Every external system and open standard Sentinel ingests from or plans to. Active integrations link straight to their admin surface; scaffolded items point at the committed code in the repo.
Primary ingest for LMR networks. Full M1 (register, authenticate, keepalive) tested against a mock gateway; M2-M4 scaffolded with codec + tests. Customer onboarding doc + disconnect-cause enum + BS422 alarm decoder shipped.
Sentinel ingests Tait vehicle events through a pre-built AppBuilder flow running on each Tait Unified Vehicle — no infrastructure access required. The flow forwards GPIO (emergency pin), text messages, emergency-mode, and presence events over MQTT to the Sentinel sim. Deployable as soon as a customer with Tait UV radios runs the flow. Ref: MNE-00034.
The DAMM-equivalent path for Tait infrastructure. Full call-event stream from the trunked controller, audio forwarding, presence, GPS. Adapter scaffold (`apps/adapter/src/sentinel_adapter/tait/enable/`) commits the module layout + spec anchors; implementation is a 4–6 week fill-in once the Northbound API reference lands from Tait's Partner Portal.
Mission-Critical Push-to-Talk over 5G is the standards-based successor to TETRA/DMR. OCUDU standardizes the open-source CU/DU compute this runs on. Adapter scaffold (`apps/adapter/src/sentinel_adapter/mcptt/`) commits the module layout + 3GPP spec anchors; implementation is a fill-in once a pilot lands.
Sentinel's architecture already fits the rApp mold: non-real-time intelligence consuming CU telemetry (O1 / A1). Helm chart + SMO onboarding manifest planned for when the first OCUDU-conformant pilot materializes.
Telewave ↔ NIST collaboration. Mission-Critical Voice Quality-of-Experience archive populated from the open-source `mcvqoe` Python suite. Trend page, KPI board, compliance PDF report, and QoE-threshold SLA rules all live today.
On-device inference for incident summarization, semantic search, and AI location extraction. No cloud dependency — every customer deployment is air-gap capable.
Voice-to-text for System Test + real-call transcription. CPU default, GPU-colocatable for AI-RAN deployments.
Whisper + Llama + embedding inference shares idle GPU cycles on the carrier's RAN baseband servers. Material TCO argument for the first MNO pilot — no dedicated Sentinel rack required.
Live STRESS badge on transcript lines — surfaces elevated stress before an explicit duress word is spoken. Heuristic today; pyannote / prosody-model upgrade path documented.
Unique Sentinel capability: pair per-UE RAN telemetry (packet loss, SINR, attach-delay) with the voice-level SLA events we already produce. No LMR product can say "this radio had 42% packet-loss during the mayday".
Map surface can run fully offline against a bundled `.pmtiles` file — air-gapped customer-site deployments stay on-brand without external tile-server dependencies.