ContractExtractor¶

Loads a document (or takes text), chunks it and makes schema-constrained model calls, returning a Contract. See Contract analysis.

`ContractExtractor` ¶

Main extraction orchestrator that coordinates document loading, chunking, LLM-based extraction, and result assembly.

`init(llm_provider: LLMProvider | None = None, llm_provider_name: str | None = None, document_loader: DocumentLoader | None = None, chunking_strategy: ChunkingStrategy | None = None, confidence_threshold: float = 0.7, parallel_processing: bool = True)` ¶

Initialize the contract extractor.

Parameters:

Name	Type	Description	Default
`llm_provider`	`LLMProvider \| None`	Custom LLM provider instance	`None`
`llm_provider_name`	`str \| None`	Full model name: `gpt-...` (OpenAI), `claude-...` (Anthropic), anything else is an Ollama model for LocalProvider. One of `llm_provider` / `llm_provider_name` is required; there is no default provider.	`None`
`document_loader`	`DocumentLoader \| None`	Custom document loader	`None`
`chunking_strategy`	`ChunkingStrategy \| None`	Custom chunking strategy	`None`
`confidence_threshold`	`float`	Minimum confidence score for extractions	`0.7`
`parallel_processing`	`bool`	Whether to process chunks in parallel	`True`

`estimate_extraction_cost(document_path: str) -> dict[str, Any]` ¶

Estimate the API cost of extracting a document without running extraction.

Loads and chunks the document, then uses the configured LLM provider's cost model to estimate the total token usage and USD cost across all extraction phases.

Parameters:

Name	Type	Description	Default
`document_path`	`str`	Path to the contract document.	required

Returns:

Type	Description
`dict[str, Any]`	Dict with keys `estimated_cost` (float, USD), `estimated_tokens`
`dict[str, Any]`	(int, input tokens), `num_chunks` (int), `llm_provider` (str,
`dict[str, Any]`	provider class name), `llm_model` (str) and `breakdown` (dict,
`dict[str, Any]`	per-phase cost and token estimates).

Example::

extractor = ContractExtractor(llm_provider=LocalProvider(model="llama3.1:8b"))
estimate = extractor.estimate_extraction_cost("contract.pdf")
print(f"Estimated cost: ${estimate['estimated_cost']:.4f}")

`estimate_extraction_cost_from_text(text: str) -> dict[str, Any]` ¶

Like estimate_extraction_cost but for already-loaded text.

`extract(document_path: str, contract_type: ContractType | None = None, analyze_risks: bool = True, extract_financial: bool = True, known_parties: list[str] | None = None) -> Contract` ¶

Extract all data from a contract document.

Parameters:

Name	Type	Description	Default
`document_path`	`str`	Path to the contract document	required
`contract_type`	`ContractType \| None`	Optional contract type hint	`None`
`analyze_risks`	`bool`	Whether to perform risk analysis	`True`
`extract_financial`	`bool`	Whether to extract financial terms	`True`
`known_parties`	`list[str] \| None`	Optional list of known party names (hints)	`None`

Returns:

Type	Description
`Contract`	Contract object with all extracted data

Raises:

Type	Description
`ExtractionError`	If extraction fails

`extract_from_text(text: str, contract_type: ContractType | None = None, analyze_risks: bool = True, extract_financial: bool = True, known_parties: list[str] | None = None) -> Contract` ¶

Like extract but for already-loaded text (no file metadata).

`extract_async(document_path: str, **kwargs: Any) -> Contract` `async` ¶

Async version of extract — runs the synchronous extractor in a thread pool.

Parameters:

Name	Type	Description	Default
`document_path`	`str`	Path to the contract document	required
`**kwargs`	`Any`	Additional arguments passed to extract()	`{}`

`extract_batch(document_paths: list[str], max_workers: int = 4, **kwargs: Any) -> list[Contract]` ¶

Process multiple contracts in parallel.

Parameters:

Name	Type	Description	Default
`document_paths`	`list[str]`	List of document paths to process	required
`max_workers`	`int`	Maximum number of parallel workers	`4`
`**kwargs`	`Any`	Additional arguments passed to extract()	`{}`

Returns:

Type	Description
`list[Contract]`	List of successfully extracted Contract objects (failed paths are skipped).

ContractExtractor¶