API docs by Redocly

Pulse API (0.4.0)

Download OpenAPI specification:

A stateless RESTful service that provides five core capabilities:

  • Batch generation of dense vector embeddings for up to 200 input strings.
  • Flexible cosine-similarity computations across strings.
  • Thematic clustering of open-ended text into concise, human-readable themes.
  • Sentiment classification of each input string.
  • Extraction of text elements matching specified themes from input strings.

Similarity

The similarity endpoint supports two scenarios:

  1. Self-similarity - Provide a single set of strings (set) and the service always returns a flattened upper-triangle similarity vector (excluding the diagonal). The response also includes the number of input strings (n) so the caller can easily re-shape the full matrix.

  2. Cross-similarity - Provide two distinct sets (set_a and set_b) and the service returns similarities only between items across the two sets (|A| x |B| matrix). The representation can be matrix (default) or flattened when the flatten=true query parameter is supplied.

Themes

The themes endpoint groups a collection of open-ended responses (e.g., survey comments, product reviews) into latent themes. Each theme now contains:

  • shortLabel - a concise (2-4 word) name for dashboards or charts.
  • label - a slightly longer descriptive title.
  • description - 1-2 sentences summarizing the common idea captured by the theme.
  • representatives - exactly two representative input strings for each theme.

Callers may optionally specify minThemes, maxThemes, and a free-text context string to steer clustering (e.g., "focus on UX issues").

Sentiment

The sentiment endpoint classifies each input string as positive, negative, neutral, or mixed and returns a confidence value ∈ [0, 1].

Versioning

The service maintains internally versioned models. When the optional version field is omitted, the latest production version is used. Supplying a version locks behaviour to that specific model version, enabling reproducible results even after future upgrades.

embeddings

Embeddings endpoint. Supports synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

  • In synchronous mode, processes up to 200 input strings and returns embeddings immediately (HTTP 200).
  • In asynchronous mode, accepts up to 2,000 input strings and returns a jobId (HTTP 202) to poll via /jobs.

Generate dense vector embeddings for input strings (synchronous or asynchronous).

Generates dense vector embeddings for input strings in a single batch. Supports synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

• Synchronous mode processes up to 200 input strings and returns embeddings immediately (HTTP 200). • Asynchronous mode accepts up to 2,000 input strings and returns a jobId (HTTP 202) to poll via the /jobs endpoint.

Authorizations:
BearerAuth
Request Body schema: application/json
required
inputs
required
Array of strings <= 2000 items

List of input strings. For synchronous (fast=true) mode, max 200; for asynchronous (fast=false or omitted) mode, max 2000.

fast
boolean

Flag indicating synchronous (true) or asynchronous (false) processing. Default false.

Responses

Response Schema: application/json
required
Array of objects (EmbeddingDocument) <= 200 items
requestId
required
string

Request samples

Content type
application/json
{
}

Response samples

Content type
application/json
{
}

similarity

Similarity endpoint. Provides self- and cross-similarity computations with sync (fast=true) and async (fast=false or omitted) modes.

  • In synchronous mode, self-similarity supports up to 100 strings; cross-similarity requires |set_a|×|set_b| ≤ 10,000 (HTTP 200).
  • In asynchronous mode, supports larger inputs (self up to 44,721 strings; cross with |set_a|×|set_b| ≤ 2,000,000,000) and returns a jobId (HTTP 202).

Compute cosine similarity between strings (self or cross).

Computes pairwise cosine similarity between input strings. Supports self-similarity (provide set) and cross-similarity (provide set_a and set_b), with synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

• In synchronous mode, self-similarity supports up to 100 input strings; cross-similarity requires |set_a|×|set_b| ≤ 10,000. Returns similarities immediately (HTTP 200). • In asynchronous mode, supports larger inputs (self up to 44,721 items; cross with |set_a|×|set_b| ≤ 2,000,000,000) and returns a jobId (HTTP 202) to poll via the /jobs endpoint.

Authorizations:
BearerAuth
Request Body schema: application/json
required
One of
set
required
Array of strings [ 2 .. 44721 ] items

Array of strings for self-similarity. For synchronous (fast=true), max 100; for asynchronous (fast=false or omitted), max 44721.

set_a
Array of strings non-empty

Array of strings for cross-similarity. For synchronous (fast=true), ensure |set_a|×|set_b| ≤ 10000; for asynchronous (fast=false or omitted), ensure |set_a|×|set_b| ≤ 2000000000.

set_b
Array of strings non-empty

Array of strings for cross-similarity. For synchronous (fast=true), ensure |set_a|×|set_b| ≤ 10000; for asynchronous (fast=false or omitted), ensure |set_a|×|set_b| ≤ 2000000000.

version
string
fast
boolean

Flag indicating synchronous (true) or asynchronous (false) processing. Default false.

flatten
boolean

For cross-similarity, flatten the matrix into a 1-D array. Ignored for self-similarity. Default false.

