> ## Documentation Index > Fetch the complete documentation index at: https://docs.runpulse.com/llms.txt > Use this file to discover all available pages before exploring further. # Run Pipeline > Parse one document (`config.step`), then run up to 10 more steps on the parse once it completes (`config.children`). A pipeline always runs async. Read its status at `GET /pipeline/{pipeline_id}`, each step's status at `GET /pipeline/{pipeline_id}/result`, and a completed step's output at its own endpoint. Each step costs what its endpoint costs, and the pipeline adds nothing. The pipeline and its steps expire together, counted from the pipeline's end. ## OpenAPI ````yaml /api-reference/v2/openapi.yml post /pipeline openapi: 3.1.0 info: title: Pulse API description: >- Agentic document infrastructure for the enterprise. The Pulse API turns complex documents into AI-ready data, such as Markdown, structured JSON and bounding boxes. version: '2' servers: - url: https://api.runpulse.com description: Default Pulse API base URL security: [] tags: - name: Parse description: Parse a document's text, tables and figures. - name: Batch parse description: Parse many documents with one config. - name: Extract description: Extract structured data from a parsed document. - name: Split description: Find each topic's pages in a parsed document. - name: Tables description: Merge a parsed document's tables across pages. - name: Charts description: Read the charts in a parsed document. - name: Classify description: Sort a document into the categories you define. - name: Pipeline description: Run a parse and its follow-on steps as one job. - name: Usage description: Check the pages and credits your organization has used. - name: Webhook description: Get notified when your organization's jobs change status. - name: Organization settings description: Read and change your organization's name, address and default API version. paths: /pipeline: post: tags: - Pipeline summary: Run Pipeline description: >- Parse one document (`config.step`), then run up to 10 more steps on the parse once it completes (`config.children`). A pipeline always runs async. Read its status at `GET /pipeline/{pipeline_id}`, each step's status at `GET /pipeline/{pipeline_id}/result`, and a completed step's output at its own endpoint. Each step costs what its endpoint costs, and the pipeline adds nothing. The pipeline and its steps expire together, counted from the pipeline's end. operationId: runPipeline parameters: - name: Pulse-Version in: header required: false schema: type: string default: '2' title: Pulse-Version examples: default: value: '2' requestBody: required: true content: application/json: schema: description: A pipeline to run on one document. A pipeline always runs async. title: PipelineRequest oneOf: - title: Inline config additionalProperties: false type: object properties: file_url: description: >- The document's URL: a public `http(s)` URL, a public `s3://` or `gs://` object, or an upload URL Pulse issued. minLength: 1 title: File Url type: string config: $ref: '#/components/schemas/PipelineConfig' description: This request's options. required: - file_url - config - title: Saved config additionalProperties: false type: object properties: file_url: description: >- The document's URL: a public `http(s)` URL, a public `s3://` or `gs://` object, or an upload URL Pulse issued. minLength: 1 title: File Url type: string config_id: description: A saved config's id, sent in place of `config`. title: Config Id type: string required: - file_url - config_id multipart/form-data: schema: oneOf: - title: Inline config type: object properties: file: type: string format: binary config: $ref: '#/components/schemas/PipelineConfig' description: This request's options. required: - file - config - title: Saved config type: object properties: file: type: string format: binary config_id: description: A saved config's id, sent in place of `config`. title: Config Id type: string required: - file - config_id encoding: config: contentType: application/json responses: '202': description: >- The pipeline started. Check `GET /pipeline/{pipeline_id}` until it ends. content: application/json: schema: $ref: '#/components/schemas/PipelineResponse' '400': description: >- The body is invalid ([`REQUEST_INVALID`](https://docs.runpulse.com/errors/request-invalid)): not JSON or a multipart form, an unknown or missing field, `file` and `file_url` both or neither, `config` and `config_id` both