Document Management & OCR

Upload & versioning

Documents can be uploaded via drag-and-drop into a matter, a contract or the central document repository. Every new version of a document is tracked as a version — the previous version remains accessible, so it can be reconstructed what state applied at what time. Deletes are soft deletes with recovery; nothing is silently removed.

Supported upload formats: PDF, DOCX, XLSX, PPTX, TXT, EML, MSG, HTML, RTF, common image formats. Storage location: EU datacenter Frankfurt, encrypted at-rest.

OCR

Scanned documents and image PDFs are useless for search without text recognition. LEGALinhouse automatically applies an OCR layer to uploaded documents, so full text becomes searchable and readable for the AI. OCR is set by default on every upload and runs in the background.

OCR itself is a pure text-extraction procedure — it reads what is on the paper, nothing more. It does not send content to a language model.

Selectable recognition quality — "Fast" or "Best"

For clean scans the fast procedure is enough. For difficult originals — faded copies, pages scanned at an angle, old faxes — a more accurate model delivers noticeably better results. You choose the level yourself:

• Two levels. Fast is the default and matches the previous behaviour; Best uses a more accurate recognition model and takes a little longer.
• Set per organisation. Administrators set the level under Settings → Company settings → "OCR quality". It then applies to every text recognition in your organisation.
• No extra cost. Text recognition consumes no credit — both levels are free. "Best" only costs a little more processing time.
• Higher resolution. Pages are now read at 300 dpi rather than 200 dpi, which improves recognition even at the "Fast" level.

AI categorization

So a document does not just sit on the server but lands in the right matter, on the right deadline and with the right document type, the AI proposes a categorization:

• Document type (contract, brief, authority letter, invoice, internal memo, ...)
• Matter and contract assignment (based on detected reference numbers, names, dates)
• Deadline detection (e.g. opposition period in an administrative decision)
• Brief summary for the file

The suggestions go to the inbox; an employee confirms them. Before any AI processing the 3-phase pseudonymization runs.

AI-free import path

For highly sensitive documents — personnel files, whistleblower notifications, criminal matter files, M&A documents — LEGALinhouse offers a fully AI-free import path. The document is uploaded, OCR text is extracted, the document is filed in the matter and made searchable — the inference pipeline is never touched.

The selection is made by tenant admin per document type or per upload source — a safe default, rather than deciding every upload individually.

Full-text search

All documents (including OCR-recognized scans) are full-text searchable — across all modules. A search for a case reference finds the authority letter, the dunning notice, the email attachment and the contract clause in the same result list. Filters by document type, matter, date, tenant and employee are available.

Output documents — an opinion is not a letter

At the end of a matter there is often a document: an internal opinion, a legal analysis, a memo. Not a covering letter to somebody, but a paper for your own organisation that records the facts.

For a long time the letter was a case's only output format — and a letter brings everything an opinion does not need: letterhead, address field, "your reference / our reference", a list of enclosures, a recipient and an outgoing-mail entry.

The case now has documents of its own. They carry your organisation's header and footer — with no letterhead, no address field, no recipient. The available document types are internal opinion, legal analysis, memo, note and statement.

• Two routes in. From an AI result via "adopt as document" — the text is taken over unchanged, there is no second AI request and no additional tokens are charged — or by hand as an empty document.
• Draft and release. While a document is a draft, every PDF carries a DRAFT notice. Releasing removes it and additionally makes the document findable by the AI within that case. Edit it later and the AI follows automatically; withdraw the release and it disappears from AI search again.
• Versions. A document can be revised; the earlier version stays identifiable as superseded. Discarding is not permanent deletion.
• Checked citations before release. If a document still holds unverified citations, the release raises a notice instead of letting it through silently. You either check them or confirm explicitly — and the confirmation is logged.
• On handover to an external law firm, released documents are passed on in full rather than as a short extract of the opening lines. Drafts are not handed over.

Taking a document out of the AI context

Some material should stay in the case but not be drawn on by the AI: a superseded version, someone else's draft, a letter whose account of the facts you do not want used as a basis. For a long time there was exactly one way to achieve that, and it was deletion — which destroyed the document as case material at the same time.

New: in the Documents tab every row carries an eye icon. One click takes the document out of the AI context, a second brings it back.

The file is fully preserved — visible, downloadable, exportable, part of the case and part of any handover. It is simply no longer found by any AI request: not in the AI chat, not in AI tasks, not in a letter draft, not in research across the case.