Skip to content
Escriba Suite Escriba Suite

What is Escriba?

Escriba is the universal translator into the language of AI. It takes any document — a PDF, a Word file, a spreadsheet, an image, an audio recording, a web page, a YouTube link — and turns it into clean, anonymized Markdown, the format that large language models read best.

It solves, in one self-hostable tool, the three headaches of feeding documents to an LLM:

  • Noisy, token-hungry input → clean, structured Markdown.
  • Sensitive-data leakage → built-in PII anonymization, with reversible pseudonymization.
  • “Will it fit? what will it cost?” → a local LLM-prep panel that counts tokens, estimates cost with live pricing, checks context-window fit and chunks for RAG.

It runs locally, in 7 languages, and is built on Microsoft MarkItDown.

Who it’s for

Section titled “Who it’s for”
  • Anyone who pastes documents into ChatGPT, Claude or Gemini and wants the text clean — and the private bits stripped out — before it leaves their hands.
  • Teams and institutions that can’t send confidential files to a third-party cloud, and need a converter that runs on their own server.
  • Builders who want a REST API, role-based access and a single Docker image with no moving parts.

What makes it different

Section titled “What makes it different”
  • The control stays on the human layer. Escriba doesn’t send your documents anywhere. It runs on your machine, deletes files right after conversion, and lets you decide what reaches a model.
  • Anonymization is reversible. Pseudonymize → send to the LLM → re-hydrate the reply locally. The restore map never leaves your browser.
  • No AI required. The smart parts — token counting, OCR, anonymization, cost estimates — all run locally. AI is strictly optional.

Next steps

Section titled “Next steps”