or neither, or more than 10 children. Or `page_range` is malformed or reaches past the document's last page ([`PAGE_RANGE_INVALID`](https://docs.runpulse.com/errors/page-range-invalid)), or a URL is not allowed ([`FILE_URL_INVALID`](https://docs.runpulse.com/errors/file-url-invalid)). `Pulse-Version` names a version this API doesn't serve ([`UNSUPPORTED_VERSION`](https://docs.runpulse.com/errors/unsupported-version)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - REQUEST_INVALID - PAGE_RANGE_INVALID - FILE_URL_INVALID - UNSUPPORTED_VERSION '401': description: >- No valid API key was sent ([`CREDENTIALS_INVALID`](https://docs.runpulse.com/errors/credentials-invalid)), or the key is inactive ([`API_KEY_INACTIVE`](https://docs.runpulse.com/errors/api-key-inactive)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - CREDENTIALS_INVALID - API_KEY_INACTIVE '402': description: >- Your organization's plan cannot cover the run: it pays per request, which API version 2 does not take ([`PAYMENT_REQUIRED`](https://docs.runpulse.com/errors/payment-required)); its account is blocked by a failed payment ([`PAYMENT_FAILED`](https://docs.runpulse.com/errors/payment-failed)), a payment that needs authentication ([`PAYMENT_AUTHENTICATION_REQUIRED`](https://docs.runpulse.com/errors/payment-authentication-required)) or a past-due subscription ([`SUBSCRIPTION_PAST_DUE`](https://docs.runpulse.com/errors/subscription-past-due)); its free tier or plan's limit is reached ([`PLAN_LIMIT_REACHED`](https://docs.runpulse.com/errors/plan-limit-reached)); the trial expired ([`TRIAL_EXPIRED`](https://docs.runpulse.com/errors/trial-expired)), the credit limit is exceeded ([`CREDIT_LIMIT_EXCEEDED`](https://docs.runpulse.com/errors/credit-limit-exceeded)) or the subscription is canceled ([`SUBSCRIPTION_CANCELED`](https://docs.runpulse.com/errors/subscription-canceled)); or a free-tier request turned storage off while your organization saves parses ([`PLAN_UPGRADE_REQUIRED`](https://docs.runpulse.com/errors/plan-upgrade-required)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - PAYMENT_REQUIRED - PAYMENT_FAILED - PAYMENT_AUTHENTICATION_REQUIRED - SUBSCRIPTION_PAST_DUE - PLAN_LIMIT_REACHED - TRIAL_EXPIRED - CREDIT_LIMIT_EXCEEDED - SUBSCRIPTION_CANCELED - PLAN_UPGRADE_REQUIRED '403': description: >- The account is suspended ([`ACCOUNT_SUSPENDED`](https://docs.runpulse.com/errors/account-suspended)), or a `file_url` in Pulse's storage is outside your organization ([`PERMISSION_DENIED`](https://docs.runpulse.com/errors/permission-denied)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - ACCOUNT_SUSPENDED - PERMISSION_DENIED '404': description: >- `config_id` names no saved pipeline config ([`RESOURCE_NOT_FOUND`](https://docs.runpulse.com/errors/resource-not-found); `errors` names the field). No endpoint answers this path ([`ROUTE_NOT_FOUND`](https://docs.runpulse.com/errors/route-not-found)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - RESOURCE_NOT_FOUND - ROUTE_NOT_FOUND '413': description: >- The document is too large ([`FILE_TOO_LARGE`](https://docs.runpulse.com/errors/file-too-large)): an upload over 100 MB (host it and send `file_url` instead), or more than the parse takes; or a form's text part is over 10 MB ([`FIELD_TOO_LARGE`](https://docs.runpulse.com/errors/field-too-large)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - FILE_TOO_LARGE - FIELD_TOO_LARGE '415': description: >- The document's type is not supported ([`FILE_TYPE_UNSUPPORTED`](https://docs.runpulse.com/errors/file-type-unsupported)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - FILE_TYPE_UNSUPPORTED '422': description: >- The document cannot be used: a corrupt ([`FILE_CORRUPT`](https://docs.runpulse.com/errors/file-corrupt)) or empty ([`FILE_EMPTY`](https://docs.runpulse.com/errors/file-empty)) file, a download that failed ([`FILE_DOWNLOAD_FAILED`](https://docs.runpulse.com/errors/file-download-failed)), or a PDF whose active content could not be removed ([`FILE_UNSAFE`](https://docs.runpulse.com/errors/file-unsafe)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - FILE_CORRUPT - FILE_EMPTY - FILE_DOWNLOAD_FAILED - FILE_UNSAFE '500': description: >- The run failed on Pulse's side ([`SERVER_ERROR`](https://docs.runpulse.com/errors/server-error)), its parse could not be started ([`PROCESSING_FAILED`](https://docs.runpulse.com/errors/processing-failed)). [`JOB_UNREADABLE`](https://docs.runpulse.com/errors/job-unreadable): Pulse