11 docs tagged with "voice-agent"

APPROVED — Stage 1 agent research, pre-populated from the AI Front Desk Provisioning handoff (2026-05-21), the eight locked Phase-0 founder decisions, the AI Front Desk and Local Authority offer pages, the voice-agent code (resolve_route, the funeral/vet vertical pattern, moderation.py), the chatbot stream route, and the confirmed `local_business_*` production schema, then resolved in the founder Stage-2 interview (2026-05-21). Defines the EXPECTED OUTPUT — what a caller, a website visitor, and the business owner experience — for the per-business AI voice agent and website chatbot that the AI Front Desk + Local Authority packages promise. The voice agent is the church-built, multi-tenant, LIFE-SAFETY-tier LiveKit agent (one deployed agent serves church / funeral / vet / local-business). All 16 decisions (8 Phase-0 architecture/scope + 8 Stage-2 resolutions) are folded in; CLAUDE.md Rule #17 is satisfied — the customer-facing build is cleared.

How Claude models (Haiku 4.5 primary, claude-sonnet fallback) are used across the chatbot, voice agent Care Agent, content generation scripts, and the Claude CLI for batch tasks

APPROVED — pre-populated from the 2026-06-08 four-agent examination (voice-agent audit, admin/data-model audit, cited best-practice research, QA acceptance audit) and three founder decisions (2026-06-08): (1) spec-then-build phased, (2) refer-first / capture-only-if-already-tried, (3) available on ALL agent tiers (chat + voice), then resolved in the Stage-2 founder interview (2026-06-08 — all open questions closed). Defines the EXPECTED OUTPUT — what a caller / chat visitor, the church benevolence team, and the church admin experience — when a person asks the church for food / money / shelter / rent / utility help (reportedly ~50%+ of church inbound). The model is REFER-FIRST: point callers to vetted local agencies + 211 first; capture for a human benevolence team only when the caller has already tried those; the AI NEVER decides worthiness. CLAUDE.md Rule #17 is SATISFIED — the build is cleared.

Cal.com scheduling integration used by the chatbot and voice agent to book pastoral appointments — per-church booking link configuration and how the booking tool is triggered

TTS for voice agent calls (Cartesia Sonic) — Cartesia provides the text-to-speech voice synthesis used during live phone calls via livekit-plugins-cartesia

Deploy updated voice agent code via Railway (git push auto-deploys); LiveKit Cloud connects automatically via the agent worker WebSocket — no manual reconnect step needed

Product overview for FuneralWiseAI — AI phone answering for funeral homes. $999 setup + $199/mo. First WiseAI Agency vertical outside of churches. Hosted in churchwiseai-web under /funeralwiseai, served via funeralwiseai.com hostname rewrite.

Stage 1 DRAFT — pre-populated from code and production research by an agent. Every touchpoint must be confirmed or corrected by the founder in a Stage-2 interview. No production deployment decisions should rely on this file alone.

Gemini 2.5 Flash usage as the primary voice agent LLM, plus Google OAuth for calendar integration and ShareWiseAI social platform connections

Update the product_knowledge table that is injected into chatbot and voice agent prompts at runtime — when pricing, features, or flows change, update here for immediate effect

Expected outputs across every touchpoint for VetWiseAI tenants — Discovery, Pre-Purchase, Email, Dashboard, Voice Agent, Public-Facing, Ongoing, and Lifecycle. Two tiers (Starter/Pro). Sales-led product with self-serve checkout as secondary path.