> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentium.in/llms.txt
> Use this file to discover all available pages before exploring further.

# Terminology

> The execution and ownership terms used throughout Agentium.

| Term | Meaning |
| - | - |
| Agent | A configured model/tool execution loop |
| Model provider | Adapter implementing model generation and/or streaming |
| Tool | Named function with a parameter schema, execution body, and optional controls |
| Toolkit | A related group of tools and their integration configuration |
| Skill | Packaged instructions and/or capabilities registered with an application |
| Ability | A reusable harness capability composed into a definition |
| Harness definition | Declarative composition of capabilities and configuration |
| Harness runtime | Runtime applying a definition, driver, grants, and aggregate controls |
| Driver | Adapter that performs work through a runtime's execution contract |
| Run | One invocation, potentially containing multiple model calls and tool roundtrips |
| Session | Conversation context associated with a session identity |
| User / tenant identity | Execution scope derived from host authentication in a hosted application |
| Grant | Host permission for a capability; a capability's existence is not permission |
| Execution policy | Host-owned decision about a validated tool call |
| Approval | A scoped decision to allow or deny a proposed action |
| Team | Multiple agents coordinated through a chosen mode |
| Workflow | Host-defined steps and control flow |
| Checkpoint | Captured state for inspection or restoration; not an undo of external effects |
| Queue job | A delivery/scheduling unit that may be retried |
| Durable task | Persisted work coordinated under explicit ownership and recovery rules |
| Durable action | Ledger-tracked operation with stable identity and reconciliation requirements |
| Reconciliation | Establishing an uncertain external outcome from authoritative evidence |
| Knowledge base | External content exposed through retrieval |
| Memory | State retained for later context, with a selected scope and lifetime |
| Observer | Subscriber/exporter that receives execution evidence |

For decisions, read [choose an approach](/learn/choose-an-approach). For how the pieces execute, read [understand a run](/learn/execution).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.