can't read this job right now; don't retry, contact support. content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - SERVER_ERROR - PROCESSING_FAILED - JOB_UNREADABLE '503': description: >- A service was briefly unavailable ([`SERVICE_UNAVAILABLE`](https://docs.runpulse.com/errors/service-unavailable)): reading your API key or organization's retention settings, starting the pipeline or its parse, or extraction. Retry. Or processing is at capacity ([`CAPACITY_EXCEEDED`](https://docs.runpulse.com/errors/capacity-exceeded)), so the pipeline was canceled; retry after the seconds in `Retry-After`. On a self-hosted install that hasn't been activated yet, [`LICENSE_NOT_ACTIVATED`](https://docs.runpulse.com/errors/license-not-activated). On a self-hosted install whose license is invalid or expired, [`LICENSE_INVALID`](https://docs.runpulse.com/errors/license-invalid). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string Retry-After: description: Seconds to wait before retrying. schema: type: string x-pulse-errors: - SERVICE_UNAVAILABLE - CAPACITY_EXCEEDED - LICENSE_NOT_ACTIVATED - LICENSE_INVALID '504': description: >- Downloading `file_url` timed out ([`FILE_DOWNLOAD_TIMEOUT`](https://docs.runpulse.com/errors/file-download-timeout)). content: application/problem+json: schema: $ref: '#/components/schemas/Problem' application/json: schema: $ref: '#/components/schemas/Problem' headers: X-Request-Id: description: The request's id; the problem's `request_id`. schema: type: string x-pulse-errors: - FILE_DOWNLOAD_TIMEOUT security: - ApiKey: [] components: schemas: PipelineConfig: additionalProperties: false description: 'The pipeline''s steps: a parse, and the steps that read it.' properties: step: $ref: '#/components/schemas/ParseStep' description: >- The parse of the document. The pipeline ends the way the parse does: `completed` if it completes, otherwise `failed` or `canceled`, whatever the other steps do. children: description: >- The steps that read the parse, started once it completes. A failed step doesn't stop the others or fail the pipeline. If the parse doesn't complete, every step is `skipped`. items: discriminator: mapping: charts: '#/components/schemas/ChartsStep' extract: '#/components/schemas/ExtractStep' split: '#/components/schemas/SplitStep' tables: '#/components/schemas/TablesStep' propertyName: type oneOf: - $ref: '#/components/schemas/ExtractStep' - $ref: '#/components/schemas/SplitStep' - $ref: '#/components/schemas/TablesStep' - $ref: '#/components/schemas/ChartsStep' maxItems: 10 title: Children type: array required: - step title: PipelineConfig { step, children } type: object PipelineResponse: properties: pipeline_id: type: string title: Pipeline Id description: The pipeline's id, `pipeline-`. created_at: type: string title: Created At description: When the job was created. additionalProperties: false type: object required: - pipeline_id - created_at title: PipelineResponse description: >- A pipeline that has started. Check `GET /pipeline/{pipeline_id}` until it ends. Problem: additionalProperties: false description: >- Why the request or the job failed. Branch on `code`; quote `request_id` to support. properties: type: description: 'This code''s page: what it means and what to do.' title: Type type: string title: description: A short summary of the code, the same for every occurrence. title: Title type: string status: description: The HTTP status the code answers with. title: Status type: integer detail: description: What happened this time. title: Detail type: string instance: description: 'This occurrence: the request''s URN, `urn:pulse:request:`.' title: Instance type: string code: description: The stable error code, e.g. `REQUEST_INVALID`. Branch on it. title: Code type: string request_id: description: >- The request that failed, as `X-Request-Id` named it; for a failed job, the request that created it. Quote it to support. title: Request Id type: string retryable: description: Whether the same request may succeed if sent again. title: Retryable type: boolean retry_after: description: Seconds to wait before retrying, as `Retry-After` says. title: Retry After type: integer errors: description: Each refused part of the request. items: $ref: '#/components/schemas/FieldIssue' title: Errors type: array resource: description: 'What wasn''t found: `job`, `batch`, `webhook`, `config`, …' title: Resource type: string reason: description: The cause within the code. title: Reason type: