The source stays where they put it.

• Reads source documents. A court filing, a leaked memo, a transcript, a regulatory submission. Open one, ask anything, cited back to the page in the source.
• Summarises long records. A long deposition, an indictment, a regulatory filing, all reduced to something a reporter can use without uploading the source anywhere.
• Translates source materials. When the source is non-English and the document can’t go to Google Translate or DeepL.
• Spreadsheet Q&A. Donor lists, lobbyist filings, campaign-finance exports. Open in Excel or CSV, ask questions in plain language, parsed locally, never uploaded.
• Edits drafts without exposing them. Tracked-change edits to a DOCX, using the actual source detail, without that detail entering a vendor’s prompt log.
• Redacts before publication. One-pass PII redaction so the published version doesn’t expose the source or third parties.

Muet works on one document at a time, locally. For corpus-wide search across thousands of files in a leak, you’ll still want a dedicated tool. Muet handles each document you read.

• Vendor logs. Every prompt and every uploaded document sits in the vendor’s system. Abuse-review staff can read it. Security investigations can read it. Acquirers can.
• Legal demands. The May 2025 court order in NYT v. OpenAI compelled OpenAI to retain “deleted” ChatGPT messages, including for Plus, Free, Team, and most API users. Whatever the vendor’s contract said, the court overrode.
• Cross-border exposure. A US-hosted vendor is reachable by US process even when the source isn’t. A reporter in one country using a vendor in another routes the document through both jurisdictions.
• The shield-law gap. Most reporter-shield protections cover direct compulsion of the journalist. They don’t automatically extend to a third-party vendor that holds the same material.

When the AI runs on the reporter’s own Mac, none of those exposures arise. There is no vendor log, no third-party storage, no cross-border routing, no parallel custodian.

• Investigative reporting on whistleblower-supplied materials.
• Cross-border reporting where the source faces real risk.
• Court-record analysis where the journalist has the documents but the public doesn’t.
• Reading individual FOIA returns without uploading them.
• Editorial review of pre-publication drafts that name real people in real situations.
• NGO and human-rights research with the same threat model as journalism.
• Documentary and book research where the source documents have to stay on the writer’s laptop.