> ## Documentation Index > Fetch the complete documentation index at: https://docs.nvisy.com/llms.txt > Use this file to discover all available pages before exploring further. # Get detection intermediates > Returns the detection's enrichment intermediates — an image's OCR layout, an audio clip's transcript, or tokenized text — as an object with a `parts` list, each part carrying its path `id`, `modality`, and the extracted `artifact`, so a client can search the content and add entities the analysis missed. A detection whose analysis ran no enricher has no intermediates (404). ## OpenAPI ````yaml /api-reference/openapi.json get /workspaces/{workspaceId}/detections/{detectionId}/intermediates openapi: 3.1.0 info: title: Nvisy API summary: Document detection and redaction platform description: >- Nvisy provides intelligent detection of sensitive content in documents, redaction pipelines, and reviewer workflows. This API enables document upload, detection analysis, redaction, and collaborative review across your workspaces. termsOfService: https://nvisy.com/legal/terms-of-service contact: name: Nvisy Support url: https://nvisy.com email: hello@nvisy.com license: name: Apache-2.0 url: https://www.apache.org/licenses/LICENSE-2.0 version: 0.1.0 servers: [] security: [] tags: - name: Health description: Service health checks - name: Capabilities description: Read-only reference data describing this deployment - name: Accounts description: Account management and profile operations - name: Identities description: 'Account sign-in methods: password and linked providers' - name: Avatars description: Public avatar image serving - name: Authentication description: Login, signup, and token management - name: Workspaces description: Workspace creation and management - name: Activities description: Workspace activity log and export - name: Documents description: Document upload, download, and management - name: Analytics description: Aggregate metrics over a workspace's documents - name: Members description: Workspace member management - name: Invites description: Workspace invitation handling - name: API Tokens description: API token management - name: Connections description: External provider connections - name: Connection Syncs description: Connection synchronization runs - name: Providers description: Inference-provider configuration - name: Pipelines description: Redaction pipeline configuration - name: Detections description: Detection analysis and redaction - name: Redactions description: Redactions produced from a detection, with reviewer edits - name: Reviews description: Document reviews, the review queue, and their discussion - name: Policies description: Redaction policy configuration - name: Webhooks description: Webhook configuration - name: Notifications description: Account notification management paths: /workspaces/{workspaceId}/detections/{detectionId}/intermediates: get: tags: - Detections summary: Get detection intermediates description: >- Returns the detection's enrichment intermediates — an image's OCR layout, an audio clip's transcript, or tokenized text — as an object with a `parts` list, each part carrying its path `id`, `modality`, and the extracted `artifact`, so a client can search the content and add entities the analysis missed. A detection whose analysis ran no enricher has no intermediates (404). parameters: - in: path name: workspaceId description: Workspace identifier. required: true schema: description: Workspace identifier. type: string format: uuid style: simple - in: path name: detectionId description: Opaque identifier of the detection. required: true schema: $ref: '#/components/schemas/DetectionId' description: Opaque identifier of the detection. style: simple responses: '200': description: '' content: application/json: schema: $ref: '#/components/schemas/ArtifactSet' '401': description: >- The serialized shape of an HTTP error: the inert wire/OpenAPI-schema view that [`Error`] renders to at the response boundary. It carries no builder logic — [`Error`] is the type handlers construct and thread through `Result`, and it builds an `ErrorResponse` directly in its `IntoResponse` impl. `context` and `status` are not part of the JSON body (`context` is logged, `status` sets the HTTP status line). [`Error`]: crate::response::Error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '403': description: >- The serialized shape of an HTTP error: the inert wire/OpenAPI-schema view that [`Error`] renders to at the response boundary. It carries no builder logic — [`Error`] is the type handlers construct and thread through `Result`, and it builds an `ErrorResponse` directly in its `IntoResponse` impl. `context` and `status` are not part of the JSON body (`context` is logged, `status` sets the HTTP status line). [`Error`]: crate::response::Error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '404': description: >- The serialized shape of an HTTP error: the inert wire/OpenAPI-schema view that [`Error`] renders to at the response boundary. It carries no builder logic — [`Error`] is the type handlers construct and thread through `Result`, and it builds an `ErrorResponse` directly in its `IntoResponse` impl. `context` and `status` are not part of the JSON body (`context` is logged, `status` sets the HTTP status line). [`Error`]: crate::response::Error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' security: - BearerAuth: [] components: schemas: DetectionId: description: Opaque detection identifier (detection_). type: string pattern: >- ^detection_[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}$ ArtifactSet: type: object properties: parts: type: array items: discriminator: propertyName: modality oneOf: - type: object properties: artifact: $ref: '#/components/schemas/Tokens' id: