An Introduction to the Hermes Agent

The Hermes Agent is an open-source AI agent developed by Nous Research, engineered to execute tasks, retain knowledge, and evolve its capabilities over time. Rather than simply generating text, it leverages tools such as the terminal, web browser, file system, web search, and code execution environments to accomplish objectives.

Understanding AI Agents

AI agents transcend basic text generation. When assigned a goal, they determine necessary actions, deploy appropriate tools, monitor outcomes, and iteratively work towards achieving that goal.

While many AI agents can already utilise tools to complete multi-step processes, the critical distinction lies in post-task behaviour. Some agents treat each interaction as an isolated event. In contrast, Hermes is architected to preserve valuable knowledge and procedures, thereby enhancing its utility with each interaction. For details on how this applies to executing agents on DaDesktop, please refer to Agentic AI.

What Distinguishes Hermes?

Hermes features an integrated learning loop. It derives skills from practical experience, refines these skills during operation, retains pertinent information, and queries past sessions when requiring data from prior work.

Reasons to Choose the Hermes Agent

Hermes is particularly advantageous for tasks involving multiple steps, tools, or applications. Instead of constantly switching between AI chat interfaces, terminals, browsers, and files, you can assign the entire objective to Hermes and allow it to coordinate the workflow.

• Complete Whole Tasks: Provide Hermes with a high-level objective rather than seeking individual instructions.
• Maintain Long-Term Context: Preserve useful information across distinct sessions.
• Create Reusable Workflows: Store effective approaches as skills to avoid re-solving identical problems from scratch.
• Automate Routine Work: Schedule tasks to execute autonomously in the background.
• Delegate Responsibilities: Employ separate sub-agents for different components of a larger project.
• Select Your Model: Connect Hermes to local models or supported cloud-based model providers.

Capabilities of Hermes

• Software Development: Analyse code, debug issues, implement changes, execute tests, and manage repositories.
• Research: Search the web, review sources, gather data, and structure findings.
• Automation: Execute scripts and schedule recurring duties such as reporting, monitoring, and other routines.
• System Administration: Utilise the terminal, install software, configure systems, and resolve technical issues.
• Browser Operations: Navigate websites, interact with web pages, and extract information.
• File Management: Read, create, edit, search, and organise files.
• Tool Integration: Connect additional services and tools via MCP.

Continuous Improvement Through Usage

Consider assigning Hermes a complex, multi-step task. During the resolution process, Hermes may identify an effective workflow. Instead of discarding this knowledge upon task completion, it can save the workflow as a skill.

When a similar task arises in the future, Hermes can utilise this saved skill as a foundational step. Should the workflow require enhancement, the skill can be updated accordingly. Its memory also retains key facts and lessons learned from previous engagements.

Memory and skills serve complementary roles. Memory preserves contextual information, while skills encapsulate procedures. Combined, they enable Hermes to accumulate deep knowledge about both its understanding and the execution of recurring tasks.

Background Automation

Hermes includes a built-in scheduler for recurring duties. You can define schedules using natural language, allowing Hermes to perform work independently without requiring you to initiate a session.

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

Local or Cloud Model Support

Hermes is not restricted to a single Large Language Model (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 local models provides greater control over the model, hardware, and data processing location. It also enables Hermes to interact with your files and other local resources without transmitting them to external model providers.

Installing Hermes

Hermes is compatible with Linux, macOS, Windows, and WSL2. You may install the desktop application or utilise the command-line interface without the desktop component.

On Linux, macOS, or WSL2, open a terminal and execute the following:

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

Following installation, run the setup wizard to configure your model provider and other preferences:

hermes setup

Consult the Hermes documentation for detailed installation options and configuration guidelines.

Executing Hermes on DaDesktop

Hermes requires an environment capable of running its tools and storing its data. DaDesktop offers a cloud desktop equipped with the necessary CPU, RAM, storage, and GPU resources to run Hermes alongside a local LLM.

This setup is particularly useful when the desired model requires more VRAM than your local machine provides. Select a GPU with sufficient VRAM, launch a DaDesktop instance, install Hermes, and run the model locally in conjunction with the agent.

You can also keep the desktop active for scheduled tasks and long-running workflows, eliminating the need to keep your personal computer on.

Learn more about running local LLMs on DaDesktop.