Knowledge Base

One knowledge base your whole team can trust

Upload contracts, handbooks, spreadsheets, recordings and videos. FlowSentric grounds every answer in what you actually gave it and cites the document and page — in English and German.

Start free Book a demo

The Knowledge Base with Documents, Web Sources and Audio & Video tabs and a drag-and-drop upload area

42file formats (28 document + 14 media)

50 MBper document, 500 MB per video

28full-text search languages incl. German

100%embeddings computed on our servers

120OCR pages per document, English & German

11connected source types

What it does

Every format you actually use

PDF, Word, Excel, PowerPoint, CSV, Markdown, HTML, JSON, images with OCR — plus MP3, WAV, MP4, MOV and more, transcribed on our own servers. Import YouTube links and podcast feeds.

Cited answers, not hallucinations

Hybrid search fuses vector similarity with full-text ranking. Every answer names its [Source: document (page N)], and the model is told exactly which documents exist.

Private by construction

Personal or organization visibility per document, folders as collections, version history with revert, and Privacy Guard masking retrieved context before it reaches a model.

How it works

01Upload documents, drag in a folder, or connect Slack, GitHub, Notion, Jira, Google Drive, RSS or email as a source.

02Files are extracted (with German-aware OCR), chunked with overlap and embedded locally — no setup required.

03Chat and agents retrieve the best passages with hybrid search and answer with page-level citations.

04Update a file and every future answer reflects the change; reindex a folder any time.

Under the hood

Retrieval

Hybrid semantic + keyword search with combined ranking
Optional reranking and query expansion for tighter results
Page-aware chunking so citations point to the right page

German-first

OCR reads English and German; umlauts and ß are preserved
Full-text search configurable to german (or several languages at once)
Contract, lease and invoice layouts handled by the extractor

Runs on your box

Embeddings computed with multilingual models on our own servers — text never leaves to be embedded
Audio and video transcribed by a local speech-recognition model
Optional local vision model for scanned pages and charts

Built for

Real-world scenarios

Client contract Q&A

Internal policy and handbook lookup

Meeting recordings turned searchable

Compliance reference library