Generate Extraction Schema
Generate a JSON schema and return a product configuration request.
ParametersExpand Collapse
ExtractGenerateSchemaParams params
ExtractV2SchemaGenerateRequest extractV2SchemaGenerateRequest
Request schema for generating an extraction schema.
ReturnsExpand Collapse
class ConfigurationCreate:
Request body for creating a product configuration.
Parameters parameters
Product-specific configuration parameters.
class SplitV1Parameters:
Typed parameters for a split v1 product configuration.
List<SplitCategory> categories
Optional<SplittingStrategy> splittingStrategy
Strategy for splitting documents.
Optional<AllowUncategorized> allowUncategorized
Controls handling of pages that don’t match any category. ‘include’: pages can be grouped as ‘uncategorized’ and included in results. ‘forbid’: all pages must be assigned to a defined category. ‘omit’: pages can be classified as ‘uncategorized’ but are excluded from results.
class ExtractV2Parameters:
Typed parameters for an extract v2 product configuration.
Optional<String> parseConfigId
Saved parse configuration ID to control how the document is parsed before extraction
Optional<String> parseTier
Parse tier to use before extraction. Defaults to the extract tier if not specified.
Optional<String> targetPages
Comma-separated page numbers or ranges to process (1-based). Omit to process all pages.
class ClassifyV2Parameters:
class ParseV2Parameters:
Configuration for LlamaParse v2 document parsing.
Includes tier selection, processing options, output formatting, page targeting, and webhook delivery. Refer to the LlamaParse documentation for details on each field.
Version version
Version for the selected tier. Use latest, or pin one of that tier’s dated versions.
Current latest by tier:
fast:2025-12-11cost_effective:2026-06-05agentic:2026-06-04agentic_plus:2026-06-04
Full list: GET /api/v2/parse/versions.
Optional<AgenticOptions> agenticOptions
Options for AI-powered parsing tiers (cost_effective, agentic, agentic_plus).
These options customize how the AI processes and interprets document content. Only applicable when using non-fast tiers.
Optional<String> clientName
Identifier for the client/application making the request. Used for analytics and debugging. Example: ‘my-app-v2’
Optional<CropBox> cropBox
Crop boundaries to process only a portion of each page. Values are ratios 0-1 from page edges
Optional<Double> bottom
Bottom boundary as ratio (0-1). 0=top edge, 1=bottom edge. Content below this line is excluded
Optional<Double> left
Left boundary as ratio (0-1). 0=left edge, 1=right edge. Content left of this line is excluded
Optional<Boolean> disableCache
Bypass result caching and force re-parsing. Use when document content may have changed or you need fresh results
Optional<JsonValue> fastOptions
Options for fast tier parsing (rule-based, no AI).
Fast tier uses deterministic algorithms for text extraction without AI enhancement. It’s the fastest and most cost-effective option, best suited for simple documents with standard layouts. Currently has no configurable options but reserved for future expansion.
Optional<InputOptions> inputOptions
Format-specific options (HTML, PDF, spreadsheet, presentation). Applied based on detected input file type
Optional<Html> html
HTML/web page parsing options (applies to .html, .htm files)
Optional<Boolean> makeAllElementsVisible
Force all HTML elements to be visible by overriding CSS display/visibility properties. Useful for parsing pages with hidden content or collapsed sections
Optional<Presentation> presentation
Optional<Spreadsheet> spreadsheet
Spreadsheet parsing options (applies to .xlsx, .xls, .csv, .ods files)
Optional<Boolean> detectSubTablesInSheets
Detect and extract multiple tables within a single sheet. Useful when spreadsheets contain several data regions separated by blank rows/columns
Optional<OutputOptions> outputOptions
Output formatting options for markdown, text, and extracted images
Optional<List<String>> additionalOutputs
Optional additional output artifacts to save alongside the primary parse output. Each value opts in to generating and persisting one extra file; the empty list (default) saves none. The three accepted values are: ‘stripped_md’ — per-page markdown stripped of formatting (links, bold/italic, images, HTML), saved as JSON for full-text-search indexing; fetch via expand=stripped_markdown_content_metadata. ‘concatenated_stripped_txt’ — all stripped pages concatenated into a single plain-text file with \n\n---\n\n between pages, useful for feeding the document into search or embedding pipelines as one blob; fetch via expand=concatenated_stripped_markdown_content_metadata. ‘word_bbox’ — raw word-level bounding boxes (one JSON object per word, with page number and x/y/w/h coordinates) saved as JSONL, useful for highlighting or grounding extracted answers back to the source document; fetch via expand=raw_words_content_metadata.
