Does Privatemode store my source code or prompts?

No. All processing happens inside hardware-isolated confidential computing environments. Your prompts and code context are encrypted before leaving your machine, decrypted only inside a hardware-isolated environment whose memory stays encrypted, and discarded after the response is generated. Prompts and responses are never stored or used for model training; only operational metadata such as token usage, timestamps, and request status is kept for billing and monitoring.

How does encryption affect response speed?

Confidential computing adds a small overhead. Response times are comparable to conventional AI services. You can also enable prompt caching with the --sharedPromptCache flag on the proxy to speed up repeated context windows.

Can I verify that the server is actually running in a confidential computing environment?

Yes. The Privatemode proxy performs remote attestation before every session, verifying the server's integrity against known cryptographic hashes. The proxy and server code are open source with reproducible builds, so you can audit exactly what runs on the server side.

Does Claude Code work the same way through Privatemode?

Claude Code's core workflow is fully supported: multi-file editing, codebase-wide search, shell command execution, and project reasoning. The Privatemode API is Anthropic-compatible, so Claude Code communicates with it natively. The only difference is the model: instead of Anthropic's Claude models, you use open-weight models running inside confidential computing.

Using Confidential AI with Claude Code

Claude Code works by reading your repository: it greps, opens files, and edits across the tree to answer a single prompt. You keep using Claude Code, but the model behind it changes: instead of Anthropic's Claude models (Sonnet, Opus, or Fable), your prompts go to open-weight models running inside confidential computing, so all of that context stays encrypted in memory while it is processed.

Introduction

The challenge

Claude Code decides what the model sees

Claude Code decides for itself which files to open, so the model sees large parts of your codebase, not just the snippet you asked about. With a conventional provider, all of it is processed in plaintext on that provider's servers.

The solution

A different model behind the same Claude Code

The Privatemode proxy serves the Anthropic Messages API, so Claude Code talks to it unmodified: one base URL and a handful of environment variables, reverted just as quickly. Behind that URL, open-weight models answer instead of Anthropic's Claude models, while Claude Code itself works as before. The proxy verifies the service through remote attestation and encrypts each request on your machine, and the model then runs in a hardware-isolated environment whose memory stays encrypted.

Benefits

Why use Privatemode AI with Claude Code?

Minutes to set up, easy to revert

End-to-end confidentiality

Learn how

Open-weight models, not Claude models

See models

Free tier available

Quickstart

Set up Claude Code in one prompt

Paste one prompt into Claude Code and it sets itself up on Privatemode, guided by the official docs and reversible at any time.

Claude Code

Anthropic's agentic CLI for your codebase. Pick a mode: try Privatemode in a single project first, or make every session confidential right away.

Try pasting this into Claude Code. It sets up this project only.

›Set up Privatemode as the confidential model backend for Claude Code - in this project only, so I can test it and cleanly undo it.

1. Get your API key from the dashboard
2. Paste the prompt into Claude Code
3. Test it: say "undo" to revert everything

No risk: Only local project settings are modified and a local proxy is created. Your overall setup stays untouched.

Prefer to do it by hand? See the manual steps

Manual setup

Or set it up by hand

The prompt above follows the setup guide in the Privatemode docs. Follow it yourself if you would rather not hand the configuration to an agent, or if Claude Code has no working provider yet to run the prompt with. Requires Claude Code v2.1.242 or later.

Get your API key

If you don't have a Privatemode API key yet, you can generate one for free here.

Set up Claude Code

The cookbook in the Privatemode docs walks you through running the local proxy and pointing Claude Code at it in ~/.claude/settings.json, including the model list that keeps the /model picker and sub-agents on Privatemode models.

Cookbook

FAQ

Frequently asked questions about Claude Code with Privatemode

Privatemode offers a range of models running inside confidential computing environments, with your data encrypted in memory during processing. See the model overview.