string parse_id: description: The parse the request recorded. title: Parse Id type: string extract_id: description: The extraction it recorded. title: Extract Id type: string split_id: description: The split it recorded. title: Split Id type: string classify_id: description: The classification it recorded. title: Classify Id type: string tables_id: description: The tables job it recorded. title: Tables Id type: string charts_id: description: The charts job it recorded. title: Charts Id type: string pipeline_id: description: The pipeline it recorded. title: Pipeline Id type: string batch_id: description: The batch it recorded. title: Batch Id type: string filename: description: The file refused. title: Filename type: string declared_type: description: The type its name declared. title: Declared Type type: string detected: description: The type its contents are. title: Detected type: string supported_types: description: The types accepted. items: type: string title: Supported Types type: array supported: description: The values accepted. items: type: string title: Supported type: array max_size_mb: anyOf: - type: integer - type: number description: The largest size accepted. title: Max Size Mb file_size_mb: anyOf: - type: integer - type: number description: The file's size. title: File Size Mb plan_name: description: Your organization's plan. title: Plan Name type: string pages_used: anyOf: - type: integer - type: number description: Pages used this period. title: Pages Used page_allowance: anyOf: - type: integer - type: number description: Pages the plan allows. title: Page Allowance upgrade_url: description: Where to change your plan. title: Upgrade Url type: string action_required: description: What to do to continue. title: Action Required type: string billing_info: additionalProperties: $ref: '#/components/schemas/JsonValue' description: Your organization's billing state, as the gate read it. title: Billing Info type: object required: - type - title - status - code - retryable title: Problem type: object ParseStep: additionalProperties: false description: 'The pipeline''s first step: a parse of the document you send.' properties: type: const: parse description: Always `parse`. The first step is always a parse of the document. title: Type type: string config: $ref: '#/components/schemas/ParseConfig' description: >- The parse's options, as `POST /parse` takes them in `config`, sent inline only. The step is checked and billed as `POST /parse` is, at 1 credit per page. required: - type - config title: ParseStep { type, config } type: object ChartsStep: additionalProperties: false description: A step that reads the charts in the pipeline's parse. properties: type: const: charts description: Always `charts`. The step runs `POST /charts` on the parse. title: Type type: string config: $ref: '#/components/schemas/ChartsConfig' description: >- The step's options, as `POST /charts` takes them in `config`, sent inline only. required: - type - config title: ChartsStep type: object ExtractStep: additionalProperties: false description: >- A step that extracts the fields of a JSON Schema from the pipeline's parse. properties: type: const: extract description: Always `extract`. The step runs `POST /extract` on the parse. title: Type type: string config: $ref: '#/components/schemas/ExtractConfig' description: >- The step's options, as `POST /extract` takes them in `config`, sent inline only. required: - type - config title: ExtractStep type: object SplitStep: additionalProperties: false description: >- A step that finds which of the parse's pages belong to each topic you name. properties: type: const: split description: Always `split`. The step runs `POST /split` on the parse. title: Type type: string config: $ref: '#/components/schemas/SplitConfig' description: >- The step's options, as `POST /split` takes them in `config`, sent inline only. required: - type - config title: SplitStep type: object TablesStep: additionalProperties: false description: A step that reads the tables in the pipeline's parse. properties: type: const: tables description: Always `tables`. The step runs `POST /tables` on the parse. title: Type type: string config: $ref: '#/components/schemas/TableConfig' description: >- The step's options, as `POST /tables` takes them in `config`, sent inline only. required: - type - config title: TablesStep type: object FieldIssue: additionalProperties: false description: One refused part of the request. properties: field: description: >- The part as sent: a body field's path (`config.enhancements.detect[1]`), or `query.