Optional<Boolean> extractPrintedPageNumber
Extract the printed page number as it appears in the document (e.g., ‘Page 5 of 10’, ‘v’, ‘A-3’). Useful for referencing original page numbers
Optional<List<GranularBbox>> granularBboxes
Bounding-box granularity levels to compute for the parse. ‘word’ computes one bounding box per detected word; ‘line’ computes one per text line; ‘cell’ computes one per table cell. Multiple levels can be requested. Empty list (default) disables granular bboxes — only item-level layout boxes are returned on the result. When set, the computed boxes are not inlined on the result items; they are written to a separate grounded_items sidecar (JSONL, one row per page) and exposed as result_content_metadata.grounded_items (a presigned download URL) on the parse result. Each row matches the GroundedJsonItem shape.
Optional<List<ImagesToSave>> imagesToSave
Image categories to extract and save. Options: ‘screenshot’ (full page renders useful for visual QA), ‘embedded’ (images found within the document), ‘layout’ (cropped regions from layout detection like figures and diagrams). Empty list saves no images
Optional<Markdown> markdown
Markdown formatting options including table styles and link annotations
Optional<Boolean> annotateLinks
Add link annotations to markdown output in the format text. When false, only the link text is included
Optional<Boolean> inlineImages
Embed images directly in markdown as base64 data URIs instead of extracting them as separate files. Useful for self-contained markdown output
Optional<Tables> tables
Table formatting options including markdown vs HTML format and merging behavior
Optional<Boolean> compactMarkdownTables
Remove extra whitespace padding in markdown table cells for more compact output
Optional<String> markdownTableMultilineSeparator
Separator string for multiline cell content in markdown tables. Example: ‘<br>’ to preserve line breaks, ’ ’ to join with spaces
Optional<SpatialText> spatialText
Spatial text output options for preserving document layout structure
Optional<Boolean> doNotUnrollColumns
Keep multi-column layouts intact instead of linearizing columns into sequential text. Automatically enabled for non-fast tiers
Optional<PageRanges> pageRanges
Page selection: limit total pages or specify exact pages to process
Optional<Long> maxPages
Maximum number of pages to process. Pages are processed in order starting from page 1. If both max_pages and target_pages are set, target_pages takes precedence
Optional<String> targetPages
Comma-separated list of specific pages to process using 1-based indexing. Supports individual pages and ranges. Examples: ‘1,3,5’ (pages 1, 3, 5), ‘1-5’ (pages 1 through 5 inclusive), ‘1,3,5-8,10’ (pages 1, 3, 5-8, and 10). Pages are sorted and deduplicated automatically. Duplicate pages cause an error
Optional<ProcessingControl> processingControl
Job execution controls including timeouts and failure thresholds
Optional<JobFailureConditions> jobFailureConditions
Quality thresholds that determine when a job should fail vs complete with partial results
Optional<Double> allowedPageFailureRatio
Maximum ratio of pages allowed to fail before the job fails (0-1). Example: 0.1 means job fails if more than 10% of pages fail. Default is 0.05 (5%)
Optional<Boolean> failOnBuggyFont
Fail the job if a problematic font is detected that may cause incorrect text extraction. Buggy fonts can produce garbled or missing characters
Optional<Boolean> failOnImageExtractionError
Fail the entire job if any embedded image cannot be extracted. By default, image extraction errors are logged but don’t fail the job
Optional<ProcessingOptions> processingOptions
Document processing options including OCR, table extraction, and chart parsing
Optional<Boolean> aggressiveTableExtraction
Use aggressive heuristics to detect table boundaries, even without visible borders. Useful for documents with borderless or complex tables
Optional<List<AutoModeConfiguration>> autoModeConfiguration
Conditional processing rules that apply different parsing options based on page content, document structure, or filename patterns. Each entry defines trigger conditions and the parsing configuration to apply when triggered
ParsingConf parsingConf
Parsing configuration to apply when trigger conditions are met
Optional<String> customPrompt
Custom AI instructions for matched pages. Overrides the base custom_prompt
Optional<Boolean> fullPageImageInPage
Trigger if page contains a full-page image (scanned page detection)
Optional<CostOptimizer> costOptimizer
Cost optimizer configuration for reducing parsing costs on simpler pages.
