Memory and privacy

My Conversations Are Sent to an LLM Provider — What Does That Actually Mean?

> ANSWER[AEO]

"The AI uses your conversation as context" means that conversation, or a structured version of it, is sent to a third-party company's servers. That is how essentially every current LLM product works, but it is a real data-sharing relationship, and the terms vary between providers in documented ways. In Closer's shipped configuration two companies receive your data: DeepInfra handles audio and transcript-derived text, and OpenAI handles embeddings and the memory engine's fact extraction.

Is "the AI uses my conversation as context" just a polite way of saying my data goes to another company?

Largely, yes, and the technical framing should not obscure it. When a product generates a response using your conversation history, that history has to be included in the request sent to whichever provider runs the model — so the provider's infrastructure receives it, at least for the duration of processing.

Independent research comparing major AI platforms found meaningful, documented differences in what happens next: whether conversation content becomes training data, what categories of information get shared onward with other third parties, and how transparent each platform actually is about either.

Does that make every AI product equally risky?

No, and treating them as interchangeable hides a real difference. The same comparative research found platforms differ in what they share and with whom — some pass contact details, location or app interaction data to third parties by default, others are more restrictive.

That is why "the AI provider has my data" is too coarse a statement to act on. The useful question is what the specific provider's terms say about retention, training use and onward sharing — not whether data transits their servers at all, which for current architectures it necessarily does.

What is worth checking about any product that uses an LLM provider?

Who receives your data in Closer, specifically?

Naming it is the mitigation, so here is the shipped configuration:

What does Closer itself store, and how well protected is it?

The parts people usually assume, stated plainly instead:

Key takeaways

  • > Using a conversation as LLM context means sending it to a third-party provider's servers — that is inherent to the architecture, not a shortcut.
  • > Providers differ measurably in retention, training use and onward sharing, so "an AI company has my data" is too coarse to act on.
  • > In Closer, DeepInfra receives audio and transcript-derived text; OpenAI receives text for embeddings and memory fact extraction.
  • > Closer stores no audio, only a hash used as a cache key — but transcripts and summaries are stored.
  • > Application-level encryption covers SMTP passwords and Google refresh tokens, not meeting content; tenant isolation is enforced in application code, not by database row-level security.

Frequently asked questions

When my conversation becomes context for an LLM, does that mean a third company now has my data?

In most current AI product architectures, yes, at least transiently: a request containing your conversation context is sent to the LLM provider's servers to generate a response.

  • What varies enormously is what that provider does after receiving it.
  • Used only to generate that one response and then discarded, retained for some period, or used to train future models — those are very different outcomes.
  • It is worth checking specifically rather than assuming.
  • Why a memory system needs to send context at all: how AI memory actually works under the hood.
Does every AI product share the same amount of data with third parties?

No, and the differences are documented, not marginal.

  • Independent research comparing major AI platforms across data usage found real variation.
  • Platforms differ in whether they use conversations for model training by default.
  • They also differ in what categories of data — location, contact details, app interaction data — get shared with third parties at all.
What should I actually check before trusting a product that sends my conversations to an LLM provider?

Three concrete things, and a general privacy policy statement alone does not answer any of them.

  • Whether the LLM provider's terms permit training on your data by default, and whether that can be turned off.
  • How long the provider retains the raw request data, as distinct from the product's own separate storage.
  • Whether the arrangement is covered by a data processing agreement that legally restricts what the provider can do with it.
  • Closer makes no claim here about provider training or a data processing agreement — those are contractual questions — and it holds no security or privacy certification.
Which providers does Closer send data to, and what do they receive?

Two, in the shipped configuration.

  • DeepInfra receives the audio and the transcript-derived text: transcription, summaries, knowledge-graph extraction, prep briefs, Ask AI, roleplay and text-to-speech all run there.
  • OpenAI receives text for embeddings and for the memory engine's fact extraction.
  • Anthropic is declared in the provider registry but is not wired to any pipeline unless an operator reconfigures it, so it receives nothing by default.
  • What Closer keeps on its own side, and what deleting removes: can you actually delete something from the memory.
Is my meeting content encrypted at rest in Closer?

Not by the application. Transcripts, summaries, memories and knowledge-graph facts are stored without any application-level encryption layer, so they get whatever encryption the database provider supplies underneath and nothing more.

  • Closer applies its own AES-256-GCM encryption to exactly two things — SMTP passwords and Google refresh tokens — because those are credentials that would let someone act as you.
  • Meeting content is not in that set.
  • Tenant isolation is enforced in application code, not by Postgres row-level security.
  • We would rather say that plainly than let "encrypted" do work it hasn't earned.

Sources