Overview

Unstructured recommends that you use the Unstructured user interface (UI) or the Unstructured API instead of the Unstructured Ingest CLI or the Unstructured Ingest Python library.

The Unstructured UI and API provide a full range of partitioning, chunking, embedding, and enrichment options for your files and data. It also uses the latest and highest-performing models on the market today, and it has built-in logic to deliver the highest quality results at the lowest cost.

The Unstructured Ingest CLI and the Unstructured Ingest Python library are not being actively updated to include these and other Unstructured API features.

You can send multiple files in batches to be ingested by Unstructured for processing. Ingestion is the term that Unstructured uses to refer to the set of activities that happens when files are input for processing. Learn more.

You can send batches to Unstructured by using the following tools:

• The Unstructured Ingest CLI, which is part of and builds upon the Unstructured open source library.
• The Unstructured Ingest Python library, which is also part of and builds upon the Unstructured open source library.

The following 3-minute video shows how to use the Unstructured Ingest Python library to send multiple PDFs from a local directory in batches to be ingested by Unstructured for processing:

The following 5-minute video goes into more detail about the various components of the Unstructured Ingest Python library:

Ingest flow

The Unstructured ingest flow is similar to an extract, transform and load (ETL) data pipeline. Because of this, a customer-defined implementation of the Unstructured ingest flow is sometimes referred to as an ingest pipeline or simply a pipeline. An Unstructured ingest pipeline contains the following logical steps:

1. Index

   Reaches out to the source location and pulls in metadata for each document.

   For example, this could include information such as the path to the files to be analyzed.

   • For the Unstructured CLI, you can control this behavior, where available for a connector, through its --input-path command option.
   • For the Unstructured Ingest Python library, you can control this behavior, where available for a connector, through its <Prefix>IndexerConfig class (where <Prefix> represents the connector provider’s name, such as Azure for Azure.)
2. Post-Index Filter

   After indexing, you might not want to download everything that was indexed.

   For example, you might want to download only files that match specific types, file names, paths, or sizes.

   For the Unstructured Ingest Python library, you can control this behavior through the FiltererConfig class.

3. Download

   Using the information generated from the indexer and the filter, downloads the content as files on the local file system for processing. This may require manipulation of the data to prepare it for partitioning.

   For example, this could include information such as the path to a local directory to download files to.

   • For the Unstructured CLI, you can control this behavior through a connector’s --download-dir command option.
   • For the Unstructured Ingest Python library, you can control this behavior through a connector’s <Prefix>DownloaderConfig class.
4. Post-Download Filter

   After downloading, if uncompression is enabled, you might not want to uncompress everything that was downloaded. The filter that was defined at the beginning is repeated here.

5. Uncompress

   If enabled, searches for any compressed files (Unstructured supports TAR and ZIP) and uncompresses them.

   • For the Unstructured CLI, you can control this behavior through the --uncompress command option.
   • For the Unstructured Ingest Python library, you can control this behavior through the UncompressConfig class.
6. Post-Uncompress Filter

   After downloading, and uncompressing if enabled, you might not want to partition everything that was downloaded, and uncompressed if enabled. The filter that was defined at the beginning is repeated here again.

7. Partition

   Generates the structured enriched content from the local files that have been downloaded, uncompressed if enabled, and filtered. Both local-based partitioning and Unstructured-based partitioning is supported, with API services-based partitioning set to run asynchronously and local-based partitioning set to run through multiprocessing.

   For the Unstructured Ingest Python library, you can control this behavior through the PartitionerConfig class.

8. Chunk

   Optionally, chunks the partitioned content. Chunking can also be run locally or through Unstructured, with asynchronous or multiprocessing behaviors set in the same approach as the partitioner.

   For the Unstructured Ingest Python library, you can control this behavior through the ChunkerConfig class.

9. Embed

   Optionally, generates vector embeddings for each element in the structured output. Most of the time, this is done through API calls to a third-party embedding vendor and therefore runs asynchronously. But it can also use a locally available Hugging Face model, which will run through multiprocessing.

   For the Unstructured Ingest Python library, you can control this behavior through the EmbedderConfig class.

10. Stage

    This is an optional step that does not apply most of the time. However, sometimes the data needs to be modified from the existing structure to better support the upload step, such as converting it to a CSV file for tabular-based destinations.

