Bug 553 (partial) — AI Panel 'Patient History' returned 'not in system'
even though the caller had 7 calls + an appointment. The model was
hallucinating instead of chaining lookup tools.
UUID safety: LLMs drop hyphens / swap chars on 36-char ids once the
context wears thin. To keep the model off the UUID path for 'this
caller' questions:
- lookup_appointments, lookup_call_history, lookup_lead_activities
now accept their id arguments OPTIONALLY
- when omitted, the sidecar resolves leadId from ctx and patientId
from the lead record (cached per-request)
- new lookup_lead_activities tool rounds out the patient-history
trio (call history + activity log + appointments)
System prompt (ccAgentHelper) tightened:
- chain call history + activities + appointments for history questions
- call lookup tools with NO arguments when using the current caller
- don't re-type UUIDs seen in CURRENT CONTEXT
- say 'feature not set up yet' when KB section is empty (packages,
etc.) instead of 'I couldn't find that'
All agent tools now emit structured [AI-TOOL] trace lines with full
UUIDs printed — tail sidecar logs to see which tool the model chose,
whether the model passed an id or used the context fallback, and how
many records came back. If the model ever hallucinates a UUID, the
resolved= field on the log line will echo it and count=0 will flag
the miss immediately.
- Team module: POST /api/team/members (in-place employee creation with
temp password + Redis cache), PUT /api/team/members/:id, GET temp
password endpoint. Uses signUpInWorkspace — no email invites.
- Dockerfile: rewritten as multi-stage build (builder + runtime) so
native modules compile for target arch. Fixes darwin→linux crash.
- .dockerignore: exclude dist, node_modules, .env, .git, data/
- package-lock.json: regenerated against public npmjs.org (was
pointing at localhost:4873 Verdaccio — broke docker builds)
- Doctor utils: shared DOCTOR_VISIT_SLOTS_FRAGMENT + normalizeDoctors
helper for visit-slot-aware queries across 6 consumers
- AI config: full admin CRUD (GET/PUT/POST reset), workspace-scoped
setup-state with workspace ID isolation, AI prompt defaults overhaul
- Agent config: camelCase field fix for SDK-synced workspaces
- Session service: workspace-scoped Redis key prefixing for setup state
- Recordings/supervisor/widget services: updated to use doctor-utils
shared fragments instead of inline visitingHours queries
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 of hospital onboarding & self-service plan
(docs/superpowers/plans/2026-04-06-hospital-onboarding-self-service.md).
Backend foundations to support the upcoming staff-portal Settings hub and
6-step setup wizard. No frontend in this phase.
New config services (mirroring ThemeService / WidgetConfigService):
- SetupStateService — tracks completion of 6 wizard steps; isWizardRequired()
drives the post-login redirect
- TelephonyConfigService — Ozonetel + Exotel + SIP, replaces 8 env vars,
seeds from env on first boot, masks secrets on GET,
'***masked***' sentinel on PUT means "keep existing"
- AiConfigService — provider, model, temperature, system prompt addendum;
API keys remain in env
New endpoints under /api/config:
- GET /api/config/setup-state returns state + wizardRequired flag
- PUT /api/config/setup-state/steps/:step mark step complete/incomplete
- POST /api/config/setup-state/dismiss dismiss wizard
- POST /api/config/setup-state/reset
- GET /api/config/telephony masked
- PUT /api/config/telephony
- POST /api/config/telephony/reset
- GET /api/config/ai
- PUT /api/config/ai
- POST /api/config/ai/reset
ConfigThemeModule is now @Global() so the new sidecar config services are
injectable from AuthModule, OzonetelAgentModule, MaintModule without creating
a circular dependency (ConfigThemeModule already imports AuthModule for
SessionService).
Migrated 11 env-var read sites to use the new services:
- ozonetel-agent.service: exotel API + ozonetel did/sipId via read-through getters
- ozonetel-agent.controller: defaultAgentId/Password/SipId via getters
- kookoo-ivr.controller: sipId/callerId via getters
- auth.controller: OZONETEL_AGENT_PASSWORD (login + logout)
- agent-config.service: sipDomain/wsPort/campaignName via getters
- maint.controller: forceReady + unlockAgent
- ai-provider: createAiModel and isAiConfigured refactored to pure factories
taking AiProviderOpts; no more ConfigService dependency
- widget-chat.service, recordings.service, ai-enrichment.service,
ai-chat.controller, ai-insight.consumer, call-assist.service: each builds
the AI model from AiConfigService.getConfig() + ConfigService API keys
Hot-reload guarantee: every consumer reads via a getter or builds per-call,
so admin updates take effect without sidecar restart. WidgetChatService
specifically rebuilds the model on each streamReply().
Bug fix bundled: dropped widget.json.hospitalName field (the original
duplicate that started this whole thread). WidgetConfigService now reads
brand.hospitalName from ThemeService at the 2 generateKey call sites.
Single source of truth for hospital name is workspace branding.
First-boot env seeding: TelephonyConfigService and AiConfigService both
copy their respective env vars into a fresh data/*.json on onModuleInit if
the file doesn't exist. Existing deployments auto-migrate without manual
intervention.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
