Suger does not use your documents or conversations to train AI models. Knowledge base content is searched at query time to ground an agent’s answer and return cited sources. The distinction between those two things — training on data and retrieving from it — is the one worth understanding, because it is what makes the answer verifiable rather than a promise.
This is the question that stops rollouts.
Not because the answer is bad, but because it is usually the fourth question asked and the first one nobody can answer precisely. Someone in security asks it, the person championing the tool says “no, I’m fairly sure it doesn’t,” and the pilot pauses for two weeks while a contract gets read.
It is worth being able to answer it exactly, and worth understanding the mechanism underneath, because “we don’t train on your data” is a sentence many vendors say and not all of them mean the same way.
The answer
No. Suger does not use your documents or conversations to train AI models. Knowledge base content is searched at query time to ground an agent’s answer and return cited sources.
Why the mechanism matters more than the promise
Training and retrieval are different operations, and knowing which one is happening tells you what is possible rather than what is promised.
Training adjusts a model’s weights. Information from the training data becomes part of the model, distributed across it, unattributable and not removable. If your document trained a model, you cannot take it back out, and you cannot tell whether it influenced any given answer.
Retrieval leaves the model unchanged. Your documents are chunked, embedded and indexed. When a question arrives, the relevant passages are found and supplied alongside it. The model reads them the way it reads the question — as input for that one answer, then it is done.
The consequences follow directly:
| If it trained | Because it retrieves | |
|---|---|---|
| Removing a document | Impossible | Removes it from future answers |
| Knowing what an answer used | Impossible | The answer cites its sources |
| Reaching another customer | Structurally possible | Not a path that exists |
| Correcting a mistake | Retrain | Edit the document |
That last row is the practical one. Because retrieval is a lookup, a wrong answer caused by a wrong document is fixed by fixing the document — and the citation tells you which document to fix.
What this means for the documents you upload
The reason this question matters is that the useful version of this tool requires your real material. A knowledge base of public marketing pages helps nobody. The value comes from the pricing policy, the escalation runbook, the contract terms — precisely the documents people hesitate over.
The retrieval model is what makes uploading them a normal decision rather than an irreversible one:
Nothing is absorbed. Remove a document and it stops appearing in answers.
Every answer is checkable. Because retrieval returns sources, you can audit what an answer was built from instead of taking it on faith. Why AI answers need citations covers what that changes about how quickly teams trust the output.
Access is still yours to control. Knowledge bases use role-based access, organisation knowledge bases require explicit per-base sharing, and user knowledge bases stay private unless shared. Uploading something does not publish it — controlling which agents see which documents covers the scoping.
What to ask any vendor
If you are evaluating more than one tool, these four questions separate them faster than a security questionnaire:
“Does customer content train your models, and does that include a version with a checkbox I could accidentally leave on?” The second half matters. A default-off setting is a different answer from a structural no.
“If I delete a document today, what happens to answers tomorrow?” A retrieval system has a clean answer. A training system has a complicated one.
“Can you show me what a given answer was built from?” Citations are the observable proof of the architecture. A system that cannot show its sources is asking you to trust a description of itself.
“Where does the content sit, and who can reach it?” Retrieval means your documents remain your documents in an index — so the access model around that index is the thing to examine.
Frequently asked questions
Is my data used to train AI models? No. Suger does not use your documents or conversations to train AI models. Knowledge base content is searched at query time to ground an agent’s answer and return cited sources.
What is the difference between training and retrieval? Training changes a model’s weights permanently. Retrieval leaves the model unchanged and supplies relevant passages alongside a question, for that one answer.
What happens if I delete a document? It stops being retrieved, so it stops appearing in answers. Nothing was absorbed into a model that would need removing.
Can I tell what an answer was based on? Yes. Retrieval returns cited sources, so an answer can be traced to the passages it used rather than accepted on trust.
Could another organisation see our documents? Knowledge bases use role-based access. Organisation knowledge bases require explicit per-base sharing, and user knowledge bases remain private unless you share them.
Why does the mechanism matter if the policy already says no? Because a policy is a promise and an architecture is a constraint. Retrieval makes deletion effective and answers auditable, which a policy alone cannot.
Takeaways
- Suger does not train models on your documents or conversations. Content is retrieved at query time.
- Training absorbs data permanently; retrieval looks it up for one answer and leaves the model unchanged.
- Because it is retrieval, deleting a document actually removes it from future answers.
- Citations are the observable proof of the architecture, not a convenience feature.
- Ask vendors what happens when you delete a document, and whether they can show an answer’s sources.
- A policy is a promise; an architecture is a constraint. Prefer the one you can verify.
The documents worth grounding an agent in are the ones people hesitate to upload. See how knowledge bases retrieve from yours without absorbing them.
Sources
Primary sources for the platform rules cited above. Last verified August 16, 2026. Cloud providers change fees, eligibility, and program terms without notice — check the source before relying on a figure.
- Suger Insulin docs: Knowledge Base — That documents are chunked, embedded and indexed for hybrid search, and retrieved at query time to ground an answer with cited sources — the mechanism this post describes
Stay Updated
Get the latest Cloud GTM insights, product updates, and marketplace strategies delivered to your inbox.