{"id":79464,"date":"2026-09-13T15:24:36","date_gmt":"2026-09-13T15:24:36","guid":{"rendered":"https:\/\/www.arimetrics.com\/?post_type=encyclopedia&#038;p=79464"},"modified":"2026-09-13T16:28:10","modified_gmt":"2026-09-13T16:28:10","slug":"hermes","status":"publish","type":"encyclopedia","link":"https:\/\/www.arimetrics.com\/en\/digital-glossary\/hermes","title":{"rendered":"Hermes"},"content":{"rendered":"<p><img decoding=\"async\" src=\"https:\/\/www.arimetrics.com\/wp-content\/uploads\/2026\/09\/hermes-agent-logo-square.jpg\" alt=\"Hermes\" class=\"boxpad alignright\" style=\"margin-top:0\" width=\"300\" height=\"300\" \/><strong>Definition:<\/strong><\/p>\n<p><strong>Hermes<\/strong> is an open-source artificial intelligence agent developed by Nous Research that combines tools, persistent memory and reusable procedures to carry out tasks. Its official name is Hermes Agent, and it can run on a self-hosted computer or server using different model providers.<\/p>\n<p>It supports work from the terminal and through connected interfaces or communication channels. One of its features is the ability to retain information and create or update instructions based on completed work for use in subsequent tasks.<\/p>\n\n<h2>Hermes Agent and the Hermes models<\/h2>\n<p>Nous Research develops both Hermes Agent and a family of language models called Hermes. They are different components: the agent is an application that organises conversations, tools and tasks; the models process context and generate responses.<\/p>\n<p>Hermes Agent does not require a model from the Hermes family. It supports different providers and connections to models hosted locally or through external services.<\/p>\n<p>The agent&#8217;s code is distributed under the MIT licence. This licence permits use, modification and redistribution under its terms, but does not determine the licences or costs of connected models and services.<\/p>\n<h2>How Hermes works<\/h2>\n<p>Hermes receives a request and processes it with the configured model, together with the available instructions, memory and context. The model can respond directly or request tools to obtain information and perform operations.<\/p>\n<p>The results of those tools are added to the conversation. The agent can use them to continue the work, correct a step or prepare a response.<\/p>\n<p>Its operation brings together several capabilities:<\/p>\n<ul>\n<li><strong>Tool use:<\/strong> reading and modifying files, running commands or looking up information, depending on the enabled integrations.<\/li>\n<li><strong>Session management:<\/strong> retaining history and retrieving previous conversations.<\/li>\n<li><strong>Persistent memory:<\/strong> storing relevant preferences, facts and decisions.<\/li>\n<li><strong>Skills:<\/strong> reusable instructions that describe procedures or collect knowledge about a task.<\/li>\n<li><strong>Delegation:<\/strong> distributing work among subagents with separate contexts.<\/li>\n<\/ul>\n<p>The available capabilities depend on the configuration. An installation can have a limited set of tools or incorporate additional connections.<\/p>\n<h2>Memory and procedural learning<\/h2>\n<p>Hermes distinguishes between user information, agent notes and conversation history. Its built-in memory uses files such as <code>MEMORY.md<\/code> and <code>USER.md<\/code>, while session search retrieves information from previous exchanges.<\/p>\n<p><strong>Skills<\/strong> serve a different purpose: they retain procedures and knowledge that the agent can load when needed. They are organised around a <code>SKILL.md<\/code> file and may include references or supporting resources.<\/p>\n<p>Hermes can create a skill from a completed task or supplied materials, and modify it later. For example, a procedure for transforming a particular file type can record its structure, the conversion steps and the necessary checks.<\/p>\n<p><strong>This learning is not the same as retraining the language model.<\/strong> What changes is the information and instructions available for future tasks, not necessarily the model&#8217;s parameters.<\/p>\n<p>Keeping a skill does not, by itself, demonstrate that the procedure is correct. Its contents can be reviewed and updated, like any other operational documentation.<\/p>\n<h2>Interfaces and scheduled tasks<\/h2>\n<p>Hermes provides a terminal interface and a desktop application. Its messaging gateway connects it to platforms such as Telegram, Discord and Slack, depending on the configured integrations.<\/p>\n<p>The gateway receives messages and returns responses through those channels. The application can run on a server even when the conversation takes place from another device.<\/p>\n<p>The scheduling system supports one-off and recurring tasks. These can be associated with one or more skills and deliver results to a configured channel or local files. A recurring task can, for example, consult selected sources and prepare a periodic summary.<\/p>\n<p>There are also scheduled tasks that run a script without involving a language model. These differ from agent tasks, in which a model interprets instructions and decides which tools to use.<\/p>\n<h2>Installation and initial configuration<\/h2>\n<p>Hermes offers options for Linux, macOS and Windows, including native Windows. Hermes Desktop provides a graphical installation route for macOS and Windows; a command-line installation is also available.<\/p>\n<p>The <a href=\"https:\/\/github.com\/NousResearch\/hermes-agent\/blob\/v2026.9.11\/website\/docs\/getting-started\/quickstart.md\" target=\"_blank\" rel=\"noopener\">getting started guide for the reviewed version<\/a> distinguishes software installation, model selection and connecting tools or channels.<\/p>\n<p>The main configuration commands are as follows:<\/p>\n<ul>\n<li><strong><code>hermes setup<\/code>:<\/strong> starts the setup wizard.<\/li>\n<li><strong><code>hermes model<\/code>:<\/strong> selects the provider and model.<\/li>\n<li><strong><code>hermes tools<\/code>:<\/strong> configures enabled tools.<\/li>\n<li><strong><code>hermes<\/code>:<\/strong> opens a conversation in the terminal.<\/li>\n<li><strong><code>hermes gateway setup<\/code>:<\/strong> configures messaging connections.<\/li>\n<li><strong><code>hermes doctor<\/code>:<\/strong> runs diagnostic checks.<\/li>\n<\/ul>\n<p>The configuration makes it possible to start with minimal capabilities and enable others later. Memory, browsing, scheduling and integration features do not all have to be active in every installation.<\/p>\n<h2>Relationship with OpenClaw and tool execution<\/h2>\n<p>Hermes and <strong><a href=\"https:\/\/www.arimetrics.com\/en\/digital-glossary\/openclaw-2\">OpenClaw<\/a><\/strong> are independent projects that share capabilities such as external models, tools, memory and messaging channels. Hermes includes a utility for importing selected OpenClaw data and settings, although that migration does not make the two systems identical applications.<\/p>\n<p>Hermes can run tools in different environments, including the local computer, containers and remote systems. The selected environment determines where operations take place and which resources are available.<\/p>\n<p>Running the agent on a self-hosted computer does not mean that all processing is local. If it uses a remote model, requests are sent to the corresponding provider; connected tools may also exchange data with other services.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>What Hermes Agent by Nous Research is: how it works, memory, skills, scheduled tasks, installation and differences from the Hermes models.<\/p>\n","protected":false},"author":33,"featured_media":79471,"template":"","encyclopedia-tag":[1385],"class_list":["post-79464","encyclopedia","type-encyclopedia","status-publish","has-post-thumbnail","hentry","encyclopedia-tag-ai-agents"],"_links":{"self":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia\/79464","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia"}],"about":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/types\/encyclopedia"}],"author":[{"embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/users\/33"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/media\/79471"}],"wp:attachment":[{"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/media?parent=79464"}],"wp:term":[{"taxonomy":"encyclopedia-tag","embeddable":true,"href":"https:\/\/www.arimetrics.com\/en\/wp-json\/wp\/v2\/encyclopedia-tag?post=79464"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}