> ## Documentation Index > Fetch the complete documentation index at: https://docs.gp.scale.com/llms.txt > Use this file to discover all available pages before exploring further. # Create Extract Job > Create a new extract job. This endpoint creates an extraction job that runs **single-pass** LLM extraction from a parse result or vector store search context. For agentic (decompose / ReAct) extraction, use ``POST …/research`` instead; requests with ``use_agentic=True`` are rejected. Args: request: Extract job request containing source_id and parameters Returns: JobEntity: The created job entity Raises: HTTPException: If source not found or workflow startup fails ## OpenAPI ````yaml https://dex.sgp.scale.com/openapi.json post /v1/projects/{project_id}/extract openapi: 3.1.0 info: title: Document Understanding API description: API for uploading and processing documents version: 0.4.5 servers: [] security: - ApiKey: [] AccountId: [] tags: - name: Projects description: Operations related to project creation and management - name: Files description: Operations related to file upload and access - name: Parse description: Operations related to starting parse jobs and accessing their results - name: Vector Stores description: Operations related to vector store creation and management - name: Extract description: Operations related to starting extract jobs and accessing their results - name: Research description: Dex Research agent kickoff and results. - name: Jobs description: Operations related to monitoring jobs and their status paths: /v1/projects/{project_id}/extract: post: tags: - Extract summary: Create Extract Job description: >- Create a new extract job. This endpoint creates an extraction job that runs **single-pass** LLM extraction from a parse result or vector store search context. For agentic (decompose / ReAct) extraction, use ``POST …/research`` instead; requests with ``use_agentic=True`` are rejected. Args: request: Extract job request containing source_id and parameters Returns: JobEntity: The created job entity Raises: HTTPException: If source not found or workflow startup fails operationId: create_extract_job_v1_projects__project_id__extract_post parameters: - name: project_id in: path required: true schema: type: string title: Project Id requestBody: required: true content: application/json: schema: anyOf: - $ref: '#/components/schemas/ExtractFromParseResultJobRequest' - $ref: '#/components/schemas/ExtractFromVectorStoreJobRequest' title: Request responses: '202': description: Successful Response content: application/json: schema: $ref: '#/components/schemas/JobEntity' '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' components: schemas: ExtractFromParseResultJobRequest: properties: parameters: $ref: '#/components/schemas/ExtractionParameters' description: Extraction parameters including schema and prompts source_id: type: string title: Source Id description: The ID of the resource to extract from (file_... or job_...) type: object required: - parameters - source_id title: ExtractFromParseResultJobRequest description: Request model for creating an extract job from a parse result. example: parameters: extraction_schema: properties: invoice_id: description: The unique invoice identifier. type: string total_amount: description: The final amount due. type: number required: - invoice_id - total_amount type: object generate_citations: true model: openai/gpt-4 system_prompt: >- You are an expert financial analyst. Extract data with high precision. user_prompt: >- The invoice total can be found near the bottom right. Pay close attention to the 'Total Due' label. source_id: job_parse_xyz... ExtractFromVectorStoreJobRequest: properties: parameters: $ref: '#/components/schemas/ExtractionParameters' description: Extraction parameters including schema and prompts search_config: $ref: '#/components/schemas/SearchConfig' description: The search config to use for the extract job type: object required: - parameters - search_config title: ExtractFromVectorStoreJobRequest description: Request model for creating an extract job from a vector store. example: parameters: extraction_schema: properties: invoice_id: description: The unique invoice identifier. type: string total_amount: description: The final amount due. type: number required: - invoice_id - total_amount type: object generate_citations: true model: openai/gpt-4 system_prompt: >- You are an expert financial analyst. Extract data with high precision. user_prompt: >- The invoice total can be found near the bottom right. Pay close attention to the 'Total Due' label. search_config: filters: file_id: file_xyz... query_type: hybrid rerank_config: model: rerank-model-v1 top_n: 5 top_k: 10 vector_store_id: vector_store_xyz... JobEntity: properties: id: type: string title: Id description: ID of the entity project_id: type: string title: Project Id description: ID of the project object: type: string const: job title: Object default: job operation: $ref: '#/components/schemas/JobOperationType' description: Operation type (e.g., 'parse') status: $ref: '#/components/schemas/JobStatus' description: Current job status source_id: anyOf: - type: string - type: 'null' title: Source Id description: Source document/file ID correlation_id: anyOf: - type: string - type: 'null' title: Correlation Id description: Request correlation ID for tracing created_at: type: string format: date-time title: Created At description: When the job was created started_at: anyOf: - type: string format: