Skip to content

Concepts

Eight words you will meet in the tutorial. Read this once, or come back when a term is unfamiliar.

flowchart LR
    P[Project] --> E[Entry]
    E --> R[Recorded answer]
    C[Call from your app] --> M{Match a tier?}
    M -- yes --> R
    M -- no, record mode --> K[Parked call]
    K -- an agent answers --> E
Project
A namespace for recordings, usually one per app. Each project has its own entries, aliases and mode. See Give each app its own project.
Entry
One stored prompt with its answer (or answers). Calls are matched against entries. You can edit, pin and template them: Edit and protect a recording.
Mode
record answers unknown prompts by holding them for an agent. replay only serves what is stored and returns an error for anything else. Set it per project.
Parked call
A call semula has no answer for, in record mode. "Parked" means the HTTP request is held open while it waits. The same call shows up as pending in lists and the dashboard. An agent or a person answers it, and the app's request then completes.
Miss
A call that matched nothing. In record mode it parks; in replay mode it returns an error (Errors).
Tier
One way of matching a call to an entry, tried cheapest first: exact, normalised, alias, words, meaning. A weak match is never served silently. It goes to review instead. See Handle near misses.
Alias
A human-approved link saying "this prompt may be answered by that entry".
Cassette
A project exported to a directory of JSON files, so recordings can live in git and be imported in CI. See Keep recordings in git.

Which word for what

The docs use one word per idea.

Idea Word Not
A call semula has no answer for and is holding open parked call held, waiting, stuck
The list of parked calls pending (semula pending) queue, backlog
A stored prompt and its answer entry recording (the act of saving), cassette (the exported files)
What a person or agent supplies, or semula replays answer response, reply

The reasoning behind these is in Why calls park and How matching works.

Advanced terms (answerer hook, record-through, claim and lease) are explained where they are used, in record through a real model, let a command answer each call and manage a busy queue.