> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trynito.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Data Retention and Visibility

> Nito does not store the contents of your calls. What it does keep, and who can see what across your harness and the model provider.

Two questions decide the privacy of any prompt: what is **kept**, and who can **see** it. This page answers both for the plugin.

## What Nito Keeps

**Nito does not store the contents of your calls.** Your prompts, the responses, reasoning, tool payloads, and attachment contents are used to serve the request and are never persisted. Conversation history, where it exists, lives on your machine.

What Nito keeps is operational metadata, the minimum needed to run and bill the service.

This holds at every level. [Private](/privacy/levels) adds the upstream provider being held to zero data retention on top of it.

## Who Can See What

Retention is separate from visibility. Even when Nito keeps nothing, the question of who *sees* a prompt in the first place depends on which command you used.

| Party                                  | Sees your prompt when                                                               | Does not see it when                                                                   |
| :------------------------------------- | :---------------------------------------------------------------------------------- | :------------------------------------------------------------------------------------- |
| **Your harness** (Claude Code / Codex) | You use [`ask`](/commands/ask-and-continue), which runs in the harness conversation | You use [`private`](/commands/private) or a Nito-routed [`session`](/commands/session) |
| **Nito**                               | Any Nito command (that is what serves the model call)                               | Not applicable                                                                         |
| **The upstream model provider**        | It serves the call, subject to the model's [level](/privacy/levels)                 | At Private and above, it is held to zero data retention                                |

The takeaway: keeping your prompt away from your **harness** is about which command you choose, while keeping it from the **provider** is about the model's level. See [What Each Command Protects](/privacy/what-each-command-protects) for the command side, and [The Four Levels](/privacy/levels) for the level side.

## Confirming It Yourself

For the strongest levels, you do not have to take retention behavior on trust: a Confidential call produces a signed attestation you can verify. See [TEE Attestation](/privacy/tee-attestation).