What is Hermes Agent?

Hermes Agent is an open-source AI agent developed by Nous Research, designed to execute tasks, retain knowledge, and improve over time. It leverages tools such as terminals, browsers, file systems, web search, and code execution to perform work, rather than merely generating text responses.

What is an AI agent?

An AI agent transcends simple text generation. When given a goal, it determines the necessary actions, utilises available tools, observes the outcomes, and continues working towards the objective.

Many AI agents are already capable of using tools to complete multi-step tasks. The distinguishing factor lies in what happens after the task is completed. While some agents treat each task as a fresh problem, Hermes is engineered to retain useful knowledge and procedures, becoming more effective with time. See Agentic AI for insights on how this applies to running agents on DaDesktop.

What makes Hermes different?

Hermes features a built-in learning loop. It can generate skills from experience, refine those skills during use, remember valuable information, and search previous sessions when it requires context from earlier work.

Why use Hermes Agent?

Hermes is particularly beneficial when a task involves multiple steps, tools, or applications. Instead of constantly switching between an AI chat, terminal, browser, and files, you can delegate the overall task to Hermes and let it coordinate the work.

• Work on complete tasks: assign Hermes an objective rather than requesting individual instructions.
• Keep context over time: retain useful information across separate sessions.
• Build reusable workflows: save successful approaches as skills rather than solving the same problem from scratch each time.
• Automate recurring work: schedule tasks to run automatically in the background.
• Delegate work: utilise separate subagents for different components of a larger task.
• Choose your model: connect Hermes to local models or supported cloud model providers.

What Hermes can do

• Software development: inspect code, debug issues, make changes, run tests, and work with repositories.
• Research: search the web, read sources, collect information, and organise findings.
• Automation: run scripts and schedule recurring tasks such as reports, monitoring, and other routines.
• System administration: use the terminal, install software, configure systems, and troubleshoot issues.
• Browser tasks: navigate websites, interact with web pages, and collect information.
• File work: read, create, edit, search, and organise files.
• Tool integrations: connect additional services and tools via MCP.

How Hermes improves with use

Suppose you give Hermes a complex task requiring several steps. While solving it, Hermes may identify an effective workflow. It can save this workflow as a skill, ensuring the knowledge is not lost when the task concludes.

Next time a similar task arises, Hermes can use that skill as a starting point. If the workflow requires refinement, the skill can be updated. Its memory can also retain useful facts and lessons learned from previous work.

Memory and skills serve distinct purposes. Memory stores useful information and context, while skills store procedures for executing tasks. Together, they enable Hermes to build comprehensive knowledge about both what it knows and how to perform recurring work.

Automate work in the background

Hermes includes a built-in scheduler for recurring tasks. You can describe a schedule in natural language, and Hermes will perform the work without waiting for you to initiate a session.

This is ideal for reports, monitoring, backups, research tasks, and other routines. Scheduled tasks can also maintain continuity between runs, allowing them to build on previous work.

Run local or cloud models

Hermes is not restricted to a single LLM. You can connect it to local models or supported cloud providers and switch models without altering the agent's tools, memory, or workflows.

Running a local model offers greater control over the model, hardware, and data processing location. It also allows Hermes to work with your files and other local resources without sending them to an external model provider.

How to install Hermes

Hermes runs on Linux, macOS, Windows, and WSL2. You can install the desktop application or use the command-line version without the desktop app.

On Linux, macOS, or WSL2, open a terminal and run:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

On Windows, run the installer from PowerShell:

irm https://hermes-agent.nousresearch.com/install.ps1 | iex

After installation, run the setup wizard to configure your model provider and other options:

hermes setup

Refer to the Hermes documentation for installation options and configuration details.

Run Hermes on DaDesktop

Hermes requires an environment where it can execute tools and store data. DaDesktop provides a cloud desktop equipped with the CPU, RAM, storage, and GPU resources necessary to run Hermes alongside a local LLM.

This is particularly useful when the model you wish to run requires more VRAM than your local machine possesses. Choose a GPU with sufficient VRAM, launch a DaDesktop desktop, install Hermes, and run the model locally alongside the agent.

You can also keep the desktop running for scheduled tasks and long-running workflows without needing to leave your personal computer on.

Learn more about running local LLMs on DaDesktop.