Hermes Cloud Now Runs in a Browser Tab: Desktop-Grade Agents on Nous's Servers
When Nous Research first shipped Hermes Cloud, the service ran as a hosted instance you reached through the desktop app. Now the company has closed that gap: as of Nov 9, you open a Cloud agent in a normal browser tab and get the same full interface as Hermes Desktop. Chats, group chats, plugins, and settings all sync across phone, web, and desktop.
The angle is what it removes. Historically, running a persistent Hermes agent meant keeping your own always-on VPS, or at least a desktop app that stayed online. Cloud changes that. Nous provisions the container, the uptime, the storage, and the browser access. You pick a name, a model, and a server size in the Nous Portal, and deployment takes about 60 seconds from the portal page at portal.nousresearch.com/cloud. Your instance runs in its own dedicated container with its own workspace, and Nous places it in the nearest region that has capacity.
Nous now operates in 14 regions, so a nearby option is usually available: Amsterdam, Chicago, Dallas, Frankfurt, London, Los Angeles, Mumbai, Paris, San Jose, Singapore, Stockholm, Sydney, Tokyo, and Toronto.
The service runs 24/7. Compute can scale to zero while the agent is idle, but scheduled jobs and messaging channels keep working with the laptop closed. Long-running projects hold their state without manual sync.
You are not locked into the web tab. A single agent can connect to Telegram, Discord, Slack, email, and the CLI, and every channel shares the same memory and workspace, so the state you see in the browser is the state on your phone. For heavier work, subagents run in separate sandboxes with their own context histories, and they do not touch the main conversation's token budget. That keeps a large multi-step run from inflating the primary context, and it means parallel tasks stay cleanly isolated from one another.
Plans: Free at $0 (usage billed separately, no cloud agent included), Plus at $20/month with 4 GB RAM and 25 GB disk, Super at $100/month with 4 GB RAM and 100 GB disk, and Ultra at $200/month with 8 GB RAM and 100 GB disk. Plus, Super, and Ultra each include one Cloud agent out of the box. Deployment requires an active subscription or at least $2 in account credits. Compute is metered to the second while running, drops to a lower storage-only rate when stopped, and model inference and paid tool usage bill separately.
The model catalog spans Anthropic Claude, OpenAI GPT, Google Gemini, Meta Llama, Mistral, DeepSeek, Qwen, xAI Grok, and Nous Research's own Hermes models. One OAuth sign-in covers access across every provider, so you can drop the per-provider API keys.
Cloud is the hosted version of a self-hosted idea. On Nov 5, v0.23.0 Hearth shipped the webapp for self-hosted setups; Cloud is that same interface in managed form. On Nov 9, the v0.23.1 patch rolled roughly 618 pull requests since v0.23.0 into the stable channel across the CLI, Microsoft Store, Docker, and Hermes Cloud.
For anyone who wants a persistent agent without babysitting infrastructure, the browser build turns that from a side project into a setting.