The memory graph engine

Everything your company knows, as a memory you can ask.

Engram reads the files you already have and keeps what they say: who did what, when, for how much, and on which page. One private database per account, hosted in Germany.

What did Thorsten Blum do last week?
09.09.2026 – Energie Ritter GmbH bestätigt die Montage der Wallbox am 16.09.2026 und bittet um die Zählernummer. 10.09.2026 – Lieferung von 12 PV-Modulen, Lieferschein 4100864600. 12.09.2026 – E-Mail: Besprechung am 14.09. um 10:00 bestätigt, Lieferverzug gemeldet. Sources: Brief_Zusage_Montage.pdf p1 · Lieferschein_scan.png p1 · mail_termin.eml p1
How it works
01

Upload

Drop a folder: PDFs, scans, e-mails, spreadsheets, photos. Structure and names are kept.

02

Read together

Related files are read in one pass, so the model sees who is who across letters, invoices and mails.

03

Remember

Every person, company, thing and dated happening is stored once, with the page and quote it came from.

04

Ask

Ask in plain words, browse the tables, or call the generated API from your own tools and agents.

Cited, always

Nothing is stored without a locator. Every value opens the file, page and quote it was read from, and disagreements between documents are kept, not hidden.

Isolated, always

Each account is its own database on a server in Germany. Sign-in is Google through Firebase; your files never leave the EU for storage.