Glossary

One Engine deployment, with one base URL, a major API version, and a set of capabilities.

A named tenant isolation boundary containing collections. Its name is unique in the cluster.

An application customer or data owner isolated from other customers. A namespace provides that boundary.

A named set of files that Search and Answer address together.

A content resource in a collection with an ID, logical path, source, metadata, and processing status.

A case-sensitive UTF-8 file name inside a collection, beginning with a slash.

The beginning of a path. A prefix ending with a slash selects a directory-like set of paths.

The content input for a file: inline text, an upload, or a supported storage URI.

A JSON object of application facts attached to a file. Metadata fields can be used in retrieval filters.

An immutable public identifier, such as ns_, col_, fil_, key_, job_, rsr_, evt_, or acr_ with a ULID suffix.

An operation that returns ranked files or passages with their spans.

An operation that returns a natural-language answer grounded in one collection, with optional sources.

An Answer that runs on its own. Start it, follow its events or read it, and cancel it if needed.

A file ID and a place in that file: pages from 1, chars or seconds from 0, from start to end, or the whole file. It holds no content.

An operation that returns the content of up to 50 spans in the parts that include names: text (the text, or transcript lines), raw (the pages, the image, or a clip), or file (the whole file).

An answer marker such as [[SRC:0001]] that names the source whose ref is SRC:0001.

The collection’s keyword, semantic, and graph flags.

An opaque identifier for indexed content used by a retrieval request.

A typed condition that selects application data. A filter does not authorize a reader.

An opaque continuation token returned by a list or Search response.

A secret credential that authenticates a program. Engine key secrets begin with gek_.

The cluster, namespace, or named collections that a key can reach.

A permitted operation category: read, write, search, or query.

An exact user or group string supplied by the trusted application, such as user:u_8812.

A list of readers allowed to retrieve content at a folder prefix or file path.

The single rule that decides whether a reader can retrieve a file.

A folder rule that governs all descendants and overrides rules below it.

The public record of an operation that continues after the request response.

A status-change notification for a job, file, collection, or research run.

An HTTPS receiver that a key registers with a scope. Engine POSTs the signed events in that scope to it.

The 30-day record of the events that one webhook got, with every attempt. Replay queues its events again.

A caller-chosen request header that identifies one logical write for safe replay.