date-time - type: 'null' title: Started At description: When the job started processing completed_at: anyOf: - type: string format: date-time - type: 'null' title: Completed At description: When the job completed result: anyOf: - additionalProperties: true type: object - type: 'null' title: Result description: Job result payload when completed progress: anyOf: - $ref: '#/components/schemas/BatchParseProgress' - type: 'null' description: Live progress payload (used by batch jobs) error: anyOf: - type: string - type: 'null' title: Error description: Error message if job failed history: anyOf: - items: $ref: '#/components/schemas/JobHistoryEvent' type: array - type: 'null' title: History description: Timeline of job execution events type: object required: - id - project_id - operation - status - created_at title: JobEntity description: Job response model representing an asynchronous operation. HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError ExtractionParameters: properties: model: type: string title: Model description: >- LLM model to use for extraction (e.g., 'openai/gpt-4', 'anthropic/claude-3-sonnet') model_kwargs: anyOf: - additionalProperties: true type: object - type: 'null' title: Model Kwargs description: Additional kwargs for the LLM model extraction_schema: additionalProperties: true type: object title: Extraction Schema description: JSON schema defining the desired output structure system_prompt: anyOf: - type: string - type: 'null' title: System Prompt description: High-level instructions or context for the extraction model user_prompt: anyOf: - type: string - type: 'null' title: User Prompt description: Specific hints about the current document generate_citations: type: boolean title: Generate Citations description: Whether to return bounding boxes for extracted values default: true generate_confidence: type: boolean title: Generate Confidence description: Whether to return confidence scores for extracted values default: true use_agentic: type: boolean title: Use Agentic description: >- Must stay False on POST …/extract (400 if True). The extract endpoint only runs single-pass LLM extraction. For agentic (decompose / ReAct) extraction, use POST …/research with ResearchJobRequest. default: false agentic_tools: anyOf: - additionalProperties: $ref: '#/components/schemas/ResearchToolConfigEntry' type: object - type: 'null' title: Agentic Tools description: >- Research-shaped tool map; ignored by the extract HTTP path (single-pass only). Use POST …/research for per-tool config. Same key shape as ResearchJobRequest.tools. execute_dependent_subtask_tool_hints: type: boolean title: Execute Dependent Subtask Tool Hints description: >- Ignored by POST …/extract (single-pass only). When True on a Research job, dependent subtasks execute their warm-start tool hints as real searches instead of converting them to intent-only guidance. default: false agent_settings: anyOf: - $ref: '#/components/schemas/AgenticSettings' - type: 'null' description: >- Ignored by POST …/extract. If ``agent_settings.use_agentic`` is True, the extract request is rejected (400); configure agentic behaviour via POST …/research instead. type: object required: - model - extraction_schema title: ExtractionParameters description: Parameters for extraction operation. SearchConfig: properties: top_k: anyOf: - type: integer - type: 'null' title: Top K description: Number of results to return filters: anyOf: - additionalProperties: anyOf: - type: string - additionalProperties: true type: object type: object - type: 'null' title: Filters description: 'Filters to apply to the search. For example, {file_id: 123}' rerank_config: anyOf: - $ref: '#/components/schemas/RerankConfig' - type: 'null' description: >- Reranking configuration. This is only applicable for SGP Vector Stores. vector_store_id: type: string title: Vector Store Id description: ID of the vector store external_id: anyOf: - type: string - type: 'null' title: External Id description: >- Engine-specific external identifier for the vector store. When set, this is passed directly to the engine's search/delete methods instead of vector_store_id. Populated from VectorStoreEntity.external_id. additionalProperties: true type: object required: - vector_store_id title: SearchConfig JobOperationType: type: string enum: - parse - batch_parse - extract - research - vector_store - chunk - summarization - create_index - update_index title: JobOperationType description: Enum for job operation values. JobStatus: type: string enum: - pending - running - succeeded - partially_succeeded - failed - cancelled title: JobStatus description: Enum for job status values. BatchParseProgress: properties: total: type: integer title: Total description: Total number of files in the batch succeeded: type: integer title: Succeeded description: Number of successfully parsed files failed: type: integer title: Failed description: Number of files that failed to parse cancelled: type: integer title: Cancelled description: Number of cancelled child jobs pending: type: integer title: Pending description: Number of files still pending (0 when job is terminal) child_jobs: items: $ref: '#/components/schemas/BatchChildJobInfo' type: array title: Child Jobs description: Per-child job status type: object required: - total - succeeded - failed - cancelled - pending - child_jobs title: BatchParseProgress description: Live progress tracking for a batch parse job. JobHistoryEvent: properties: step: type: string title: Step description: Human-readable step name timestamp: type: string format: date-time title: Timestamp description: When this event occurred duration_ms: anyOf: - type: integer - type: 'null' title: Duration Ms description: Duration in milliseconds (for completed steps) status: anyOf: - type: string - type: 'null' title: Status description: Event status (e.g., 'completed', 'failed') details: anyOf: - additionalProperties: true type: object - type: 'null' title: Details description: Additional event details type: object required: - step - timestamp title: JobHistoryEvent description: A single event in the job execution timeline. ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type input: title: Input ctx: type: object title: Context type: object required: - loc - msg - type title: ValidationError ResearchToolConfigEntry: properties: enabled: type: boolean title: Enabled description: When False, the tool is not offered to the agent default: true config: additionalProperties: true type: object title: Config description: Tool-specific configuration (e.g. top_k, filters, rerank_config) type: object title: ResearchToolConfigEntry description: >- Per-tool enablement and opaque config (e.g. search_config for similarity_search). AgenticSettings: properties: use_agentic: type: boolean title: Use Agentic description: >- Legacy mirror of ExtractionParameters.use_agentic. On POST …/extract, True causes a 400 rejection (extract is single-pass only). For agentic runs use POST …/research with ResearchJobRequest. default: false planning_model: anyOf: - type: string - type: 'null' title: Planning Model description: >- Model for planning steps (deciding what to search, decomposing tasks). Defaults to the main extraction model. acting_model: anyOf: - type: string - type: 'null' title: Acting Model description: >- Model for acting steps (producing the final extracted output). Defaults to the main extraction model. reflection_model: anyOf: - type: string - type: 'null' title: Reflection Model description: >- Model for reflection steps (synthesizing search results, aggregating subtasks). Defaults to the main extraction model. max_turns: type: integer title: Max Turns description: >- Maximum number of agent turns (decompose→run→aggregate cycles) before forcing completion. default: 2 max_context_tokens: type: integer title: Max Context Tokens description: Max tokens for LLM context (system + messages). Trims when exceeded. default: 100000 max_reasoning_hops: anyOf: - type: integer - type: 'null' title: Max Reasoning Hops description: >- Max reasoning hops (search or extract actions) per subtask in the ReAct loop. Default is 10. max_waves: anyOf: - type: integer - type: 'null' title: Max Waves description: >- Max dependency waves per turn. Subtasks beyond this depth are promoted to the last wave. Default is 3. max_validation_fix_attempts: type: integer title: Max Validation Fix Attempts description: Max attempts to fix schema validation errors via LLM reflection. default: 2 decompose_prompt_template: anyOf: - type: string - type: 'null' title: Decompose Prompt Template description: Custom decompose prompt template. When None, uses the default. decompose_rules: anyOf: - type: string - type: 'null' title: Decompose Rules description: >- Extra rules appended to the decompose prompt. When None, uses built-in rules. search_rules: anyOf: - type: string - type: 'null' title: Search Rules description: >- Extra rules appended to the reasoning step search action. When None, uses built-in rules. extraction_rules: anyOf: - type: string - type: 'null' title: Extraction Rules description: >- Extra rules appended to the reasoning step extract action. When None, uses built-in rules. type: object title: AgenticSettings description: >- Configuration for the agentic extractor (models, turns, hop limits, prompt templates). Tool enablement and search-summary scope for vector-store agentic extract live on :class:`ExtractionParameters` as ``agentic_tools`` (Research-shaped map); dependent-hint execution uses ``ExtractionParameters.execute_dependent_subtask_tool_hints``. RerankConfig: properties: model: anyOf: - type: string - type: 'null' title: Model description: Reranking model to use (uses system default if not specified) top_n: anyOf: - type: integer minimum: 1 - type: 'null' title: Top N description: Number of results to keep after reranking (defaults to top_k) instruction: anyOf: - type: string - type: 'null' title: Instruction description: >- Custom instruction for the reranking model (e.g., 'Given a medical question, retrieve relevant clinical passages'). Only applies to instruction-following rerankers like Qwen3. type: object title: RerankConfig BatchChildJobInfo: properties: source_document_id: type: string title: Source Document Id description: Source document ID job_id: type: string title: Job Id description: Child job ID status: $ref: '#/components/schemas/JobStatus' description: Status of the child job parse_result_id: anyOf: - type: string - type: 'null' title: Parse Result Id description: Parse result ID if succeeded error: anyOf: - type: string - type: 'null' title: Error description: Error message if failed type: object required: - source_document_id - job_id - status title: BatchChildJobInfo description: Status of a single child job within a batch. securitySchemes: ApiKey: type: apiKey in: header name: x-api-key description: API key for authentication AccountId: type: apiKey in: header name: x-selected-account-id description: Selected Account ID ````