How Do I Use Private (Confidential) AI Models?

Add :private to the end of a supported model's name, and Bankr runs it inside a Trusted Execution Environment (TEE): secure hardware where your prompts and the model's replies are processed so that even the company running the model can't see them. This is called private (confidential) inference.

How to Use It

Pick a supported model and add :private to its name (<model-name>:private). That's the whole switch, and it works anywhere you choose a model: web terminal, CLI, or API. In an API request you can also send "private": true in the request body.

It fails closed: if you ask for a :private model and no secure enclave is available, the request fails with an error instead of quietly running without privacy.

What it protects: the enclave keeps your prompts from the model provider. Bankr's gateway still handles the request, so this isn't end-to-end encryption.

Which Models Support It

Some open-weight models support private inference. Proprietary models from providers such as Anthropic, OpenAI, and Google don't. The list changes as capacity comes and goes, so check the live list rather than relying on a fixed one:

• CLI: bankr llm models --private

• Web: the model lineup at bankr.bot/llm

Good to Know

• Responses are verifiable. Private responses include attestation headers that prove the work happened inside a verified enclave.

• Error 422 (confidential_unavailable): that model doesn't support private inference. Pick one from the live list.

• Error 503 (attestation_unverified): the gateway couldn't verify the hardware's security certificate, so it blocked the request to protect your data. Try again.

For API examples, Claude Code setup, and how to verify attestations, see Private Inference on docs.bankr.bot.