Guides

What is Hermes Agent?

A practical overview of Hermes Agent by Nous Research: what it does, why builders use it, which models it supports, and how official skills work.

Updated June 1, 2026

Illustration of a winged AI agent standing in front of a bright blue and pink background.

Hermes Agent is an open-source AI agent from Nous Research. It runs as a long-lived assistant that can use tools, remember prior work, create skills from experience, and operate from the terminal, dashboard, or messaging apps such as Telegram, Discord, Slack, WhatsApp, Signal, and Teams.

Hermes Agent is getting attention because it is built for agent workflows that keep running beyond one chat tab. It supports persistent memory, tool use, model routing, subagents, messaging gateways, scheduled automations, MCP servers, and a first-class skills system. For builders experimenting with Claude Code, Codex, Cursor, OpenCode, local models, and open agent infrastructure, Hermes combines several agent patterns in one open-source project.

Why Hermes Agent is getting popular

Hermes Agent fits a specific moment in AI tooling. Builders are moving from single-turn chat to persistent agents that can keep context, use tools, and repeat workflows. Hermes gives that workflow a local home.

It is open source, with the core project in the NousResearch/hermes-agent GitHub repository.
It is model agnostic and can use Nous Portal, OpenRouter, Anthropic, OpenAI, Gemini, and custom endpoints.
It works with local OpenAI-compatible runtimes such as Ollama.
It has persistent memory, so sessions and memories can survive across interactions.
It supports skills natively, including discovery, loading, updating, and reuse.
It reaches the user outside the terminal through messaging surfaces.
It supports subagents for parallel tasks with isolated contexts.

This combination makes Hermes attractive for builders who want a personal agent, a research assistant, a coding agent, or an automation runner that can live on a laptop, VPS, or serverless backend.

Which models does Hermes Agent support?

Hermes Agent supports hosted and local models through several provider paths: Nous Portal, OpenRouter, Anthropic, OpenAI, Gemini, custom OpenAI-compatible endpoints, and local runtimes such as Ollama. Hermes uses a main model for user messages and tool-call loops, plus auxiliary models for context compression, vision, web extraction, approval scoring, MCP routing, session titles, and skill search.

What is Hermes Agent used for?

Hermes Agent is useful when the task benefits from continuity. A normal chatbot answers a message. A Hermes-style workflow can keep memory, use tools, hand work to subagents, load skills, and return through a messaging gateway.

Coding workflows: planning, debugging, PR review, GitHub issue handling, tests, and codebase inspection.
Research: web search, paper discovery, local knowledge base search, OSINT-style public record workflows, and recurring monitoring.
Personal productivity: notes, reminders, documents, email, and messaging.
MLOps and model work: fine-tuning, serving, evaluation, vector databases, structured generation, and GPU workflows.
Creative work: diagrams, video pipelines, Blender control, memes, SVG diagrams, and HTML-based video compositions.
Operations: Docker management, tunnels, webhooks, watchers, and deployment workflows.
Security: 1Password CLI usage, OSS forensics, username research, and authorized web pentest workflows.

Hermes becomes more useful when a workflow has steps, tools, files, credentials, memory, or repeatable procedures.

How Hermes Agent works with skills

Hermes skills are reusable instruction packages the agent can load when needed. Hermes stores skills in ~/.hermes/skills/ and supports external skill directories such as a shared ~/.agents/skills/ folder. Because Hermes skills use the open SKILL.md format, many of them can also be used with other skill-aware agents. Give your AI agent the install source and a clear prompt such as: “Install this skill, read its SKILL.md, and adapt it to your skill folder and command format.”

Each skill usually contains a SKILL.md file with frontmatter and instructions. A skill can also include references, scripts, templates, and assets. Hermes follows a progressive disclosure pattern: it can list available skills cheaply, then load the full skill only when the task calls for it.

Hermes makes installed skills available as slash commands:

/github-pr-workflow create a PR for the auth refactor
/plan design a rollout for migrating our auth provider
/excalidraw

It also supports natural-language skill discovery:

hermes chat --toolsets skills -q "What skills do you have?"
hermes chat --toolsets skills -q "Show me the axolotl skill"

Hermes stores skills as files on disk, so useful workflows can survive beyond a single chat. Through its skill_manage tool, Hermes can create, patch, edit, and delete skills, then reuse the saved workflow later.

How Hermes Skills Hub works

Hermes Skills Hub is the skill-management layer inside Hermes, exposed through the hermes skills CLI commands and documented in the Hermes skills docs. It handles browsing, searching, installing, auditing, updating, and uninstalling skills. It can install official optional skills, skills from skills.sh, well-known skill endpoints, and direct SKILL.md URLs.

hermes skills browse --source official
hermes skills search kubernetes
hermes skills inspect openai/skills/k8s
hermes skills install official/security/1password
hermes skills audit
hermes skills update

Skills are executable context. A skill can shape tool use, load scripts, use credentials, and change how the agent behaves. Browsing, inspecting, auditing, and updating skills are part of the workflow.

