Answers with source citations – from your documents.
MadeForPrivacy makes file shares, contracts, Handbooks and mailboxes searchable for agents, with RAGFlow as the Knowledge layer running in your installation. Every answer shows where it comes from – and only people allowed to see the source get it.
How documents become knowledge
Ingest, split, retrieve – traceably.
In MadeForPrivacy, Knowledge is a tool step like any other: ingestion writes to a source, retrieval reads from it. Both run in the pipeline, with a data zone and tracking.
Ingest
Files, text or entire file shares – PDF, Office, scans. RAGFlow recognizes layout and tables instead of just scraping text.
ingestionSplit & index
Documents are split into meaningful sections and stored as vectors. Even large collections remain reliably searchable.
chunks + embeddingsSearch
Keyword and vector search combined, results weighted against each other, deduplicated and re-ranked.
hybrid retrievalCite
Results go into the briefcase as items – with a reference to the original document. The next step answers based on them, not on guesswork.
memory itemsSources & access
Separated by team, department, client – and person.
Every Knowledge source is a catalog entry with a type, data zone and groups. Agents only search sources the requesting person is authorized to access – nothing gets mixed up.
- DatasetsDocument indexes for file shares, contracts, product documentation, tickets. Ingestion via pipeline or bulk import of entire drives.
- Tied to AD groupsAnyone in the group in your directory sees the source. Anyone who leaves the group no longer sees it – no maintenance in the product.
- Personal sourcesUser-based sources keep a separate area for each person. Neva's memory is built exactly the same way.
- Handbooks as a curated sourceGit-backed document collections with read and write permissions per group – maintained knowledge instead of file sprawl.
Documents
PDF, Word, Excel, scans, emails – with layout and table recognition.
Groups
Access per source via your AD groups.
Data zone
Source and agent each carry a zone; local stays local.
Handbooks
Versioned collections with permissions, used as a source.
Memory
Neva remembers what matters to you. Only for you.
Besides documents, there is a second pillar: memories. Before every answer, the personal assistant decides separately whether to look something up and whether to remember something – deterministically and isolated per person.
rawThe conversation turn as it happened – question and answer.
semanticFacts and preferences: responsibilities, formats, personal preferences.
episodicWhat happened when – the course of events, not just the result.
proceduralHow things get done: recurring procedures.
Storing and looking up are two separate decisions – Neva stores only your message, never her own answer, and tells you when she has remembered something.
Handbooks
Knowledge that agents not only read but also maintain.
A Handbook is a versioned document collection: Markdown pages with metadata in a Git working directory, validated against a schema and rendered as a searchable website. For quality management documentation, processes, operating manuals – and as a personal Handbook for each person.
Git, not a file share
Every Handbook is its own Git repository. Every change is a commit – with the person as author, even if an agent wrote it. History, diffs and rollback included; push to a remote if you want.
Schema instead of free text
Pages carry front matter based on the Handbook's YAML schema: status, owners, progress, review date. The build validates every page against it; anything that doesn't match is not published.
Permissions per group
Read and write access is tied to AD groups in the admin catalog – not to a file in the repository. Every person automatically gets their own Handbook ("My Handbook").
Readable in the browser
The working directory is turned into a static, searchable website, with navigation based on the folder structure and version history for each page – accessible with your normal sign-in, no client required.
For agents, Handbooks are tools: an explorer tool finds pages, a reader tool puts them into the briefcase, and a writer tool creates or edits pages – as a commit in the person's name, locked against concurrent runs. So an agent can extend an inspection log or update a work instruction, and Knowledge makes the same Handbook searchable for questions.
The Knowledge layer
RAGFlow – open source, installed on your infrastructure.
RAGFlow is an open-source engine for retrieval-augmented generation with deep document understanding. MadeForPrivacy runs it as part of the stack: content never leaves your infrastructure, and every search result comes with a reference to the original.
Layout & tables
Recognizes structure in PDFs and scans, not just running text.
Hybrid search
Keyword and vector results weighted, deduplicated and ranked.
Source citations
Every result points to the document and the exact passage.
Self-hosted
Runs as a container in the MadeForPrivacy stack, GPU optional.
Live demo
Which file share should answer first?
Bring a folder, a Handbook or a mailbox – we’ll show you live how agents turn it into answers backed by sources.