> ## 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. # Privacy Levels in Depth > The four privacy levels as a ladder, and how the level follows the model you pick. ## Each Level Builds on the Last Read the four as a ladder. Each rung adds a stronger guarantee and, in exchange, narrows the models and features available to you. Climb only as high as the call needs. | Level | How you reach it | What it adds | | :--------------- | :------------------------------------ | :---------------------------------------------------------- | | **Anonymous** | Pick a model whose level is Anonymous | Your identity detached before the provider sees the request | | **Private** | Pick a model whose level is Private | Upstream provider held to zero data retention | | **Confidential** | Pick a model that offers Confidential | Sealed execution with a verifiable attestation | | **Encrypted** | Coming soon to the plugin | Client-side encryption to the sealed environment | ## The Level Follows the Model There is no privacy flag on your command. The level is part of the model's name. Anonymous and Private are a model's own level, and its ID is just provider and model. A model that also offers a stronger level gets a separate entry in the catalog, whose name ends in `:confidential` or `:encrypted`. ```text theme={"theme":{"light":"github-light","dark":"github-dark"},"languages":{"custom":["typescript","python","curl"]}} z-ai/glm-5.2 the Private entry z-ai/glm-5.2:confidential the Confidential entry z-ai/glm-5.2:encrypted the Encrypted entry, coming soon ``` One underlying model, several entries, one level each. You pick the level by picking the entry, and you never add the marker yourself: it is part of a name the catalog gives you. See [Model ID Convention](/models/model-id-convention). Browse with the [`models`](/commands/choosing-a-model) command, then set your model with `model` or `--model`. Set it once and it stays until you change it, and the level comes with it. ## Why a Model Reaches Only Certain Levels No model offers every level, and the reason is what can be enforced on it. * **Open-weight models can be served with zero data retention enforced, or run inside confidential hardware.** That is what makes everything above Anonymous possible: retention switched off upstream, sealed execution, an attestation you can check. They start at Private and climb from there. * **Frontier models always route through their provider.** What can be done is detach your identity from the request: the provider gets it from Nito, without your identity attached. That is Anonymous, and it is as far as a frontier model goes. So the ladder is not a quality ranking. It reflects what can be enforced on a given model. ## It Never Silently Downgrades If a call cannot be served at the chosen model's level, it **fails with a clear message** rather than dropping to a weaker one. A guarantee that quietly weakens is worse than no guarantee, so the call is refused instead. ## The Four Pages The frontier rung. Your identity is detached before the provider sees the request. Privately served models with no upstream retention. Sealed execution with an attestation you can fetch and verify. The top rung, adding client-side encryption to the sealed environment. Coming soon to the plugin.