type: array items: type: string minItems: 1 modality: type: string const: text required: - id - modality - artifact - type: object properties: artifact: $ref: '#/components/schemas/Layout' id: type: array items: type: string minItems: 1 modality: type: string const: image required: - id - modality - artifact - type: object properties: artifact: $ref: '#/components/schemas/Transcription' id: type: array items: type: string minItems: 1 modality: type: string const: audio required: - id - modality - artifact - type: object properties: artifact: $ref: '#/components/schemas/Tokens' id: type: array items: type: string minItems: 1 modality: type: string const: tabular required: - id - modality - artifact required: - parts ErrorResponse: description: >- The serialized shape of an HTTP error: the inert wire/OpenAPI-schema view that [`Error`] renders to at the response boundary. It carries no builder logic — [`Error`] is the type handlers construct and thread through `Result`, and it builds an `ErrorResponse` directly in its `IntoResponse` impl. `context` and `status` are not part of the JSON body (`context` is logged, `status` sets the HTTP status line). [`Error`]: crate::response::Error type: object properties: message: description: User-friendly error message safe for client display. type: string name: description: The error name/type identifier. type: string required: - name - message Tokens: description: >- Owning token sequence: the [`Text`](super::Text) modality's enrichment artifact, produced by a tokenizing enricher and read by a context enhancer. Tokens are sorted by `offset.start` (producers should emit them in order; consumer-side code assumes this). A context enhancer borrows the underlying slice via [`as_slice`](Tokens::as_slice) and walks it by count when scoring an entity's neighbourhood. Empty ([`Default`]) until a tokenizing enricher fills it, the enhancer then tokenizes on demand instead. type: array items: $ref: '#/components/schemas/Token' Layout: description: >- An image's recognized text, laid out in space. An ordered set of [`LayoutBlock`]s (the recognized text regions). The flat [`text`], the blocks joined, is what a recognizer inspects; [`resolve`] maps a byte range of that text back to the [`ImageLocation`] it occupies, using the blocks' (and their words') bounding boxes. Empty when the backend recognized nothing. [`text`]: Self::text [`resolve`]: Self::resolve type: object properties: blocks: description: Blocks in reading order. type: array items: $ref: '#/components/schemas/LayoutBlock' text: description: >- The blocks' text joined by [`BLOCK_SEPARATOR`], cached so recognition and byte-range resolution share one flat string. type: string required: - blocks - text Transcription: description: |- Timestamped transcript of an audio stream. An ordered set of [`TranscriptSegment`]s. The flat [`text`], the segments joined, is what a recognizer inspects; [`resolve`] maps a byte range of that text back to the [`TimeSpan`] it occupies, using the segments' (and their words') timings. Empty when the backend produced nothing (silence, or a no-op backend). [`text`]: Self::text [`resolve`]: Self::resolve type: object properties: segments: description: Segments in stream order. type: array items: $ref: '#/components/schemas/TranscriptSegment' text: description: |- The segments' text joined by [`SEGMENT_SEPARATOR`], cached so recognition and byte-range resolution share one flat string. Each segment's text begins at a known offset within it (see [`segment_offsets`]). [`segment_offsets`]: Self::segment_offsets type: string required: - segments - text Token: description: >- One token produced by an upstream tokenizer. `lemma` falls back to `text` when the producer has no lemmatizer, so callers that want lemma-aware matching can read `token.lemma` uniformly without checking which engine produced the artifact. type: object properties: is_punct: description: Producer-asserted punctuation flag. type: boolean is_stop: description: |- Producer-asserted stopword flag (e.g. "the", "a", "of" for English). Producers without a stopword list set this to `false`. type: boolean lemma: description: >- Lemma when the producer emitted one; otherwise == [`text`](Self::text). type: string offset: description: >- Byte range this token occupies in the source text. Use this to map back to substrings of the original input. allOf: - $ref: '#/components/schemas/Range_of_uint' text: description: Surface form as it appears in the source text. type: string required: - text - lemma - offset - is_stop - is_punct LayoutBlock: description: |- One recognized region of an image: its bounding box and text, optionally broken into per-word boxes. type: object properties: region: description: Bounding region of the block in image coordinates. allOf: - $ref: '#/components/schemas/ImageLocation' text: description: Recognized text for this block. type: string words: description: |- Per-word boxes within the block, when the backend emitted them. Empty otherwise; resolution then falls back to the block region. type: array default: [] items: $ref: '#/components/schemas/LayoutWord' required: - region - text TranscriptSegment: description: |- One segment of a [`Transcription`]: a span of audio and the text recognised within it, with optional diarization, language, confidence, and per-word timings. `speaker_id` is populated only by backends with diarization; `language` by backends that emit per-segment language detection; `confidence` when the backend reports one; `words` when it emits a word-level breakdown (which is what lets a sub-segment range resolve to a tighter span than the whole segment). type: object properties: confidence: $ref: '#/components/schemas/Confidence' description: Backend confidence in the segment, when reported. language: