SENTINEL AI logo SENTINEL AI TAHAI Web Services
SENTINEL AI

Local-first boundaries

Data Handling

The data posture is built around user-approved sources, matter privacy, no-training defaults, and reviewable provenance.

User-approved roots

Local folder access should be limited to explicit allowlists.

Separate outputs

Generated organized files should be written outside original input roots.

Source manifests

Every source can carry path, hash, date, class, parser status, and review state.

No-training default

Private matter data should be used for retrieval and inference only unless explicitly opted in.

Retention labels

Traces, assets, and documents need retention and deletion policies.

Redaction

Sensitive prompts, traces, and exports should support redaction and review.

Private work stays privateLocal-first design is a practical privacy advantage.
Provenance mattersA grounded answer is only as strong as its source record.