Questions, answered honestly

Yes. Walter's brain is Claude Code, Anthropic's agent, and you bring your own Claude plan or Anthropic API key, billed to you by Anthropic. We never resell, include or share Claude access, and our prices don't cover it. Light use fits a Claude Pro plan; an always-busy household may hit its limits and want a bigger plan or an API key with a budget.

No. You sign in through Anthropic's own sign-in for Claude Code, on Walter's screen with a keyboard plugged in, and Walter only asks Claude Code whether it's signed in. If you use an API key instead, it's stored only on your box and handed to Claude Code, never shown or logged.

No. Walter is an independent product that runs Claude Code with your own Claude account.

It listens for its name on the device. Nothing is sent anywhere until you say its name (or reply right after it speaks, or talk in the first minute after it starts up). There's a mute key, and it mutes itself during quiet hours overnight.

The words of your requests go to Claude under your account. Cloud speech, Google and remote access only if you turn them on. See the full map.

Partly. The face, the wall board, reminders and on-device speech keep working, but thinking needs the internet, so it can't answer questions while offline.

A 64-bit Intel or AMD laptop or mini PC that can be wiped, a screen, and a microphone and speaker (a USB speakerphone is best). We recommend 4 CPU cores and 8 GB of memory. Two-core machines work, but the voice can stutter when it's busy. Details

Not yet. The beta installs its own Linux system on a dedicated 64-bit PC.

The install itself is a USB stick and a wait. Claude sign-in is a few minutes. Connecting Google and remote access take extra steps in the beta, and we help beta households with them.

It's a real trade-off, so Walter gets a computer of its own. Today the AI can run commands on that machine without asking; a more restricted permission mode is planned before public launch. Admin access to the machine is off unless you turn it on. Anything with real-world consequences (spending money, sending messages for you, locks and alarms) needs your spoken yes. Web pages and emails can try to trick AI agents, so connect only the accounts you need.

It updates overnight to tested versions and rolls itself back if an update fails. Cryptographically signed releases are being added for the beta. Encrypted memory backups are planned.

The owner must be an adult. It's built for a household, but it isn't a children's product, and nobody should be enrolled for face or voice recognition without their consent (or a parent's).

Honestly: English only; one household per box; no smart-home control out of the box; it can't make phone calls or send email on its own; music services with copy protection (like Amazon Music) don't play; no Raspberry Pi or Mac support; one-tap remote access setup and encrypted backups are still planned. It's an AI, so it can be wrong; check anything important.

The code runs on your own machine and you can read it there. The license for public release is still being decided.

Free during the beta. After that, the software stays free and the optional Cloud Link is planned at $[5] a month or $[49] a year (placeholders). Your Claude plan is separate.