Step 1 · The documents are foundWhere does the knowledge live today?
Enterprise AI assistantYou ask your company documents a question, and the answer comes with its source.
Contracts, procedures, price lists and past correspondence gather behind one search box. You type the question in your own words, and under the answer it says which page of which document it came from. You check it yourself instead of asking three people.
Connected sources
- Google Drive
Notion
Gmail
Slack
PostgreSQL
Claude- File server
How it is built
Your documents join one searchin three steps.
First we find out which folders and which mailboxes the knowledge lives in today. Then we connect those sources to one search and carry each source's access rights over unchanged. Last, we start tracking which documents people turn to and how often.
Step 2 · A question is askedThe answer comes with its source.
Step 3 · Use is trackedThe most used documents become visible.
What it does
The answer is foundand its source is shown next to it.
No answer is given without a document behind it. The documents in the search, the answers given and the questions left unanswered all sit in one panel.
Every document goes into one search
PDFs, Word files, spreadsheets, decks and scanned documents go into the same search. Whether a file sits in Drive, in Notion or in an email attachment, you ask the question in one place.
Every answer shows its source
Under the answer it says which document and which page it came from. You click the link, open the document and see for yourself that the sentence is really there.
Unknown questions get no invented answer
When the documents hold nothing on a question, the reply says so and the question goes onto the gap list. That list shows which knowledge has not been written down yet.
Old and new versions do not mix
When a document is updated the new version applies and the old one stays on record. If two documents say different things on the same subject, both are shown side by side and you choose which one applies.
You see what the team asks most
How often each type of document is consulted sits in the panel. A subject people ask about often is the first candidate for writing down properly.
New staff find answers on their own
Procedures, prices and past proposals are found without asking anyone. The panel tracks how much the search is used month by month.
Monthly report
Which documents are used, and how often,is reported every month.
The report shows the share of answers given with a source, the number of documents in the search and the documents consulted most. Unanswered questions are listed too, which tells you which document needs writing.
- Share of answers with a source
- Most consulted documents
- Questions left unanswered
Frequently asked
Knowledge platform Frequently asked questions
An enterprise AI assistant is a search tool that lets you ask your company's own documents a question. You type the question in your own words, and the answer comes with the page of the document it came from. Unlike a general tool such as ChatGPT, it answers only from your documents.
No. The search runs on your own infrastructure and your documents are not used to train models. When an answer is written, only the question itself and the relevant passage go out, never the full archive.
The permissions at the source are kept exactly. Someone who cannot see a folder cannot see an answer drawn from a document in it either. Role-based permissions are written during the build and every query is logged.
Every answer shows which section of which document it came from. You click the link and open the source. An answer with no document behind it is not given.
It is restricted so that it does not. When the documents hold nothing on a question, the reply says so. The question then goes onto the gap list, which shows which document is missing.
Both. The conflict is flagged and the two passages are placed side by side with their dates. A person picks the version that applies, and that choice is recorded.
PDF, Word, Excel, decks, plain text and scanned documents. Text is extracted from scans; a page that cannot be read is flagged rather than silently skipped.
In about a week of discovery we find out which folders hold the knowledge and what the access rule is for each source. Then we show you a demo running on your own documents. The first live workflow opens within 4-8 weeks; the range depends on document volume and the number of sources. Your own timeline is written into the discovery report.
Discovery week measures two things: the time it takes to find information and the number of people asked per question. Both are measured again after search opens and sit in the report next to the first figures.
Connected sources
Wherever the knowledge sits,that is where it is read from.
Each source is connected with its own access rights. Someone who cannot see a folder cannot see answers drawn from it either. The platform opens no new door for anyone.
Google Drive
Contracts, procedures and decks join the search the moment they change, and folder permissions are kept as they are.
Notion
Internal documents and guidelines sync instantly, and workspace permissions carry over.
Email attachments
Proposal, invoice and correspondence attachments are scanned every day; mailbox permissions are kept.
Slack channels
Decision and announcement history joins the search every day, and channel permissions carry over.
Your own database
Product, price and customer records are read every hour; access stays role-based.
File server
Technical drawings and archives are scanned overnight; a page that cannot be read is flagged rather than silently skipped.
Terms explained
- MCP serverMCP servers are programs that expose an outside system's tools, data and prompt templates to AI applications such as Claude or ChatGPT, using the Model Context Protocol (MCP). MCP is an open-source standard for connecting AI applications to external systems. Because every server describes itself in the same format, one server can be used by any application that supports the protocol.
- Large language modelLarge language models (LLMs) are AI models that learn from very large amounts of text and write new text by predicting what comes next. Assistants such as ChatGPT and Claude run on them. An LLM can answer questions, summarise, classify and draft. Its knowledge stops where its training data ends, and where it is unsure it can still produce an answer that reads well but is wrong.
Next step
We write the proposal after we have found the problem together
In a thirty-minute call we listen to how you work and talk through where time and customers are slipping away. If you do not need a digital fix, we will tell you that too.
30 minutes · no preparation needed
Discovery call
Root causefound together
Optionstalked through
Demoafter discovery
Next stepin writing
