creating-abstractions

To build custom OpenBench components for data ingestion, extend the DataSource abstract base class and implement required properties and methods. This enables OpenBench to fetch and format data from unique locations or proprietary APIs.

Yes, you can integrate external AI frameworks like LangChain by creating a framework adapter. This adapter connects external AI logic to the workbench, allowing you to run custom agents within OpenBench workflows.

To implement a custom vector store, create a Data Store component by extending the provided abstract base classes. This defines how OpenBench indexes and retrieves your processed data, including custom validation methods for robust pipeline execution.

Defining custom agents requires adhering to abstract property implementations and dataclass return types. You must also include validation methods to ensure robust pipeline execution when integrating your user-defined AI logic.

To format and export processed data, create a custom Output Generator by extending its abstract base class. This specifies exactly how processed data should be formatted and exported from the OpenBench pipeline.