feat(ai): finish engine wiring — briefings via provider, flyer honors key

app/api/ai/briefing/route.ts

@@ -1,9 +1,7 @@
 import { NextRequest, NextResponse } from 'next/server';
 import { createServer } from '@/lib/supabase/server';
 import { requireUserContext } from '@/lib/supabase/auth';
-import Anthropic from '@anthropic-ai/sdk';
-
-const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
+import { resolveProvider } from '@/lib/ai/provider';
 
 export async function POST(req: NextRequest) {
   try {
@@ -146,14 +144,15 @@ Rules:
 - For weekly briefing, populate weeklyHighlights and weeklyConflicts
 - If data is sparse, be honest but still encouraging`;
 
-    const response = await anthropic.messages.create({
-      model: process.env.AI_MODEL ?? 'claude-haiku-4-5-20251001',
-      max_tokens: 2000,
+    const provider = await resolveProvider();
+    const completion = await provider.complete({
       system: systemPrompt,
       messages: [{ role: 'user', content: `Generate ${type} briefing for ${firstName}.\n\nData:\n${context}` }],
+      tools: [],
+      maxTokens: 2000,
     });
 
-    const text = response.content[0].type === 'text' ? response.content[0].text : '{}';
+    const text = completion.text || '{}';
 
     let briefing: Record<string, unknown>;
     try {

app/api/ai/flyer/route.ts

@@ -1,13 +1,12 @@
 import { NextRequest, NextResponse } from 'next/server';
 import Anthropic from '@anthropic-ai/sdk';
-import { createServer } from '@/lib/supabase/server';
+import { createServer, createServiceClient } from '@/lib/supabase/server';
 import { requireUserContext } from '@/lib/supabase/auth';
 import { rateLimit, clientIp } from '@/lib/server/rate-limit';
+import { getAIConfig } from '@/lib/ai/settings';
 
 export const runtime = 'nodejs';
 
-const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
-
 const IMAGE_TYPES = ['image/jpeg', 'image/png', 'image/webp', 'image/gif'];
 
 type ProposedEvent = {
@@ -99,8 +98,18 @@ Rules:
       ? { type: 'document' as const, source: { type: 'base64' as const, media_type: 'application/pdf' as const, data } }
       : { type: 'image' as const, source: { type: 'base64' as const, media_type: mediaType as 'image/jpeg' | 'image/png' | 'image/webp' | 'image/gif', data } };
 
+    // Flyer parsing needs PDF/vision input, which is Anthropic-specific — use the
+    // admin-configured Anthropic key/model (Admin → AI Engine), with env fallback.
+    const aiConfig = await getAIConfig(createServiceClient());
+    const apiKey = aiConfig.anthropicKey ?? process.env.ANTHROPIC_API_KEY;
+    if (!apiKey) {
+      return NextResponse.json({ error: 'Flyer scanning needs an Anthropic API key. Add one in Admin → AI Engine.' }, { status: 503 });
+    }
+    const model = aiConfig.provider === 'anthropic' && aiConfig.model ? aiConfig.model : 'claude-sonnet-4-6';
+    const anthropic = new Anthropic({ apiKey });
+
     const response = await anthropic.messages.create({
-      model: process.env.AI_MODEL ?? 'claude-sonnet-4-6',
+      model,
       max_tokens: 1500,
       messages: [{ role: 'user', content: [filePart, { type: 'text', text: prompt }] }],
     });

app/api/ai/weekly-briefing/route.ts

@@ -3,9 +3,7 @@ import { createServer } from '@/lib/supabase/server';
 import { requireUserContext } from '@/lib/supabase/auth';
 import { planLevel } from '@/lib/constants/plans';
 import { weekWindow, weekRangeLabel, choreCompletionRate, bucketByDay, dayLoad } from '@/lib/ai/weekly';
-import Anthropic from '@anthropic-ai/sdk';
-
-const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
+import { resolveProvider } from '@/lib/ai/provider';
 
 /**
  * Plus-tier Weekly AI Briefing. Distinct from the daily briefing: it reads a
@@ -166,14 +164,15 @@ Rules:
 - Score categories 0-100 honestly from the data; be encouraging if data is sparse but never invent events.
 - Emojis: 🏥 medical, ⚽ sports, 📚 school, ✈️ travel, 🍽️ dinner, 💼 work, 🎂 birthday.`;
 
-    const response = await anthropic.messages.create({
-      model: process.env.AI_MODEL ?? 'claude-haiku-4-5-20251001',
-      max_tokens: 2600,
+    const provider = await resolveProvider();
+    const completion = await provider.complete({
       system: systemPrompt,
       messages: [{ role: 'user', content: `Generate the weekly briefing for ${firstName}.\n\nData:\n${context}` }],
+      tools: [],
+      maxTokens: 2600,
     });
 
-    const text = response.content[0]?.type === 'text' ? response.content[0].text : '{}';
+    const text = completion.text || '{}';
 
     let briefing: Record<string, unknown>;
     try {

docs/AGENT_HANDOFF.md

@@ -1,7 +1,7 @@
 # Agent Handoff — Bubaly / FamilyOS
 
 Living context doc so another agent can continue without re-deriving everything.
-Last updated after PR #81. Keep this updated as you ship.
+Last updated after PR #84. Keep this updated as you ship.
 