When enabled, the parser analyzes each page and routes simpler pages to faster, cheaper processing while preserving quality for complex pages. Only works with ‘agentic’ or ‘agentic_plus’ tiers.
Optional<Boolean> disableHeuristics
Disable automatic heuristics including outlined table extraction and adaptive long table handling. Use when heuristics produce incorrect results
Optional<Ignore> ignore
Options for ignoring specific text types (diagonal, hidden, text in images)
Optional<Boolean> ignoreDiagonalText
Skip text rotated at an angle (not horizontal/vertical). Useful for ignoring watermarks or decorative angled text
Optional<OcrParameters> ocrParameters
Optional<List<WebhookConfiguration>> webhookConfigurations
Webhook endpoints for job status notifications. Multiple webhooks can be configured for different events or services
Optional<List<String>> webhookEvents
Events that trigger this webhook. Options: ‘parse.success’ (job completed), ‘parse.error’ (job failed), ‘parse.partial_success’ (some pages failed), ‘parse.pending’, ‘parse.running’, ‘parse.cancelled’. If not specified, webhook fires for all events
Optional<WebhookHeaders> webhookHeaders
Custom HTTP headers to include in webhook requests. Use for authentication tokens or custom routing. Example: {‘Authorization’: ‘Bearer xyz’}
class SpreadsheetV1:
Typed parameters for a spreadsheet v1 product configuration.
Optional<String> extractionRange
A1 notation of the range to extract a single region from. If None, the entire sheet is used.
Optional<Boolean> flattenHierarchicalTables
Return a flattened dataframe when a detected table is recognized as hierarchical.
Optional<Boolean> generateAdditionalMetadata
Whether to generate additional metadata (title, description) for each extracted region.
Optional<Boolean> includeHiddenCells
Whether to include hidden cells when extracting regions from the spreadsheet.
Optional<List<String>> sheetNames
The names of the sheets to extract regions from. If empty, all sheets will be processed.
Optional<String> specialization
Optional specialization mode for domain-specific extraction. Supported values: ‘financial-standard’, ‘financial-enhanced’, ‘financial-precise’. Default None uses the general-purpose pipeline.
Generate Extraction Schema
package com.llamacloud_prod.api.example;
import com.llamacloud_prod.api.client.LlamaCloudClient;
import com.llamacloud_prod.api.client.okhttp.LlamaCloudOkHttpClient;
import com.llamacloud_prod.api.models.configurations.ConfigurationCreate;
import com.llamacloud_prod.api.models.extract.ExtractGenerateSchemaParams;
import com.llamacloud_prod.api.models.extract.ExtractV2SchemaGenerateRequest;
public final class Main {
private Main() {}
public static void main(String[] args) {
LlamaCloudClient client = LlamaCloudOkHttpClient.fromEnv();
ExtractV2SchemaGenerateRequest params = ExtractV2SchemaGenerateRequest.builder().build();
ConfigurationCreate configurationCreate = client.extract().generateSchema(params);
}
}{
"name": "x",
"parameters": {
"categories": [
{
"name": "x",
"description": "x"
}
],
"product_type": "split_v1",
"splitting_strategy": {
"allow_uncategorized": "include"
}
}
}Returns Examples
{
"name": "x",
"parameters": {
"categories": [
{
"name": "x",
"description": "x"
}
],
"product_type": "split_v1",
"splitting_strategy": {
"allow_uncategorized": "include"
}
}
}