description: Detected language for this segment, when the backend reported one. type: string span: description: Time span the segment covers within the stream. allOf: - $ref: '#/components/schemas/TimeSpan' speaker_id: description: Diarization speaker label, when the backend assigned one. type: string text: description: Recognised text for this segment. type: string words: description: |- Per-word timings within the segment, when the backend emitted them. Empty otherwise; resolution then falls back to the segment span. type: array default: [] items: $ref: '#/components/schemas/TranscriptWord' required: - span - text Range_of_uint: type: object properties: end: type: integer format: uint minimum: 0 start: type: integer format: uint minimum: 0 required: - start - end ImageLocation: description: |- Region within image content. An axis-aligned [`BoundingBox`] in pixel coordinates locates the region; an optional [`Polygon`] captures a rotated or quadrilateral shape when the source produced one (OCR engines that emit 4-point polygons), and an optional page number addresses multi-page documents. type: object properties: bounding_box: description: Axis-aligned bounding box of the region, in pixel coordinates. allOf: - $ref: '#/components/schemas/BoundingBox' page: description: 1-based page number, for multi-page documents like PDFs. type: integer format: uint32 minimum: 0 polygon: $ref: '#/components/schemas/Polygon' description: |- Polygon vertices when the region is rotated or quadrilateral. Axis-aligned-only sources leave this unset. required: - bounding_box LayoutWord: description: One word within a [`LayoutBlock`], with its own bounding box. type: object properties: confidence: $ref: '#/components/schemas/Confidence' description: Per-word confidence, when reported. region: description: Bounding region of the word in image coordinates. allOf: - $ref: '#/components/schemas/ImageLocation' text: description: The word text, as it appears in the block text. type: string required: - region - text Confidence: description: |- Confidence score in the closed range `0.0..=1.0`. Carried by every provenance [`AuditEvent`] (the `before`/`after` of a recognition, fusion, or calibration) and by the effective confidence of an [`Entity`]. The newtype enforces the range at construction so no downstream code has to defend against values outside `[0, 1]`. Distinct from [`ConfidenceThreshold`] so the two cannot be confused at a call site: a score is *produced* by detection, a threshold is a *cutoff* configured to filter scores. Compare the two with [`ConfidenceThreshold::passes`]. [`AuditEvent`]: crate::entity::audit::AuditEvent [`Entity`]: crate::entity::Entity type: number format: float TimeSpan: description: |- Half-open `[start, end)` stream interval, measured in microseconds. The coordinate a time-addressed medium (audio, video) uses to locate a region: a transcribed segment, a redacted span. Microsecond precision is finer than both word-level speech timings and per-sample audio resolution, so a span never loses precision being carried as a `TimeSpan`; the endpoints are non-negative offsets by construction. Half-open like a byte range: `[start, end)`, so two intervals that merely touch (`a.end == b.start`) do not [`overlap`]. [`from_millis`] and [`as_millis`] bridge the millisecond-based APIs that surround it (audio durations, provider timings reported in ms). [`overlap`]: Self::overlaps [`from_millis`]: Self::from_millis [`as_millis`]: Self::start_millis type: object properties: end_us: description: |- Microseconds from the start of the stream where the interval ends (exclusive). type: integer format: uint64 minimum: 0 start_us: description: Microseconds from the start of the stream where the interval begins. type: integer format: uint64 minimum: 0 required: - start_us - end_us TranscriptWord: description: One word within a [`TranscriptSegment`], with its own time span. type: object properties: confidence: $ref: '#/components/schemas/Confidence' description: Per-word confidence, when reported. span: description: Time span the word covers within the stream. allOf: - $ref: '#/components/schemas/TimeSpan' text: description: The word text, as it appears in the segment text. type: string required: - span - text BoundingBox: description: |- Axis-aligned rectangle, given by its minimum and maximum corners. The location type for the image and document modalities: where a detected entity sits within a rendered page. [`min`] is the top-left corner and [`max`] the bottom-right under the usual screen convention (y grows downward), though the box itself is agnostic to coordinate orientation. [`min`]: Self::min [`max`]: Self::max type: object properties: max: description: Maximum corner (bottom-right, conventionally). allOf: - $ref: '#/components/schemas/Point' min: description: Minimum corner (top-left, conventionally). allOf: - $ref: '#/components/schemas/Point' required: - min - max Polygon: description: |- Closed polygon, given by its ordered vertices. A richer location than a [`BoundingBox`] for detections whose extent is not rectangular: rotated text, a region traced by a vision model, a signature. The boundary is implicitly closed, so the last vertex connects back to the first. [`BoundingBox`]: super::BoundingBox type: array items: $ref: '#/components/schemas/Point' Point: description: |- Point in a 2-D coordinate space. The coordinate basis is left to the consumer: pixel coordinates for a raster image, normalized `0.0..=1.0` coordinates for a resolution-independent region, or page units for a document. The model only requires the two scalars. type: object properties: x: description: Horizontal coordinate. type: number format: double 'y': description: Vertical coordinate. type: number format: double required: - x - 'y' ````