Cortex by Arc

Your data

Where it lives, where it goes, and what Arc never sees.

On your machine

  • Cortex reads your agent transcripts and never changes them.
  • It writes only to one folder, .cortex/, on your disk.
  • The compiled record (what was said, what it means now, what changed) lives there and nowhere else.

To your model provider, when you compile

  • Compiling sends the relevant turns to a language model, under your own subscription or API key.
  • You choose the provider. Point Cortex at a local model and nothing leaves your machine.
  • Nothing is sent without your say: Cortex estimates the cost, asks, and stops at a hard cap.
  • Your provider's own privacy terms apply to what you send it.

To Arc

  • By default, nothing. The preview key ships inside Cortex and is checked on your machine.
  • Only if you sign in to an Arc account, or refresh an account key, Cortex sends four fields: your account id, the product name, a device fingerprint and the Cortex version. Never a conversation, a file, or anything from the record.
  • There is no usage tracking in Cortex.

What Cortex never does

  • It never uploads your conversations, your files or your record to Arc.
  • It never acts in your world. It changes only its own record, and you can inspect and undo every change.
  • It never locks you out of your own work. If a license lapses, your continuity keeps working.
  • It never switches off remotely. There is no kill switch.

The full text is the Privacy Notice. Every statement on this page is checked against the code before each release.