The alpha line is supported on a best-effort basis until the first stable release.

Please do not open public issues for vulnerabilities, leaked secrets, or privacy-impacting bugs. Email the project maintainers or use the private advisory channel once the GitHub repository is published.

• Telemetry is disabled by default.
• Demo model output is deterministic and local.
• Real provider credentials must not be returned by the API.
• Real model credentials should stay inside the user's own Agent gateway or platform Agent environment, not inside Study Anything.
• Logs and traces should use user hashes and redact secrets.
• Generate .env with python3 scripts/setup_env.py; never deploy placeholder values from .env.example.
• APP_ENV=production should pass python3 scripts/check_env.py --strict before public exposure.
• Local backup manifests are checksum verified and reject unsafe file paths.
• Sync restore preview must remain non-destructive and return only counts, hashes, conflicts, and warnings rather than plaintext learning data.

Run the current security recovery verifier before public release or risky local upgrades:

python3 scripts/verify_security_recovery_hardening.py

The verifier covers backup manifest tamper detection, path traversal rejection, wrong-passphrase redaction, restore-preview privacy, and the disabled destructive restore API boundary.

The public, metadata-only audit kit is under security/audit/. Generate and verify the portable package with:

python3 scripts/generate_external_security_audit_pack.py --check
python3 scripts/verify_external_security_audit_pack.py --check

ready_for_independent_audit means an external human reviewer can begin from a pinned commit. It does not mean an audit or penetration test has run or passed. Private reproduction details must use the vulnerability reporting channel, not the public audit-tracking issue.

See docs/security.md for the local-first security model and operational recovery expectations.