Hermes How-To · Reference
Hermes Glossary
Plain-English definitions of the terms used across the Hermes How-To series. No jargon left unexplained. Each term links to the episode that shows it in action.
New to running an AI agent in a terminal? This page explains every term the series uses, in plain language. Skim it once, or jump back whenever a word trips you up. Each definition links to the episode where you see the idea in action.
55 terms, plain English
- .env file
- A hidden file that holds secrets like API keys, separate from your regular settings. Hermes keeps it in ~/.hermes so your keys stay out of shared config.Learn it in: Env Var Cheat Sheet, Configuration Deep Dive
- ACP (Agent Client Protocol)
- A shared standard that lets Hermes run inside code editors like VS Code, Zed, and JetBrains. Chat, file changes, and terminal output all show up inside your editor instead of a separate window.Learn it in: In Your Editor (ACP)
- Agent (AI agent)
- A program that uses an AI model to actually do things for you, not just chat. It can read files, run commands, search the web, and take real actions on your machine. Hermes is an AI agent.Learn it in: What It Actually Is
- AGENTS.md
- A plain-text file you drop in a project so Hermes knows the rules and layout of that project: where things live, what conventions to follow, what to avoid. It is about the work, not the agent's personality.Learn it in: AGENTS.md
- Allowlist
- A list of exactly who is allowed to talk to your agent when it runs as a chat bot. Anyone not on the list is ignored. It is how you keep a public-facing bot from answering strangers.Learn it in: Security · Approvals, Hermes on Telegram
- API
- Short for Application Programming Interface: a way for one program to talk to another over the internet. Hermes calls an AI provider's API to get answers from a model.Learn it in: The Provider Landscape
- API key
- A secret password-like string that proves you are allowed to use a paid service such as an AI provider. You paste it in once and Hermes stores it locally. Treat it like a password and never share it.
- API server mode
- Running Hermes so other apps can talk to it the same way they talk to OpenAI. This lets chat front-ends like Open WebUI or LibreChat use Hermes as their brain.Learn it in: As an API Server
- Batch processing
- Running the agent across hundreds or thousands of prompts at once instead of one at a time. Useful for bulk jobs and for generating training or evaluation data.Learn it in: Batch Processing
- Bot token
- A secret code a chat platform (like Telegram) gives you when you create a bot. Hermes uses it to log in and act as that bot. Anyone who has the token controls the bot, so keep it secret.Learn it in: Hermes on Telegram
- Browser automation
- Letting the agent drive a real web browser: open pages, click, fill forms, and read what is on screen. It is how Hermes does things on websites that have no API.Learn it in: Browser Automation
- Checkpoint / Rollback
- Hermes quietly takes a snapshot of your files before it changes them. If a change goes wrong, /rollback puts things back the way they were. It is an undo button for the agent's edits.Learn it in: Checkpoints & Rollback
- CLI (command-line interface)
- Driving a program by typing commands instead of clicking buttons. Hermes is primarily a CLI tool: you talk to it by typing in your terminal.Learn it in: Your First Chat, Five Slash Commands
- Code execution
- A mode where the agent writes a short Python script that calls its own tools, so a multi-step job happens in one turn instead of many. It runs in a sandbox for safety.Learn it in: Code Execution: Runs Python
- config.yaml
- The main settings file for Hermes, kept in the ~/.hermes folder. It holds your preferences: which model, which tools, how things behave. You can edit it by hand or change it with one command.Learn it in: Configuration Deep Dive
- Context file
- A file Hermes loads automatically to shape how it works in a project, such as AGENTS.md, SOUL.md, or CLAUDE.md. It is how the agent picks up your rules without you re-explaining them every time.Learn it in: AGENTS.md
- Context window
- How much text a model can hold in mind at once, measured in tokens. When a conversation gets longer than the window, older parts have to be summarized or dropped.Learn it in: Pick Your Brain
- Credential pool
- A set of several API keys for the same provider that Hermes rotates between. If one key hits a rate limit or fails, it automatically tries the next.Learn it in: Credential Pools
- Cron / Scheduled task
- A way to make the agent run on a schedule: every morning, every hour, or once at a set time. You can ask in plain language and Hermes sets it up. The name comes from a classic Unix scheduler.Learn it in: Schedule With Cron
- Delegation / Subagent
- Hermes can spin off helper agents to work on parts of a task in parallel, each with its own limited tools and context. The main agent collects their results.Learn it in: Subagent Delegation
- Docker / Container
- A way to run software inside a sealed box that cannot touch the rest of your computer. Running Hermes' shell in Docker sandboxes what the agent can reach.Learn it in: Hermes in Docker
- Environment variable
- A named value your operating system or a tool reads at startup, like a setting passed in from outside. Hermes reads several to find your keys and tweak behavior.Learn it in: Env Var Cheat Sheet
- Fallback provider
- A backup provider Hermes switches to automatically if your main one errors out, so a single outage does not stop your work.Learn it in: Fallback Providers
- Gateway (messaging gateway)
- The piece of Hermes that connects your agent to a chat app like Telegram or Discord, so you can message it from your phone. You set it up once, then run it.Learn it in: Hermes on Telegram, Hermes on Discord
- HERMES_HOME
- The folder where one Hermes setup keeps everything: config, secrets, memory, skills, and sessions. By default it is ~/.hermes. Pointing it somewhere else is how separate profiles stay isolated.Learn it in: Profiles, Configuration Deep Dive
- Hook
