How Agents Work: The Mental Model
Everything an agent can do comes from four kinds of building block. Keep them
straight and the rest of the app is easy.
- Abilities = what it can do. An ability is a switch that unlocks a *bundle of
tools*. Turn on "Web Access" and the agent gains search, weather, and maps tools.
See Abilities.
- Tools = the concrete actions. Each tool is one callable action. Tools always
arrive via an ability. How much detail about each tool is shown to the agent (and
whether it's loaded on demand) is the tool's exposure mode. See Tools & How They're Exposed.
- Skills = what it knows. A skill is a written playbook — guidance for a task —
that can be always-on or pulled in only when needed, so an agent can carry a big
library without bloating every message. See Skills (On-Demand Knowledge).
- Prompts = how it behaves. The standing instructions that set the agent's voice,
role, and rules. See Prompts & Context.
On top of those sit two cross-cutting systems:
- Memory — facts that persist across conversations and are recalled automatically.
See Memory.
- The agent loop — the repeating cycle of think → call tools → read results →
think again that produces each answer. See The Agent Loop.
The slogan: **abilities = tools, skills = knowledge, prompts = behaviour, memory =
recall.**