Glossary
Quick definitions. No fluff. Click through to the full page if you want more depth.
| Term | Definition |
|---|---|
| Assistant | Your personal AI entity. Has a name, personality, memory, and the ability to take actions on your behalf. |
| Workspace | The folder at ~/.vellum/workspace/ containing your assistant's configuration, identity, and skills. Your assistant's brain, in plain text. |
| Skill | A bundle of tools and instructions that gives your assistant a specific capability (e.g., Email, Weather, DoorDash). Loaded on demand. |
| Tool | A single atomic action your assistant can take (e.g., web_search, file_read, browser_navigate). The building blocks that skills are made of. |
| Channel | A way to communicate with your assistant. Currently: desktop app and voice. More coming. |
| Memory | Information your assistant saves for future reference. Facts, preferences, decisions, and learnings. Stored locally, searchable. |
| SOUL.md | The file defining your assistant's personality, principles, preferences, and behavioral boundaries. |
| IDENTITY.md | The file containing your assistant's name, role, personality summary, and emoji. |
| USER.md | The file where your assistant stores what it's learned about you. |
| LOOKS.md | The file controlling your assistant's visual avatar appearance. |
| Home Base | Your personalized dashboard in the desktop app. A starting point with quick actions and suggested tasks. |
| Sandbox | A secure, isolated environment where your assistant runs code and tools without touching your actual system. |
| Host tools | Tools that run on your actual machine (outside the sandbox). These require explicit permission via Allow / Don't Allow prompts. |
| Permission prompt | The Allow / Don't Allow dialog that appears when your assistant needs to do something sensitive on your machine. |
| Playbook | Trigger-action automation rules for handling incoming messages. |
| Watcher | A polling system for monitoring external sources and triggering actions when something changes. |
| Schedule | A recurring automated task that runs on a cron or RRULE pattern (e.g., “every weekday at 9am”). |
| Reminder | A one-time notification that fires at a specific future time. |
| Task | A work item in your personal task queue. Can be ad-hoc or based on a reusable task template. |
| AgentMail | The email service that gives your assistant its own email address, separate from yours. |
| Dynamic UI | Interactive visual surfaces (cards, tables, apps, dashboards) that your assistant creates on the fly. |
| Token scopes | Granular permission controls that define what your assistant can access and do within connected services. |
That's the conceptual foundation. You now know more about how your assistant works than most people know about their own email client. Next up: Guides for the practical “how do I...” stuff, or Trust & Security if you want the full picture on privacy and permissions.
On this page