unstract

Unstract uses LLMs to extract structured JSON from documents (PDFs, images, scans) based on natural-language prompts, and lets you deploy the result as an API or an ETL pipeline. It is aimed at teams in finance, insurance, healthcare and KYC or compliance. The README's comparison says that instead of writing regex and per-vendor templates you write a prompt once, a new document type takes minutes in Prompt Studio, and any provider (OpenAI, Anthropic, Bedrock, Ollama) can be plugged in.

Key pieces are Prompt Studio for defining extraction schemas, API deployment (send a document over REST, get JSON back), an ETL pipeline that pulls documents from sources such as S3, MinIO, Google Cloud Storage, Azure Blob, Google Drive, Dropbox or SFTP and loads them into a warehouse, an MCP server for agents such as Claude, and an n8n node. To run it locally you need Linux or macOS, Docker and Docker Compose, Git and at least 8 GB of RAM, then clone the repository and run `./run-platform.sh`. A managed cloud and an enterprise edition with SSO and role-based access control are offered, and the open-source code is AGPL-3.0.