jxnl/instructor
Instructor
Instructor is a library designed to parse, validate, and map unstructured language model responses into strongly typed, schema-compliant data objects. It provides a framework for structured data extraction that uses data modeling classes to enforce strict type constraints on model outputs, ensuring that generated content consistently matches expected structures.
The library distinguishes itself through an automated error recovery system that manages the lifecycle of failed extraction attempts. When a model output fails to meet defined schema requirements, the framework automatically triggers a recursive feedback loop, sending specific error messages back to the language model to prompt self-correction until the output is valid. This process is supported by a provider-agnostic interface layer that standardizes communication protocols, allowing for a consistent extraction workflow across different language model services.
Beyond core extraction and validation, the library supports real-time data processing through incremental stream parsing. This capability allows for the consumption of structured objects as they are generated, reducing latency in complex workflows. The project is distributed as a Python library with documentation focused on integrating these validation and mapping patterns into application environments.
Features
- LLM Response Parsers - Automates the extraction and validation of JSON from language models with built-in retry and error feedback mechanisms.
- Error-Correction Feedback Loops - Implements recursive feedback loops that provide structural error messages to the model for iterative self-correction.
- Structured Data Extraction - Maps unstructured text into validated objects using schema definitions to ensure consistent output formats.
- Schema-Validated Data Structures - Enforces schema constraints on unstructured text using data models to ensure returned content matches expected structures.
- LLM Schema Outputs - Enforces strict data types and structural constraints on model outputs to ensure reliable data processing.
- Pydantic Schema Outputs - Uses data modeling classes to define expected output structures and validate model responses against strict type constraints.
- LLM-to-Structured Data Converters - Maps unstructured text from language models into validated objects using schema definitions for type safety.
- LLM Provider Interfaces - Standardizes the interface for structured data extraction across multiple language model services.
- Multi-Provider Abstractions - Standardizes the interface for structured data extraction across various language model providers.
- Provider-Agnostic Model Interfaces - Provides a standardized interface layer to maintain consistent extraction workflows across different language model providers.
- Incremental JSON Parsing - Parses partial JSON fragments into structured objects in real time as they are generated.
- Real-Time Text Streaming - Processes and yields structured data objects incrementally as they are generated to reduce latency.
- Structured AI Output Streams - Yields structured data incrementally as it is generated to allow for real-time processing of nested responses.
- Agent Error Recovery - Automatically retries failed extraction attempts when output does not match the expected schema.
- Automatic Task Retries - Automatically re-executes failed extraction requests to ensure schema compliance without manual intervention.
- Response Type Mapping - Transforms unstructured natural language responses into strongly typed objects for reliable data integration.
- RAG Frameworks and Toolkits - Library for extracting structured data from text using LLMs.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
