The employee asks and gets an answer only from the documents he can open.
The company's knowledge lies in documents and in the heads of several people. The employee asks in plain words. The system searches only in documents to which this person has access and provides an answer with the source.
Tested on sample data, without customer data.
- answers from documents to which the questioner does not have access
- 0
- questions checked in the test, each with a different access card
- 750
- half of the local model's responses are ready faster
- 521 ms
Syntalith test results on 600 test documents and five fictitious accounts. Last checked: October 22, 2026.
Employees asked the same people over and over again.
Procedures, price lists and arrangements are in many places. A regular search engine doesn't understand the question and doesn't keep track of who can see what.
- 01
Just a question
Without knowing the file or folder name.
- 02
Source
The response indicates the document and its version.
- 03
Permissions
A confidential document is not responded to by a person without access.
From question to answer in four steps.
- 01
The employee asks
In plain words, after logging in.
- 02
The system checks access
He knows who this person is and what he can open.
- 03
He searches within her scope
The database returns only documents to which it has access.
- 04
Answer with a source
Or a short refusal when the source is missing.
What the system does and what stays with people.
Makes a system
- Looks for answers in the questioner's documents
- Provides the source and version of the document
- It records who asked and what they got
People decide
- Who has access to which document
- Which version of the document is valid
- What to do when two sources say different things
Statistics: company knowledge base with authorizations in 2026.
According to the Syntalith test (2026), none of the 750 answers came from a document to which the questioner did not have access.
Half of the local model's responses were ready in 521 ms or less.
With the local model, the saved mileage cost nothing for the external API.
Questions asked in different words than in the document were answered worse. Before implementation, we tune the search based on the customer's questions.
According to the Central Statistical Office, the average salary in the enterprise sector in September 2026 was PLN 9,266.48 gross. With employer contributions and 168 hours a month, an hour of work costs the company PLN 66.45.
Scenario: 600 questions per month
- By hand
- 600 × 25 min of search = 250 h of work, PLN 16,613
- Local model
- server with GPU and its maintenance instead of a token bill
- He stays with people
- permissions and selection of the applicable version
Estimate. Assumptions: 600 questions per month and 25 minutes of searching for answers are sample numbers. We price the server cost for the local model separately.
What stays with the owners of the documents.
- Access
- Access to a document is decided by its guardian.
- Conflicting sources
- The system does not select the version itself. He refers the matter for clarification.
- Read only
- The system reads the documents and does not change anything in them.
This is what the answer looks like to the employee.
Screens from the test on sample data.
01Zadaj pytanieOdpowiedź ma źródło. 
02Otwórz źródłoPracownik sprawdza odpowiedź w dokumencie. 
03Zobacz, co widzi inna osobaPoufny dokument nie trafia do odpowiedzi.
Are your employees also looking for answers in folders?
Describe your process to us. We will show the prototype on your process and your data. The analysis and prototype cost €0, and you know the fixed implementation price before the contract.
