Hermes Agent review

Hermes Agent is an open-source agent runtime, not a hosted API or a model. You install it on a computer or deploy an agent to the cloud through Nous Portal, then connect it to a model provider. The agent sits between that provider, its tools and the interfaces where you use it: desktop, terminal, or messaging platforms such as Telegram, Discord, Slack, WhatsApp, Signal and email.

The runtime keeps conversations, memories and skills between sessions. It can schedule tasks through its gateway, browse the web, use media tools and delegate work to subagents with separate conversations and terminals. For execution isolation, it offers local, Docker, SSH, Singularity and Modal backends.

That makes Hermes an orchestration layer for persistent, tool-using workflows rather than a drop-in inference endpoint. You can run it on your own computer or choose optional cloud hosting. A local process stops when its machine stops; saved data remains, but tasks needing an active agent will not run until it is started again.

The documented integration surface is the Hermes CLI, desktop app and platform connections. The homepage provides a terminal install command for macOS, Linux and Windows, desktop downloads for macOS 12+ and Windows 10/11, and setup steps for connecting a model provider. It also points users to documentation for configuring messaging integrations.

The homepage does not name a public REST API or language-specific SDK. It does describe Python RPC scripts as an option for isolated subagents, but that is not evidence of a general-purpose Python SDK. Confirm SDK support and any programmatic API in the vendor's docs before designing an integration around them.

For developer workflows, the CLI and gateway are the clearest entry points. Supported messaging surfaces include Telegram, Discord, Slack, WhatsApp, Signal and email. The tool also includes web search, browser automation, vision, image generation and text-to-speech. Expect to configure the provider and integrations as part of setup rather than call Hermes as a hosted API.

The software is free and open source under the MIT license, and Hermes can run on your own computer. The Free Nous Portal plan costs $0 per month and is for free models only; it has standard rate limits, no hosted tools and no cloud agent hosting. You can also use your own provider, which has its own pricing.

For hosted tools and a cloud agent, the pricing page lists Plus at $20 per month, Super at $100 and Ultra at $200. Each includes one cloud agent: Plus provides 4 GB RAM and 25 GB disk; Super, 4 GB RAM and 100 GB disk; Ultra, 8 GB RAM and 100 GB disk. Paid tiers include shared usage, free and paid models, and high rate limits. Super is listed at 5× Plus usage and Ultra at 10×.

Pay-as-you-go credit top-ups are available. Extra or larger cloud agents are billed daily from credits, so compute needs can add to the monthly plan. Annual prices are not published. The vendor does not publish numeric request limits, so check current Portal terms before estimating a production workload.

Hermes Agent is open source under the MIT license. You can run it on your own computer, and the homepage lists local, Docker, SSH, Singularity and Modal sandbox backends. Cloud hosting through Nous Portal is optional, not a requirement for using the agent.

Hermes stores conversations, memories and skills between sessions. The vendor recommends keeping a backup of the Hermes data directory when moving to another computer. That gives you a practical migration path for saved data, though the homepage does not describe a formal export format or a one-click migration tool.

The model provider and hosted services are separate choices: the vendor says you can use your own provider or connect to Nous Portal for model and tool credits. That reduces dependence on one hosted service, but workflows may still need adjustment when you change providers, tools or runtime environments.

The main production constraint is runtime availability. In a Reddit thread about hosting, a developer reports that long-running or scheduled work stops when a laptop sleeps. The vendor says tasks needing an active agent will not run while it is stopped; an always-on deployment needs a host that stays available or optional cloud hosting.

The sandbox choices include Docker, SSH and other backends, and the homepage says they include container hardening and namespace isolation. Treat that as a starting point, not a substitute for reviewing permissions, secrets handling and your own threat model. The vendor does not publish numeric rate limits or an uptime commitment on the homepage, so confirm those before planning a service around hosted tiers.

Reddit users also emphasize model selection and memory configuration. One commenter says a small default memory setup did not retain enough context for their work; others report that model fit affects agent task quality. Test your own workflows, schedule monitoring and back up the Hermes data directory.