Cortex is a personal AI memory system. It is built so the public repository stays generic and all personal/private data lives outside version control.

No real personal data in tracked files. That includes code, comments, docs, prompts, example files, READMEs, commit messages, and this file. The 2026-06 incident that motivated this policy happened because real client names and personal facts were baked into tracked code and docs; the gitignored files never leaked. The fix is to keep every personal value in the gitignored zone below.

Personal / instance-specific values live ONLY in:

• ~/.cortex/cortex.local.toml (gitignored): your sensitivity + category rules, instance settings, and ingest lists. Loaded at runtime by config_loader.
• ~/.cortex/secrets.toml (gitignored): API keys.
• memory/core/USER.md, OVERSEER.md, APP.md, and dossier/ (gitignored): your identity and context.
• The local databases under plugins/overseer/data/ (gitignored).

The repo ships *.example templates for these. Copy them, fill in your own values locally, and they will never be committed.

With no sensitivity config, imports gist-and-drop: capture structural signal, never store raw content. You opt IN to keeping raw by adding your own rules. The default off-box ceiling is internal, so confidential/restricted data never leaves the machine unless you explicitly raise it.

• As a user: open a private GitHub security advisory; do not file a public issue.
• As a maintainer of your own fork: scrub it from history with git filter-repo and rotate any exposed credential. A history rewrite does NOT un-compromise a key that was ever committed; treat it as compromised and rotate at the provider.

Report security or privacy issues via a private GitHub security advisory on this repository, not a public issue.