Unstructured

At a glance

Availability: Experimental (how to enable).
Auth: API key.
Connection: The key from UNSTRUCTURED_API_KEY is sent as the unstructured-api-key header.
Docs: https://docs.unstructured.io/api-reference/partition/overview

Set these per environment. See Connect an integration.

Variable	Required	Description
`UNSTRUCTURED_API_KEY`	Yes	Unstructured API key, sent as the unstructured-api-key request header Docs.

Create an Unstructured account: Sign up at https://platform.unstructured.io - new accounts include a free trial of the serverless API, enough for testing without billing.
Get an API key: In the Unstructured UI, open the API Keys section, create a key, and copy it.
Store the key: Add UNSTRUCTURED_API_KEY=… to your .env file. Requests authenticate with the unstructured-api-key header (no Bearer prefix).
Verify access: Run the Partition Document tool with a small text or PDF file and strategy auto. A 401 means the key is wrong; a 422 usually means the files field is missing or unreadable.

The Partition endpoint processes one local file per request and does not fetch remote URLs - download the file first and pass its content
Your account may be issued a dedicated API URL; the default https://api.unstructuredapp.io/general/v0/general works for standard serverless accounts
hi_res and vlm strategies are slower and cost more per page than fast/auto but handle complex layouts and tables much better
Use chunking_strategy by_title with max_characters tuned to your embedding model for RAG-ready chunks

Tool	Access	Description
Partition Document	Read	Parse one local document into structured JSON elements (titles, narrative text, tables, lists) with optional chunking for RAG. The file content is uploaded as the multipart/form-data ‘files’ field together with the other fields; remote URLs are not accepted - fetch the file first and pass its content

Partition the document I provide with Unstructured using by_title chunking and summarize the resulting elements.
Partition my document with the hi_res strategy and extract the tables it contains.