Responses

Response Schema: application/json
scenario
required
string
Enum: "self" "cross"
mode
required
string
Enum: "matrix" "flattened"
n
required
integer
flattened
required
Array of numbers <float> <= 2000000000 items [ items <float > ]
matrix
Array of numbers <= 2000000000 items [ items <float > <= 2000000000 items [ items <float > ] ]
requestId
required
string

Request samples

Content type
application/json
Example
{
}

Response samples

Content type
application/json
{
}

themes

Themes endpoint. Clusters text into themes with sync (fast=true) and async (fast=false or omitted) modes.

  • In synchronous mode, processes up to 200 input strings and returns themes immediately (HTTP 200).
  • In asynchronous mode, accepts up to 500 input strings and returns a jobId (HTTP 202).

Cluster open-ended text responses into thematic groups.

Groups input strings into latent themes using LLM-based clustering. Supports synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

  • In synchronous mode, processes up to 200 input strings and returns themes immediately (HTTP 200).
  • In asynchronous mode, accepts up to 500 input strings and returns a jobId (HTTP 202) to poll via the /jobs endpoint.

Each theme includes a shortLabel, label, description, and exactly two representative input strings. Optionally control theme count with minThemes, maxThemes, and steer focus via context.

Authorizations:
BearerAuth
Request Body schema: application/json
required
inputs
required
Array of strings [ 2 .. 500 ] items

List of input strings. For synchronous (fast=true) mode, max 200; for asynchronous (fast=false or omitted) mode, max 500.

minThemes
integer >= 1
maxThemes
integer <= 50
context
string
version
string
prune
integer [ 0 .. 50 ]
fast
boolean

Flag indicating synchronous (true) or asynchronous (false) processing. Default false.

Responses

Response Schema: application/json
required
Array of objects (Theme) <= 50 items
requestId
required
string

Request samples

Content type
application/json
Example
{
}

Response samples

Content type
application/json
{
}

sentiment

Sentiment endpoint. Classifies sentiment with sync (fast=true) and async (fast=false or omitted) modes.

  • In synchronous mode, processes up to 200 input strings (HTTP 200).
  • In asynchronous mode, accepts up to 10,000 input strings and returns a jobId (HTTP 202).

Classify sentiment of each input string.

Classifies the sentiment of each input string as positive, negative, neutral, or mixed, with confidence scores ∈ [0,1]. Supports synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

• In synchronous mode, processes up to 200 input strings and returns results immediately (HTTP 200). • In asynchronous mode, accepts up to 10000 input strings and returns a jobId (HTTP 202) to poll via the /jobs endpoint.

Optionally supply version for reproducible outputs.

Authorizations:
BearerAuth
Request Body schema: application/json
required
inputs
required
Array of strings [ 1 .. 10000 ] items

List of input strings. For synchronous (fast=true) mode, max 200; for asynchronous (fast=false or omitted) mode, max 10000.

version
string
fast
boolean

Flag indicating synchronous (true) or asynchronous (false) processing. Default false.

Responses

Response Schema: application/json
required
Array of objects (SentimentResult) <= 10000 items
requestId
required
string

Request samples

Content type
application/json
Example
{
}

Response samples

Content type
application/json
{
}

extractions

Extractions endpoint. Extracts elements matching themes with sync (fast=true) and async (fast=false or omitted) modes.

  • Both modes support up to 200 input strings and up to 50 themes.
  • Synchronous returns immediate results (HTTP 200); asynchronous returns a jobId (HTTP 202).

Extract elements matching themes from input strings.

Extracts substrings from inputs that match the provided themes. Supports synchronous (fast=true) and asynchronous (fast=false or omitted) modes.

• Both modes support up to 200 input strings and up to 50 themes. • Synchronous mode returns extraction results immediately (HTTP 200). • Asynchronous mode returns a jobId (HTTP 202) to poll via the /jobs endpoint.

Returns a 3-dimensional array where extractions[i][j] contains matching elements for input i and theme j.

Authorizations:
BearerAuth
Request Body schema: application/json
required
inputs
required
Array of strings [ 1 .. 200 ] items
themes
required
Array of strings [ 1 .. 50 ] items
version
string
fast
boolean

Flag indicating synchronous (true) or asynchronous (false) processing. Default false.

Responses

Response Schema: application/json
extractions
required
Array of strings <= 1000 items [ items <= 1000 items [ items <= 1000 items ] ]

3D array of extracted elements, shape [inputs.length][themes.length][K]

requestId
required
string

Request samples

Content type
application/json
{
}

Response samples

Content type
application/json
{
}

jobs

Jobs endpoint. Poll job status for asynchronous requests generated by other endpoints.

  • Retrieve status and result URL for completed jobs.

Retrieve status of an asynchronous job.

Retrieves the status of a previously submitted long-running job. Returns pending, completed, or failed. When completed, includes a resultUrl to download results.

Authorizations:
BearerAuth
query Parameters
jobId
required
string

Unique identifier for the job.

Responses

Response Schema: application/json
jobId
string
status
required
string
Enum: "pending" "completed" "failed"
resultUrl
string

Response samples

Content type
application/json
{
}