 ## Product & stack
 - **Bubaly / FamilyOS** — a family operating system. Next.js 15 App Router + TS +
@@ -122,6 +122,7 @@ npx vitest run tests/<your>.test.ts   # full suite currently 363 passing
 - #79 Configurable AI engine (Claude/Anthropic OR ChatGPT/OpenAI) + admin UI for keys
 - #80 Domain canonicalization: `theagoras.com` → `www.bubaly.com` (next.config redirect)
 - #81 Removed "Loved by N families" social-proof badge from the marketing hero
+- #82 handoff regen · #83 support@bubaly.com everywhere · #84 finished AI-engine wiring (briefing/weekly/flyer)
 
 Latest migration applied to prod: **0048**. Next migration number: **0049**.
 
@@ -139,19 +140,22 @@ Latest migration applied to prod: **0048**. Next migration number: **0049**.
 - **Wired through:** all provider-based AI routes use `await resolveProvider()` (briefings
   via provider, conflict, home AI, marketing AI, social, accident, import) AND the main
   assistant `app/api/ai/chat/route.ts`.
-- **Still on the Anthropic SDK directly** (follow-up): `app/api/ai/briefing/route.ts`,
-  `app/api/ai/weekly-briefing/route.ts`, `app/api/ai/flyer/route.ts`
-  (verify: `grep -rln "@anthropic-ai/sdk" app lib`). Migrate the same way; `flyer` uses
-  image input which the generic `complete()` interface doesn't model yet (needs extension).
+- `complete()` takes an optional `maxTokens` (default 1024) — set it for long JSON outputs.
+- **AI wiring is now complete (#84):** `briefing` + `weekly-briefing` go through
+  `resolveProvider()`. `flyer` stays on the Anthropic SDK on purpose (PDF/vision input is
+  Anthropic-specific) but reads the admin-configured Anthropic key/model via `getAIConfig`
+  and returns 503 with a clear message if no Anthropic key is set. If you switch the engine
+  to OpenAI, flyer still needs an Anthropic key (or build an OpenAI-vision path; note: no PDF).
 - To use ChatGPT: `/admin/ai` → pick **ChatGPT (OpenAI)**, choose a model (gpt-4o…),
   paste the OpenAI key, Save. Env fallbacks: `OPENAI_API_KEY`, `AI_PROVIDER=openai`, `AI_MODEL`.
   (The OpenAI account/key must have active billing or calls 401/429.)
 
 ## Backlog (prioritized, each a clean PR)
 1. Remaining marketing pillars: **A/B testing**, **Lead scoring**, lifecycle journeys.
-2. Finish AI-engine wiring: migrate `briefing` / `weekly-briefing` / `flyer` to `resolveProvider()`.
-3. Broader UX brief (Phases 3/4/5/9/11): mobile-first polish, theme-token audit,
+2. Broader UX brief (Phases 3/4/5/9/11): mobile-first polish, theme-token audit,
    Family Command Center home, AI-native touches, performance.
+- (DONE #84) Finish AI-engine wiring: briefing/weekly-briefing → resolveProvider; flyer
+  reads admin Anthropic key.
 
 ## How to continue (quick start for the next agent)
 1. Read this whole file. Recreate `/tmp/sbq.mjs` if missing (see migration section);

lib/ai/provider.ts

@@ -19,24 +19,26 @@ export type AICompletion = {
   toolCalls: { name: string; args: Record<string, unknown> }[];
 };
 
+export type AICompleteInput = {
+  system: string;
+  messages: AIMessage[];
+  tools: AITool[];
+  /** Max output tokens. Defaults to 1024. */
+  maxTokens?: number;
+};
+
 export interface AIProvider {
   readonly id: string;
   readonly model: string;
-  complete(input: {
-    system: string;
-    messages: AIMessage[];
-    tools: AITool[];
-  }): Promise<AICompletion>;
+  complete(input: AICompleteInput): Promise<AICompletion>;
 }
 
 // --- Anthropic implementation ---
 export class AnthropicProvider implements AIProvider {
   id = 'anthropic';
   constructor(public model = 'claude-sonnet-4-6', private apiKey = process.env.ANTHROPIC_API_KEY!) {}
 
-  async complete({ system, messages, tools }: {
-    system: string; messages: AIMessage[]; tools: AITool[];
-  }): Promise<AICompletion> {
+  async complete({ system, messages, tools, maxTokens = 1024 }: AICompleteInput): Promise<AICompletion> {
     const res = await fetch('https://api.anthropic.com/v1/messages', {
       method: 'POST',
       headers: {
@@ -46,7 +48,7 @@ export class AnthropicProvider implements AIProvider {
       },
       body: JSON.stringify({
         model: this.model,
-        max_tokens: 1024,
+        max_tokens: maxTokens,
         system,
         tools: tools.map((t) => ({ name: t.name, description: t.description, input_schema: t.input_schema })),
         messages: messages
@@ -72,13 +74,11 @@ export class OpenAIProvider implements AIProvider {
   id = 'openai';
   constructor(public model = 'gpt-4o', private apiKey = process.env.OPENAI_API_KEY ?? '') {}
 
-  async complete({ system, messages, tools }: {
-    system: string; messages: AIMessage[]; tools: AITool[];
-  }): Promise<AICompletion> {
+  async complete({ system, messages, tools, maxTokens = 1024 }: AICompleteInput): Promise<AICompletion> {
     if (!this.apiKey) throw new Error('OpenAI API key is not configured');
     const body: Record<string, unknown> = {
       model: this.model,
-      max_tokens: 1024,
+      max_tokens: maxTokens,
       messages: [
         ...(system ? [{ role: 'system', content: system }] : []),
         ...messages