Agents begin responding then stop mid-response ("virtual silence"); effectively unusable for ~24h.
1. The stall is a model/provider issue, not infrastructure.
api.deepseek.com right now: 1.1–1.4s per call, all returning OK2. No agent traffic reached the app in the last 24h.
0 "Agent chat request" log lines — the chat endpoint was never hit3. Root cause for #2: partner subdomains rejected by ALLOWED_HOSTS.
window.location.origin, so on jorja-essex.gp.veripath.co.uk / maple-surgery.gp.veripath.co.uk they call those hostsALLOWED_HOSTS env omits them → DisallowedHost errors flood the logsjorja-essex exists but maple-surgery does NOT (yet nginx cert + traffic reference it)4. Secondary issues.
Redis Authentication required — scheduled tasks not dispatchingALLOWED_HOSTS (or use a wildcard entry)maple-surgery partner org exists