Fact Engine
“I think the DB is down” never becomes root cause. Beliefs stay hypotheses until evidence and a human confirm them.
Sentinel joins the war room like a teammate. It separates what the team knows from what it believes, tracks who owns what, flags contradictions — and never touches production without a human.
“I think the DB is down” never becomes root cause. Beliefs stay hypotheses until evidence and a human confirm them.
Two responders disagree — Sentinel flags it and asks for verification instead of picking a side.
“Someone check the deploy” gets an owner, a priority, and a follow-up when it goes quiet.
Rollbacks, restarts, failovers are proposed — never executed — until a commander approves.
Every belief renders what supports it, what contradicts it, and its live confidence.
Six scores computed from state, never asserted by the model. Low root-cause confidence stays low.
Every fact, action, decision, conflict and tool run becomes an immutable timeline event.
Impact, findings, decision ledger, unresolved risks — generated the second the incident resolves.
Responders talk in the Agora RTC voice room. Sentinel joins as a participant.
→Agora's ares ASR converts speech; RTM delivers speaker-attributed transcripts.
→The engine + Gemini extract facts, hypotheses, actions, conflicts as typed ops.
→Safe tools auto-run. Production actions wait at the human approval gate.
→WARN · ASK · SUMMARIZE · PROPOSE — voiced into the room via managed TTS.
ASR: AGORA ARES · LLM: GEMINI 2.5 FLASH (CUSTOM VENDOR → OUR ENGINE) · TTS: AGORA-MANAGED · TRANSCRIPTS: RTM DATA CHANNEL
THEN RUN THE PAYMENT-OUTAGE DEMO — OR JOIN VOICE AND TALK