Official Nous Research skills

Nous Research ships two kinds of skills for Hermes Agent: bundled skills included with Hermes and copied into ~/.hermes/skills/ during install, plus optional official skills shipped under optional-skills/ and installed explicitly with hermes skills install official/<category>/<skill>.

Notable Hermes skills worth knowing

Hermes has a large official skill catalog, so the useful starting point is the subset that gives the agent practical behavior it can repeat.

Skill	Why it is worth using
`dogfood`	Use it to make Hermes test a web app like a QA pass: open pages, inspect console errors, collect screenshots, classify issues, and return a structured report.
`adversarial-ux-test`	Use it when a product needs stress testing from the perspective of a frustrated or confused user. It helps surface unclear copy, fragile flows, and UX assumptions.
`github-pr-workflow`	Use it to keep pull request work consistent: branch, commit, open the PR, monitor CI, and prepare the merge path.
`subagent-driven-development`	Use it for larger engineering tasks where separate agents can investigate, implement, or review isolated parts of the work.
`axolotl`	Use it for LLM fine-tuning workflows that need structured YAML configuration and repeatable training steps. It reflects Nous Research’s focus on open model work.

The dogfood skill is useful because it gives Hermes a testing procedure, browser-tool checklist, evidence format, and report template. It turns web application QA into a reusable agent workflow.

Official optional Nous Research skills by category

The optional catalog is broad. A category view makes the catalog easier to scan.

Category	Why the category is useful	Examples
Autonomous AI agents	Useful when Hermes needs to coordinate with other agent CLIs, memory systems, or autonomous coding environments.	`openhands`, `blackbox`, `grok`, `honcho`, `antigravity-cli`
Communication	Useful for decision-heavy work where the agent needs a repeatable framework for trade-offs and recommendations.	`one-three-one-rule`
Creative	Useful when the agent needs to generate visual or media assets with defined tools and output formats.	`blender-mcp`, `concept-diagrams`, `hyperframes`, `meme-generation`
DevOps	Useful when Hermes needs to operate local infrastructure, containers, tunnels, watchers, or AI app command-line tools.	`docker-management`, `pinggy-tunnel`, `watchers`, `inference-sh-cli`
Dogfood	Useful for testing products with repeatable QA and UX review workflows.	`adversarial-ux-test`
Finance	Useful when the agent needs structured spreadsheet models, valuation workflows, decks, or market context.	`excel-author`, `dcf-model`, `lbo-model`, `stocks`
MCP	Useful when Hermes needs to build, connect, or operate MCP servers and tools.	`fastmcp`, `mcporter`
MLOps	Useful for model builders working on fine-tuning, inference, vector search, structured generation, GPU cloud, data curation, or evaluation.	`axolotl`, `trl-fine-tuning`, `unsloth`, `vllm`, `qdrant`
Research	Useful when the agent needs repeatable search, OSINT, code graph, scientific, or local knowledge retrieval workflows.	`duckduckgo-search`, `osint-investigation`, `qmd`, `gitnexus-explorer`
Security	Useful when Hermes needs guided workflows for secrets tooling, supply-chain investigation, username research, or authorized web testing.	`1password`, `oss-forensics`, `sherlock`, `web-pentest`
Software development	Useful when the agent needs to document a codebase or debug REST and GraphQL APIs with a repeatable process.	`code-wiki`, `rest-graphql-debug`
Web development	Useful when the agent needs to control web app UI through natural-language instructions.	`page-agent`
Blockchain	Useful for read-only chain analysis, wallet inspection, market data, and transaction context.	`evm`, `solana`, `hyperliquid`

Bundled Hermes skills worth knowing

Bundled skills are active after installation. They cover everyday workflows such as planning, GitHub PRs, code review, TDD, systematic debugging, subagent execution, Obsidian, Google Workspace, Notion, MCP, arXiv, and LLM wiki workflows. The bundled catalog gives Hermes a practical baseline. The optional catalog expands it into heavier workflows that users can install when they need them.

How to find official Hermes and Nous Research skills

Use the Hermes docs: the bundled skills catalog and optional skills catalog.
Check the NousResearch/hermes-agent GitHub repository.
Use hermes skills browse --source official inside Hermes.
Inspect before installing with hermes skills inspect.
Run hermes skills audit after installing third-party or hub skills.
Browse the Skillscout official skills directory when comparing official AI agent skills across vendors, docs, and GitHub sources.

Official skills are useful because they come from the project owner and match the way Hermes is designed to work. The same security rule still applies: inspect a skill before installing it, especially when it can run scripts, access credentials, or call external services.