How JournalInt Extracts, and What Data It Keeps
What the product does and the path of your documents, plainly. An extraction is a lead, not a finding.
How It Extracts
JournalInt reads the full text of each document in two stages, with your confirmation in between.
Scanned PDFs go through OCR first. Then:
- Entities: an AI model proposes people, organisations and places. You review and confirm them.
- Relations: with the confirmed entities, the model proposes the relations between them, from a closed vocabulary of types.
- Quote: every relation is kept with the sentence of your document that supports it.
The Verbatim Quote
Every relation carries the exact sentence of the document that supports it, and JournalInt checks that the sentence is in the text before showing it. No quote, no relation.
The quote is there for you to read in context. It does not replace checking the claim against other sources.
Human Review
You confirm the entities before the relations are extracted. The relations come afterwards, and you can correct or delete any of them.
Every item carries its provenance label: detected by AI, confirmed by a person or detected by a person.
Data and Providers
Your account, your files, the extracted text, the entities, the relations and the graph are stored in the EU. Nobody but you sees your projects.
To extract the relations, the text of your documents is sent to AI model and numeric-vector providers based in the United States. We use paid accounts where training on your documents is not enabled.
A record of each call to the model, with the text sent, is kept to debug errors and to know the cost. When you delete your account it is removed, normally within the hour, along with your documents, your graph and your files.
Visits to this site are counted with Vercel Web Analytics, anonymously and without third-party cookies.
What Is Coming
This is what is planned for the next phases, not what exists today. No dates: the order may change.
MVP
- A full graph to explore: filters by entity and relation type, search and a panel with the evidence for each relation.
- Graph editing: merge duplicate entities, review queues and an audit log of the changes.
- Two assistants that cite: one to ask the graph and one to ask your documents.
- Graph analysis: who is central and which communities there are.
V1
- Source quality: a three-level assessment of each document.
- Enrichment with registries and sanctions lists (OpenSanctions).
- Audio and video transcription.
- Your own AI model with your own keys, and account settings.
- Daily discovery of sources on the web.
- Chain of custody for your documents.
The providers’ names are given to anyone who asks, and the full privacy policy is given to invited people. Write to .