- Custom code that runs automatically at a key moment, such as before a tool runs or when a message arrives. Hooks add logging, alerts, or guardrails without changing Hermes itself.Learn it in: Event Hooks
- Image generation
- Creating pictures from a text description. Hermes can do this through image models, and you pick which one to use.Learn it in: Image Generation
- LLM (large language model)
- The AI brain that reads and writes text. It is what Hermes sends your messages to and gets answers back from. ChatGPT, Claude, and Gemini are examples of large language models.Learn it in: What It Actually Is, The Provider Landscape
- Local model
- An AI model that runs on your own computer instead of a company's servers. It is private and free to run, but needs a capable machine. Tools like Ollama make it easy.Learn it in: Run Hermes with Ollama, Run Local LLMs on Mac
- MCP (Model Context Protocol)
- A standard way to plug ready-made tools into Hermes, such as GitHub, databases, or internal APIs, without writing code. You point Hermes at an MCP server and it discovers the tools.Learn it in: Connect Anything With MCP
- Memory
- Notes Hermes keeps so it remembers your setup and preferences across sessions, stored in MEMORY.md and USER.md. It curates these itself so you stop re-explaining your stack.Learn it in: It Remembers You
- Memory provider
- An optional external service that gives Hermes a richer, longer-term memory than the built-in files. You plug one in when you want more personalization.Learn it in: Memory Providers
- Model
- A specific AI brain you choose, like Claude Opus or a local Qwen. Different models have different strengths, speeds, and costs. You can switch models any time.Learn it in: The Provider Landscape, Configuring Models
- Nous Portal
- Nous Research's subscription gateway. One sign-in gets you hundreds of models plus web search, image generation, text-to-speech, and browser tools, with no separate API keys to manage. The easiest start for newcomers.Learn it in: Run Hermes with Nous Portal
- Open source
- Software whose code is public and free to use and change. Hermes is open source under the MIT license, so you can read it, run it, and modify it. You only pay for the AI model usage.Learn it in: What It Actually Is
- Personality
- The voice and manner the agent uses: direct or chatty, formal or casual. You set a durable personality in SOUL.md and can switch tone for a single chat with /personality.Learn it in: SOUL.md
- Plugin
- An add-on that gives Hermes new tools, hooks, or integrations without editing its core code. Some come built in; you can also build your own.Learn it in: Built-In Plugins Tour, Build Your Own Plugin
- Profile
- A completely separate Hermes setup with its own memory, sessions, skills, and bot, all on one machine. Use one profile for work and another for personal, and they never mix.Learn it in: Profiles
- Prompt
- What you type to the agent: your question, instruction, or request. The quality of your prompt shapes the quality of the answer.Learn it in: Your First Chat
- Provider
- The company or server that hosts the AI model you use, such as OpenAI, Anthropic, OpenRouter, or your own local server. Hermes works with any OpenAI-compatible provider.Learn it in: The Provider Landscape, Provider Routing
- Provider routing
- Rules that decide which provider handles a request, so you can optimize for cost, speed, or quality. You can prefer, rank, or block providers.Learn it in: Provider Routing
- Session
- One saved conversation with the agent, including its full history and tool actions. Hermes saves every session so you can resume, rename, or search them later.Learn it in: Sessions
- Shell
- The program that runs the commands you type in a terminal, such as bash, zsh, or PowerShell. When Hermes runs a command for you, it runs it in a shell.Learn it in: Your First Chat, Configuration Deep Dive
- Skill
- A reusable playbook the agent loads only when it needs it: step-by-step instructions for a task. Each skill also works as a slash command you can call by name.Learn it in: Skills
- Slash command
- A shortcut you type starting with / inside a chat, like /help or /model, to trigger a specific action. Every skill is also a slash command.Learn it in: Five Slash Commands
- SOUL.md
- The file that defines who your agent is: its identity, tone, and how it handles uncertainty. It is the first thing in the agent's system prompt and stays consistent across all your projects.Learn it in: SOUL.md
- Terminal
- The text window where you type commands to your computer instead of clicking. On Mac it is Terminal, on Windows it is PowerShell, on Linux it is your shell. Hermes lives here.
- Token (LLM token)
- The small chunks of text a model reads and writes, roughly a word or part of a word. Cost and the size of the context window are both measured in tokens.Learn it in: Pick Your Brain
- Tool
- Something the agent can do beyond talking: run a command, read a file, search the web, drive a browser, remember, schedule. Tools are what turn a chatbot into an agent.Learn it in: Tools & Toolsets
- Toolset
- A named group of related tools you can switch on or off together, such as web, terminal, file, or memory. You give the agent only the toolsets a job needs.Learn it in: Tools & Toolsets
- TTS (text-to-speech)
- Turning the agent's written replies into spoken audio. Hermes supports several voices and providers, including a free option.Learn it in: TTS, Voice Mode
- Vision
- The ability to understand images. Paste a screenshot or photo and ask the agent to read, describe, or work with it, using any vision-capable model.Learn it in: Vision · Paste
- Voice mode
- Talking to the agent with your microphone and hearing spoken replies, instead of typing. It works in the CLI and on messaging platforms.Learn it in: Voice Mode
- Web search
- Letting the agent look things up on the live web so its answers reflect current information, not just what the model already knew.Learn it in: Web Search
- Worktree (git worktree)
- A way to check out more than one branch of a code project at the same time, in separate folders. It lets the agent work on several things at once without them colliding.Learn it in: Git Worktrees