Skip to content

Air-Gap Mode

Air-Gap Mode is a Sovereign feature. It replaces the ocultar cloud NER with a bundled local Small Language Model (SLM) so that PII detection never requires an internet connection.

Go to Settings → AI NER and switch from ocultar cloud to Local SLM (Sovereign). Ki! will load the bundled model on the next prompt — the first load takes 2–3 seconds; subsequent loads are instant.

In air-gap mode, the masking pipeline is entirely on-device. No data leaves your machine at any point during a session.

BehaviourCloud modeAir-gap mode
NER backendocultar cloud sidecarBundled local SLM
Internet required for maskingYesNo
Licence checkMonthly hash check90-day grace period
LLM API callsStill requires internet unless using OllamaStill requires internet unless using Ollama

To go fully offline, combine air-gap mode with a local LLM provider such as Ollama configured in Settings → API Provider.

Ki! checks the licence server once every 30 days. If the check cannot reach api.getki.ai (offline, firewall, DNS failure), Ki! starts a 90-day grace period. Sovereign features remain fully active during this period.

The grace period resets each time a successful licence check completes. For permanently air-gapped environments, the grace period effectively never expires as long as the licence is renewed before the 90-day window closes.

Grace period start: the first failed licence check after a previously successful one.

Grace period end: 90 days after start, or a successful licence check, whichever comes first.

If the 90-day grace period expires without a successful licence check:

  • Sovereign features (local SLM, batch scrubbing, policy sync, compliance bundle) are disabled.
  • Community features (chat masking, document scrubbing, vault viewer, egress log) continue without interruption.
  • A banner in the app prompts you to restore connectivity and re-verify the licence.

For enterprises with no internet access:

  1. Enable air-gap mode on all seats.
  2. Configure a local Ollama instance as the AI provider.
  3. Set up a proxy that allows outbound HTTPS to api.getki.ai on port 443 only — this is the minimum needed for the licence check. Alternatively, plan for a quarterly “check-in” window where seats briefly connect to perform the licence verification.
  4. Distribute the ki-policy.json via an internal file server — no external URL needed.