    For the Unstructured Ingest Python library, you can control this behavior, where available for a connector, through its <Prefix>UploadStagerConfig class.

11. Upload

    Writes the local content to the destination. If no destination is provided, the local one will be used, which writes the final result to a location on the local filesystem. If batch uploads are needed, this will run in a single process with access to all documents. If batch uploads are not supported, all documents can be uploaded concurrently by using the asynchronous approach.

    For the Unstructured Ingest Python library, you can control this behavior through a connector’s <Prefix>UploaderConfig class.

Generate Python code examples

You can connect any available source connector to any available destination connector. However, the source connector code examples in the documentation show connecting only to the local destination connector. Similarly, the destination connector code examples in the documentation show connecting only to the local source connector.

To quickly generate an Unstructured Ingest Python library code example that connects any available source connector to any available destination connector, do the following:

1. Open the Unstructured Ingest Code Generator webpage.

2. Select your input (source) location type from the Get unstructured documents from drop-down list.

3. Select your output (destination) location type from the Upload RAG-ready documents to drop-down list.

4. Select your chunking strategy from the Chunking strategy drop-down list:

   • None - Do not chunk the data elements’ content.
   • basic - Combine sequential data elements to maximally fill each chunk. However, do not mix Table and non-Table elements in the same chunk.
   • by_title - Use the basic strategy and also preserve section boundaries. Optionally preserve page boundaries as well.
   • by_page - Use the basic strategy and also preserve page boundaries.
   • by_similarity - Use the sentence-transformers/multi-qa-mpnet-base-dot-v1 embedding model to identify topically similar sequential elements and combine them into chunks. This strategy is availably only when calling Unstructured.

   To learn more, see Chunking strategies and Chunking configuration.

5. For any chunking strategy other than None:

   • Enter your chunk size in the Chunk size (characters) box, or leave the default of 1000 characters.
   • If you need to apply overlapping to the chunks, enter the chunk overlap size in the Chunk overlap (characters) box, or leave default of 20 characters.

   To learn more, see Chunking configuration.

6. To generate vector embeddings, select the provider in the Embedding provider drop-down list.

   To learn more, see Embedding configuraton.

7. Click Generate code.

8. Copy the example code from the Generated Code pane into your code project.

9. The code example will contain one or more environment variables that you must set for the code to run correctly. To learn what to set these variables to, click the documentation links that are below the Generated Code pane.

Pricing

Calls to the Unstructured CLI or Unstructured Ingest Python library that are routed to Unstructured’s software-as-a-service (SaaS) for processing (for example, by specifying an Unstructured API key and an Unstructured SaaS URL) require an Unstructured account for billing purposes.

Unstructured offers three account pricing plans:

• SaaS Cloud-hosted - Processing happens on Unstructured’s software-as-a-service (SaaS) cloud infrastructure in a multi-tenant environment.
• Private SaaS - Processing also happens on Unstructured’s SaaS cloud infrastructure, but your data stays protected in a dedicated cloud environment, maintaining strict data privacy.
• VPC - Sometimes referred to as self-hosted, an instance of the Unstructured SaaS is deployed into your own virtual private cloud (VPC), providing complete data ownership and infrastructure control, full customization, and dedicated technical support.

For more details, see the Unstructured Pricing page.

Some of these plans are billed on a per-page basis.

Unstructured calculates a page as follows:

• For these file types, a page is a page, slide, or image: .pdf, .pptx, and .tiff.
• For .docx files that have page metadata, Unstructured calculates the number of pages based on that metadata.
• For all other file types, Unstructured calculates the number of pages as the file’s size divided by 100 KB.
• For non-file data, Unstructured calculates a page as 100 KB of incoming data to be processed.

Learn more

• Ingest configuration settings enable you to control how batches are sent and processed.
• Source connectors enable you to send batches from local or remote locations to be ingested by Unstructured for processing.
• Destination connectors enable Unstructured to send the processed data to local or remote locations.

See also

• The Unstructured UI enables you to send batches to Unstructured from remote locations, and to have Unstructured send the processed data to remote locations, all without using code or a CLI.

Assistant

Responses are generated using AI and may contain mistakes.