.svg.png){kind=small}
.svg.png){kind=icon}
CAD integration
Engineering data from Solidworks, Inventor and other CAD systems directly searchable.
HOW IT WORKS
From document to source-accurate answer.
At its core, KoAssist is a retrieval system: documents become knowledge building blocks, every question becomes a targeted search, and the answer is built from the most relevant sources. Four steps, each transparent.
PIPELINEFOUR STAGES
How documents become answers.
- 01 · INGEST
Documents are indexed.
Standards, specs and project documents are ingested and broken into semantic building blocks: small knowledge units that can be searched precisely later. OCR for scanned PDFs is built in.
VDI 2230 Blatt 1 (240 pp.) → 1,842 text segments → Indexed in 4 min - 02 · UNDERSTAND
The question is understood.
The question is converted into a semantic representation. KoAssist doesn't search for exact keywords but for meaning: technical terms, synonyms and rephrased variants are all considered.
"Torque M16?" → Meaning recognized (tightening torque, bolted joint, preloaded, M16 thread) → ~0.2s → ready to search - 03 · RETRIEVE
Relevant sources are found.
From the entire knowledge base, the five to fifteen most relevant passages are retrieved, ranked by semantic similarity to the question. These sources form the basis for the answer.
Top hits: 94% VDI 2230 Bl.1 · 87% VDI 2230 Bl.2 · 71% Eng. HB §4.2 - 04 · GENERATE
The answer is built from the sources.
The answer is formulated from the retrieved passages, with every statement anchored in a specific source (file, page). If the sources don't provide a clear answer, that is reported transparently instead of guessed.
"At least property class 8.8[1]. Safety-critical: 10.9 / 12.9[2]." + 2 sources linked
WHAT'S SUPPORTED
Formats, languages, sources.
File formats
PDF (incl. OCR)DOCXXLSXPPTXTXT, MD
Languages
GermanEnglishmore on request
Source systems
SharePointGoogle DriveConfluenceNetwork drivesERP / PDM / PLM / CRM (separate)
OUTLOOK
What's coming next.
In progress· 3 Module
Extended PLM integrations
Bills of materials, change statuses and approvals from Teamcenter, Windchill and other PLM systems.
Knowledge from emails
Bring targeted email conversations into the knowledge space. No blanket mailbox connection, only relevant threads.
Planned for Q3· 3 Module
English interface
KoAssist fully in English, for international engineering teams.
Duplicate part search
Analyze drawings, automatically identify reuse potential.
Additional model providers
Support for additional AI models, including local inference for maximum data sovereignty.
Concept· 3 Module
Project documentation
Simplified documentation of technical decisions.
Operating manuals
The tool drafts, the engineer reviews and approves.
Standards and compliance
Automatically check and document technical requirements.
SETUP
From initial call to roll-out.
01approx. 60 min
Discovery call
Understand your data landscape, clarify integrations.
025-15 business days
Indexing
Documents are indexed, knowledge base prepared.
032 weeks
Pilot with a small team
3-5 engineers test, feedback is incorporated.
04continuous
Roll-out
Scaled to the full department.
- 01approx. 60 min
Discovery call
Understand your data landscape, clarify integrations.
- 025-15 business days
Indexing
Documents are indexed, knowledge base prepared.
- 032 weeks
Pilot with a small team
3-5 engineers test, feedback is incorporated.
- 04continuous
Roll-out
Scaled to the full department.
TECHNICAL QUESTIONS
Answers for IT and engineering.
Which languages are supported?
German and English by default. Additional languages are technically possible and configured on request. Answers appear in the language of the question.
How large can the document base be?
Up to 10,000 pages on the Solo plan, up to 75,000 pages on the Department plan, up to 250,000 pages on the Plant plan. Larger document bases on the Enterprise tier on request. Indexing time scales linearly: roughly 4 minutes per 100 pages of standard text.
How current are the answers?
KoAssist works on a snapshot of the document base that is regularly re-indexed. The update frequency is configurable: daily, weekly or on change via webhook. Real-time indexing is under development.
What about encrypted or scanned PDFs?
Scanned PDFs are processed via OCR (Optical Character Recognition), with German and English text recognition built in. Password-protected PDFs must be decrypted beforehand. KoAssist does not bypass encryption.
Can we export answers and sources?
Yes. Every answer can be copied with its linked sources or exported as a small document. A full conversation history is available per user and can be downloaded from the system.
How is KoAssist different from a traditional search?
Traditional search finds documents that contain a keyword. KoAssist finds the specific passage that answers a question, summarizes it and cites the source with file and page. It also understands synonyms and technical terms, which pure keyword search cannot do.