Knowledge
Transparency is central to us at Intric. When you upload a document to Knowledge, specific processes are in place to ensure your privacy and data sovereignty throughout the entire chain — from file upload to a searchable knowledge index.
The process from your file upload to an indexed, searchable document occurs through a secure interaction between the Intric platform (where your data is processed and stored) and the embedding model you have selected (e.g. Berget or OpenAI).
Step-by-step: How your data is handled
All transfers between Intric and its sub-processors occur over secure, encrypted connections.
Step 1 — User uploads a file to Intric
The user uploads a document (e.g. a PDF, DOCX, or text file) via Intric’s web interface or API.
Data sent to Intric’s server:
- The file’s complete binary content
- File name and metadata (size, file type)
The file is transferred over an encrypted connection (HTTPS) to Intric’s servers. No data is forwarded to external services at this stage.
Step 2 — Intric processes and stores the document
Intric extracts the text content from the file, stores the original securely, and prepares the content for indexing. All processing in this step happens internally on Intric’s servers.
What happens:
- Text content is extracted from the document (PDF, DOCX, etc.)
- The raw original file is stored in Intric’s object storage (S3) in Sweden
- Extracted text, metadata, and page structure are saved in Intric’s database (PostgreSQL)
- The text is split into smaller segments (chunks) to enable semantic search
Nothing leaves Intric’s server at this stage — all processing and storage happens internally on the platform.
Step 3 — Intric sends text content to the embedding model
In this step, the text content from each chunk is sent to the configured embedding model to be converted into searchable vectors.
Data sent from Intric’s server:
- The text content of each chunk (actual document text — not hashes or anonymized data)
What happens at the embedding model: A numeric vector (a list of floating-point numbers) is generated based solely on the text provided. The embedding model has zero context about the user’s or organization’s identity.
Step 4 — Response to Intric
The embedding model’s response (the vectors for each chunk) is sent to Intric’s server, which receives and stores the information encrypted in its database.
Data sent from the embedding model to Intric:
- The numeric vectors for each text chunk (for semantic search)
Immediately after the result is sent back to Intric, both the user’s input and the generated output are deleted from the embedding model’s server.
Step 5 — User can use the file in Knowledge
The document becomes searchable and can be used in your knowledge collections in Intric.
Data stored on Intric’s servers:
- Extracted full text and page structure
- Text chunks with associated embedding vectors
- File metadata (name, size, file type, timestamp)
The embedding vectors are stored together with the original chunk text in Intric’s database in Sweden, hosted by a Swedish sub-processor.
Data sharing and privacy
To protect your and your organization’s privacy, we apply the principle of data minimization. This means the sub-processor only gets access to the content absolutely necessary to perform the task — no user identity ever leaves your infrastructure.
We have strict zero data retention clauses in all our contracts with language model sub-processors. This guarantees that content sent to the embedding model is never saved by the provider after the vectors are returned, nor is the information used to train their AI models.
In the table below, you can see exactly what data is sent to the sub-processor and what does not leave Intric’s servers.
| Sent to the embedding model | Not sent to the embedding model |
|---|---|
|
|
Data retention and deletion
All file storage and indexing happens on Intric’s infrastructure in Sweden — no external service stores your documents, extracted text, or vectors.
When a document is deleted from Knowledge, all parts are removed — the original file, extracted text, chunks, and vectors. In the audit log, administrators can see when documents are uploaded and deleted, and by whom.