` / `header.`. title: Field type: string pointer: description: >- A body field as a JSON Pointer (RFC 6901), e.g. `#/config/enhancements/detect/1`. Absent for a query parameter or header. title: Pointer type: string issue: description: What is wrong with it. title: Issue type: string required: - field - issue title: FieldIssue type: object JsonValue: {} ParseConfig: additionalProperties: false description: The options for a parse. properties: model: description: The model to parse with. enum: - ultra-2 title: Model type: string storage: $ref: '#/components/schemas/Storage' description: Whether the result is saved, and where. page_range: description: >- The pages to parse, listed as numbers and ranges from 1, such as `1-3, 6, 8-end` (`end` is the last page). Defaults to every page. A spreadsheet's pages are its sheets, hidden ones included. pattern: ^ *[0-9]+ *(- *([0-9]+|end) *)?(, *[0-9]+ *(- *([0-9]+|end) *)?)*$ title: Page Range type: string prompt: description: Extra instructions for reading the document, added to Pulse's own. minLength: 1 title: Prompt type: string extensions: $ref: '#/components/schemas/Extensions' default: additional_formats: [] footnotes: false document_metadata: false word_level_bounding_boxes: false chunking: strategies: [] description: Extra outputs to add to the result. enhancements: $ref: '#/components/schemas/Enhancements' default: detect: [] figures: describe: [] include: false prompt: null refine: modes: [] spreadsheet: cell_formatting: false include_hidden: [] raw_values: false trim_empty: [] required: - model title: ParseConfig { model, storage, page_range, 3 more } type: object ChartsConfig: additionalProperties: false description: How to read the parse's charts and which files to build from them. properties: page_range: description: >- The pages to read charts from, listed as numbers and ranges from 1, such as `1-3, 6, 8-end` (`end` is the last page). Defaults to every page. Pages are numbered as in the parse's result. pattern: ^ *[0-9]+ *(- *([0-9]+|end) *)?(, *[0-9]+ *(- *([0-9]+|end) *)?)*$ title: Page Range type: string data_points: default: 20 description: >- How many points to keep per series for a chart read from pixels. A spreadsheet chart returns every point the workbook holds. maximum: 500 minimum: 2 title: Data Points type: integer export_formats: anyOf: - const: all type: string - items: enum: - excel - csv - las type: string type: array default: - excel description: >- `excel` builds one workbook of every chart, `csv` one file of every chart, and `las` one file per well-log chart. `all` builds all three. title: Export Formats enhancements: $ref: '#/components/schemas/ChartsEnhancements' default: zoom: false description: Extra reading steps. title: ChartsConfig { page_range, data_points, export_formats, 1 more } type: object ExtractConfig: additionalProperties: false description: Options for extracting structured data from a completed parse. properties: page_range: description: >- The pages to read, listed as numbers and ranges from 1, such as `1-3, 6, 8-end` (`end` is the last page). Defaults to every page. Pages are numbered as in the parse's result. pattern: ^ *[0-9]+ *(- *([0-9]+|end) *)?(, *[0-9]+ *(- *([0-9]+|end) *)?)*$ title: Page Range type: string prompt: description: Extra instructions for filling in `input_schema`. minLength: 1 title: Prompt type: string input_schema: additionalProperties: $ref: '#/components/schemas/JsonValue' description: >- The JSON Schema (Draft 7) to fill in, with an object root, at least one property and at most 5,000 properties in all. title: Input Schema type: object effort: default: low description: '`high` reads the parse more carefully, at an extra charge per page.' enum: - low - high title: Effort type: string confidence: default: standard description: >- `agentic` gives more reliable confidence scores, at an extra charge per page. It must be enabled for your organization. enum: - standard - agentic title: Confidence type: string format: anyOf: - const: all type: string - items: enum: - enriched - raw type: string type: array default: - enriched description: >- `enriched` returns values with citations and confidence, `raw` returns the values alone, and `all` returns both. title: Format required: - input_schema title: ExtractConfig { page_range, prompt, input_schema, 3 more } type: object SplitConfig: additionalProperties: false description: 'The options for a split: the topics to find and how carefully to read.' properties: topics: description: >- The topics to find, each named once. The result lists them in this order. items: $ref: '#/components/schemas/Topic' maxItems: 200 minItems: 1 title: Topics type: array effort: default: low description: >- How carefully the document is read. `low` costs 1 credit per page, and `high` costs 6. enum: - low - high title: Effort type: string required: - topics title: SplitConfig { topics, effort } type: object TableConfig: additionalProperties: false description: Options for merging a parse's tables. properties: format: anyOf: - const: all type: string - items: enum: - html - json - md type: string type: array default: - html description: >- `html` returns the table's HTML, `json` its headers and rows as text, and `md` a Markdown table. `all` returns all three. title: Format page_range: description: >- The parse result's pages to take tables from, listed as numbers and ranges from 1, such as `1-3, 6, 8-end` (`end` is the last page). Defaults to every page. Tables merge only within these pages. pattern: ^ *[0-9]+ *(- *([0-9]+|end) *)?(, *[0-9]+ *(- *([0-9]+|end) *)?)*$ title: Page Range type: string title: TableConfig { format, page_range } type: object Storage: additionalProperties: false description: >- Whether the result is saved to your organization's Pulse dashboard, and in which folder. properties: save: description: >- `true` saves the result, and `false` keeps it only until `expires_at`, 24 hours by default. If omitted, it's saved unless your organization turned saving off, which also overrides `true`. Free plans can't send `false`. title: Save type: boolean folder_id: description: >- The dashboard folder to save the result in. Sending it saves the result, so it's refused with `save: false`. Another organization's folder is ignored. minLength: 1 title: Folder Id type: string title: Storage { save, folder_id } type: object Extensions: additionalProperties: false description: >- Extra outputs to add to the result, each under the result's `extensions`. properties: additional_formats: default: [] description: >- Other formats to return the document in, under `extensions.additional_formats`. Choose from `html` (the Markdown as HTML), `xml` (the elements, by page) and `tagged_md` (the Markdown with element ids). items: enum: - html - xml - tagged_md type: string title: Additional Formats type: array footnotes: default: false description: >- Set to `true` to find the document's footnotes and link each one to the elements that cite it, in `extensions.footnotes`. title: Footnotes type: boolean document_metadata: default: false description: >- Set to `true` to return the metadata the file carries, such as its title, authors or a PDF's form fields, in `extensions.document_metadata`. title: Document Metadata type: boolean word_level_bounding_boxes: default: false description: >- Set to `true` to return every word with its outline, as `Word` elements in `extensions.word_level_bounding_boxes`. This option must be enabled for your organization. title: Word Level Bounding Boxes type: boolean chunking: $ref: '#/components/schemas/Chunking' default: strategies: [] title: Extensions { additional_formats, footnotes, document_metadata, 2 more } type: object Enhancements: additionalProperties: false description: Optional passes that add to the parse or correct it. properties: detect: default: [] description: >- Send `["selections"]` to detect checkboxes, radio buttons and similar controls, and mark each ticked one `selected: true`. items: const: selections type: string title: Detect type: array figures: $ref: '#/components/schemas/Figures' default: describe: [] include: false prompt: null description: Descriptions and images of the document's figures. refine: $ref: '#/components/schemas/Refine' default: modes: [] description: Passes that correct the parse. spreadsheet: $ref: '#/components/schemas/Spreadsheet' default: include_hidden: [] trim_empty: [] raw_values: false cell_formatting: false description: Options for spreadsheets. title: Enhancements { detect, figures, refine, 1 more } type: object ChartsEnhancements: additionalProperties: false description: Extra reading steps for charts in PDFs and images. properties: zoom: default: false description: >- Set to `true` to re-read small or dense regions of each chart up close. Values are more accurate, but the job is slower. Ignored for