Skip to main content
Different commands protect different things. This page is the command-by-command map of what your harness provider sees and what it does not. The key distinction is the trust boundary: your harness can see some Nito prompts and not others, independent of the model’s privacy level.

​
At a Glance

​
Ask

ask runs inside your current harness conversation. That means your harness can see the ask prompt, the same way it sees anything else you type into that session. Its usefulness is model access: it reaches a model through Nito without switching your harness’s model. Use it when the sensitivity of the prompt does not require keeping it from your harness. The Nito-supported model’s own privacy level still applies to the call once it reaches Nito, following the model you chose.

​
Private

private is the command built for the trust boundary. A private turn is not sent to your harness’s own model provider. Be precise about what that covers:
  • What it protects: the turn is kept out of your harness’s upstream model provider.
  • What it does not do by itself: it is not a local sandbox and does not, on its own, stop local disk writes. A no-prompt private session is the incognito form that avoids persisting harness transcript and prompt history; the options and their honest limits are on the Private page.
Pair private with a model whose level matches how sensitive the content is: the command controls what your harness sees, and the model controls the level of the call at Nito.

​
Session

A session is a whole harness session powered entirely by a model through Nito. Every model turn is Nito-routed rather than going to your harness’s own provider, for as long as the session runs. Its purpose is sustained work on a model through Nito, not the throwaway secrecy of a private turn; the two differ on persistence, which the Session page compares directly.

​
Fusion

fusion sends one prompt to two or three models in the Nito catalog. Two things decide its privacy. Where you run it. Fusion inherits the privacy of whatever it runs inside.
  • Inline, in your ordinary harness conversation, it behaves like ask: your harness sends the prompt and the answers upstream, and they sit in its normal history.
  • Inside a private session or a Nito-routed session, it does not. The run stays out of what your harness sends, and it keeps whatever that session keeps, which is nothing in a no-prompt private session and your history in a session.
The levels of the participants. Anonymous and Private models can mix, and the run lands at the least-private level among them, so a single Anonymous participant pulls the whole Fusion to Anonymous. A Confidential participant cannot be mixed with a lower level at all; that combination is refused rather than quietly downgraded.

​
Context

context is not a model call and does not change what your harness sees. It controls how much of your prompt’s surroundings travel with an ask: by default only your prompt is sent, and turning context on lets a short generated summary of the task go along. It is prompt-only by default for a reason, so treat turning it on as a deliberate choice.

​
The Rule of Thumb

Two questions, two controls:
  1. Should my harness see this prompt? If not, use private (or a session), not ask.
  2. How private must the call itself be? Choose a model at the level that matches, since the level follows the model.