptime, storage remains billable while stopped, and model inference and tool use are extra. Teams should estimate the full workload rather than compare only subscription tiers.
The multi-channel setup also makes a single agent reachable from more than one interface, while scheduled execution points toward unattended routines such as reports and backups. Those capabilities can reduce operational friction, but they do not establish how reliably the agent completes a task or what approvals and safeguards apply. A useful pilot would define bounded tasks, track actual usage charges, and inspect what the agent can access before connecting production accounts.
Limits and uncertainty
Nous's page does not give current per-size instance prices in the captured pricing table; it directs users to the deployment portal. It does not quantify typical compute or storage bills, inference costs, tool charges, uptime guarantees or task success rates. The page describes containers as hardened but does not detail isolation mechanisms or independent security testing. Its statements about persistent memory and increasing capability over time are product claims, not independently measured results. Availability of listed regions depends on capacity.
Skill to build now
Before deploying an always-on agent, write a small cost model with separate lines for running compute, stopped storage, inference and tool use. Test one scheduled task with a defined completion check, stop the instance when it is not needed, and compare the bill with the cost of your current hosting approach.
Reader takeaway
Hermes Cloud packages a persistent Hermes Agent, dashboard, scheduling and messaging access into a managed instance. The practical decision is not simply whether to subscribe: estimate runtime, storage and model/tool usage separately, and verify the boundaries of the agent's workspace before assigning unattended work.