spreadsheet charts. title: Zoom type: boolean title: ChartsEnhancements type: object Topic: additionalProperties: false description: >- A topic to find in the document, named as you want it back in the result. properties: name: description: >- The topic's name in the result, such as `Balance sheet`. It needs at least one letter or digit. Names that differ only in case, punctuation or spacing count as the same topic and are refused. minLength: 1 title: Name type: string description: description: >- What belongs to this topic, such as `Statement of operations and its notes`. Without it, the name alone decides which pages match. minLength: 1 title: Description type: string required: - name title: Topic { name, description } type: object Chunking: additionalProperties: false properties: strategies: default: [] items: $ref: '#/components/schemas/SemanticStrategy' title: Strategies type: array title: Chunking type: object Figures: additionalProperties: false description: >- What to return for the document's figures: its charts, diagrams, photos and other pictures. properties: describe: default: [] description: >- Send `["all"]` to describe every figure in its `description`. A figure that can't be described gets none, with a `FIGURE_DESCRIPTIONS_SKIPPED` warning on the result. items: const: all type: string title: Describe type: array include: default: false description: >- Set to `true` to return each figure's image, base64-encoded, in its `figure_base64`. title: Include type: boolean prompt: description: >- Extra instructions for the figure descriptions. Used only with `describe`. minLength: 1 title: Prompt type: string title: Figures { describe, include, prompt } type: object Refine: additionalProperties: false description: Passes that check the parse against the page images and correct it. properties: modes: default: [] description: >- The passes to run, each mode at most once. A page they can't correct is returned as parsed, with a `REFINE_SKIPPED` warning on the result. items: $ref: '#/components/schemas/RefineMode' title: Modes type: array title: Refine type: object Spreadsheet: additionalProperties: false description: Options for spreadsheets. Other documents ignore them. properties: include_hidden: default: [] description: >- The hidden `rows`, `columns` or `sheets` to read. By default, hidden parts are skipped. items: enum: - rows - columns - sheets type: string title: Include Hidden type: array trim_empty: default: [] description: >- Send `rows`, `columns` or both to drop empty but formatted ones past a sheet's last value. Some exported workbooks carry thousands of them. Kept cells keep their addresses. Applies to `.xlsx` and `.xlsm` files only. items: enum: - rows - columns type: string title: Trim Empty type: array raw_values: default: false description: >- Set to `true` to add each cell's underlying value as its `raw_value`, such as `1201.67` where `content` shows `$1,202`. title: Raw Values type: boolean cell_formatting: default: false description: >- Set to `true` to add each cell's formatting (font, fill, borders and alignment) as its `style`. title: Cell Formatting type: boolean title: Spreadsheet { include_hidden, trim_empty, raw_values, 1 more } type: object SemanticStrategy: additionalProperties: false description: >- Chunks cut where the meaning of neighboring sentences drifts apart. A chunk longer than `max_characters` is cut again. Without `max_characters`, chunk length has no cap. properties: strategy: const: semantic title: Strategy type: string max_characters: maximum: 100000 minimum: 20 title: Max Characters type: integer required: - strategy title: SemanticStrategy { strategy, max_characters } type: object RefineMode: additionalProperties: false description: >- One pass that checks the parsed content against the page images and corrects it. properties: mode: description: >- `tables` fixes cells and structure. `formatting` fixes bold, headings, lists and math, not words. `text` fixes misread words and numbers, not formatting. `layout` fixes a figure or table read as the wrong kind. enum: - tables - formatting - text - layout title: Mode type: string prompt: description: >- Extra instructions for this pass. A prompt sent with `layout` is refused (`REQUEST_INVALID`). minLength: 1 title: Prompt type: string required: - mode title: RefineMode { mode, prompt } type: object securitySchemes: ApiKey: type: apiKey in: header name: x-api-key ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.