v1.1.0:3 — AI upgrades: history context, test connection, cost estimator, streaming preview

Four incremental upgrades to the AI program generator. No schema change, no /data migration.

1. History as context (the killer feature)
   - lib/ai/historyContext.ts builds a 90-day per-exercise rollup:
     frequency, recent weights, estimated 1RM (Epley), avg RPE,
     days-since-last, plus a STAGNANT flag when the heaviest weight in
     the new half doesn't beat the old half.
   - Generate page surfaces an "Include my workout history as context"
     checkbox (default on at >=10 logged workouts). When checked, the
     ~1-3 KB summary is appended to the system prompt so the model can
     recommend things like "you've stalled bench at 245 — try paused reps."
   - We deliberately don't ship raw set logs (privacy + token cost).

2. Test connection
   - POST /api/ai/test sends a tiny "say hi in 3 words" prompt and
     reports latency + first sample, or the error inline.
   - "Test connection" button next to "Save AI config" in
     Settings -> AI integration. Verifies provider/model/key/baseUrl
     without going through full program generation.

3. Cost estimator
   - lib/ai/pricing.ts ships a price table for major models
     (Claude 3.5/3.7/4/4.5, GPT-4o/5/o1/o3/o4-mini, Gemini 1.5/2.0/2.5).
     Ollama always returns 0; openai-compatible returns null.
   - Generation history shows per-row cost + a 30-day rolling total
     at the top of the page.

4. Streaming preview render
   - lib/ai/lenientJson.ts: stack-aware partial-JSON parser that
     auto-closes open strings/brackets/braces in reverse-of-opening
     order, drops dangling key:value pairs and partial keywords.
     Returns a best-effort snapshot of the program-so-far on each chunk.
   - Generate UI now renders a live "Building program..." panel that
     updates as weeks/days/exercises arrive instead of just showing
     raw text and waiting for stream end.

Tests: 26 new (ai-historyContext.test.ts, ai-lenientJson.test.ts,
ai-pricing.test.ts). 161 total pass.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

proof-of-work/lib/ai/historyContext.ts

@@ -0,0 +1,245 @@
+import type { PrismaClient } from '@prisma/client';
+
+/**
+ * Build a compact workout-history summary the AI can use as
+ * context for personalized program generation.
+ *
+ * We DELIBERATELY don't ship raw set logs — that would be tens of
+ * KB per request and burn tokens. Instead we compute per-exercise
+ * aggregates over a recent window (default 90 days):
+ *
+ *   - totalSets in window
+ *   - distinct workouts
+ *   - daysSinceLast
+ *   - lastWeight, lastReps (from the most-recent set)
+ *   - bestWeight (heaviest set in window)
+ *   - estimated 1RM (Epley formula on the heaviest weighted set)
+ *   - rpe trend (avg RPE over recent sets, if logged)
+ *   - stagnation flag (heaviest weight unchanged for 4+ weeks AND
+ *     ≥3 sessions of that exercise in those 4+ weeks)
+ *
+ * Plus a top-level summary: total workouts, frequency, primary
+ * exercise types touched.
+ *
+ * The output is JSON-stringifiable, ~5-15 KB for a typical user.
+ */
+
+export interface HistoryExerciseSummary {
+  name: string;
+  type: string;
+  totalSets: number;
+  distinctWorkouts: number;
+  daysSinceLast: number;
+  lastWeight: number | null;
+  lastReps: number | null;
+  bestWeight: number | null;
+  estimated1RM: number | null;
+  avgRpe: number | null;
+  stagnant: boolean;
+}
+
+export interface HistorySummary {
+  windowDays: number;
+  totalWorkouts: number;
+  workoutsPerWeek: number;
+  primaryTypes: string[]; // exercise types by descending volume
+  exercises: HistoryExerciseSummary[];
+}
+
+/** Epley estimated 1RM: weight * (1 + reps / 30) */
+function epley1RM(weight: number, reps: number): number {
+  return Math.round(weight * (1 + reps / 30));
+}
+
+export async function buildHistorySummary(
+  prisma: PrismaClient,
+  userId: string,
+  windowDays = 90,
+): Promise<HistorySummary> {
+  const cutoff = new Date(Date.now() - windowDays * 86_400_000);
+
+  // Pull every set log in the window with its exercise + workout
+  // date. One query, one result-set walk.
+  const sets = await prisma.setLog.findMany({
+    where: {
+      workout: {
+        userId,
+        deletedAt: null,
+        date: { gte: cutoff },
+      },
+    },
+    select: {
+      reps: true,
+      weight: true,
+      rpe: true,
+      exerciseId: true,
+      workoutId: true,
+      workout: { select: { date: true } },
+      exercise: { select: { name: true, type: true } },
+    },
+    orderBy: { workout: { date: 'desc' } },
+  });
+
+  if (sets.length === 0) {
+    return {
+      windowDays,
+      totalWorkouts: 0,
+      workoutsPerWeek: 0,
+      primaryTypes: [],
+      exercises: [],
+    };
+  }
+
+  const workoutIds = new Set(sets.map((s) => s.workoutId));
+  const totalWorkouts = workoutIds.size;
+  const weeks = windowDays / 7;
+  const workoutsPerWeek = Math.round((totalWorkouts / weeks) * 10) / 10;
+
+  // Group by exercise
+  const byExercise = new Map<
+    string,
+    {
+      name: string;
+      type: string;
+      sets: typeof sets;
+    }
+  >();
+  for (const s of sets) {
+    if (!byExercise.has(s.exerciseId)) {
+      byExercise.set(s.exerciseId, {
+        name: s.exercise.name,
+        type: s.exercise.type,
+        sets: [],
+      });
+    }
+    byExercise.get(s.exerciseId)!.sets.push(s);
+  }
+
+  // Per-exercise summaries
+  const now = Date.now();
+  const exercises: HistoryExerciseSummary[] = [];
+  for (const [, group] of byExercise) {
+    const groupSets = group.sets;
+    const distinctWorkouts = new Set(groupSets.map((s) => s.workoutId)).size;
+    const mostRecent = groupSets[0]; // already date-desc
+    const daysSinceLast = Math.floor(
+      (now - mostRecent.workout.date.getTime()) / 86_400_000,
+    );
+
+    const weightedSets = groupSets.filter(
+      (s): s is typeof s & { weight: number; reps: number } =>
+        typeof s.weight === 'number' && typeof s.reps === 'number',
+    );
+    const bestWeightSet = weightedSets.reduce<
+      | { weight: number; reps: number }
+      | null
+    >((best, s) => {
+      if (!best || s.weight > best.weight) return s;
+      return best;
+    }, null);
+    const bestWeight = bestWeightSet?.weight ?? null;
+    const estimated1RM =
+      bestWeightSet != null ? epley1RM(bestWeightSet.weight, bestWeightSet.reps) : null;
+
+    const rpeSets = groupSets.filter(
+      (s): s is typeof s & { rpe: number } => typeof s.rpe === 'number',
+    );
+    const avgRpe =
+      rpeSets.length > 0
+        ? Math.round(
+            (rpeSets.reduce((sum, s) => sum + s.rpe, 0) / rpeSets.length) * 10,
+          ) / 10
+        : null;
+
+    // Stagnation: best weight in oldest half == best weight in newest half
+    // AND ≥3 distinct sessions in the window.
+    let stagnant = false;
+    if (distinctWorkouts >= 3 && bestWeight != null && weightedSets.length >= 4) {
+      const sortedByDate = [...weightedSets].sort(
+        (a, b) => a.workout.date.getTime() - b.workout.date.getTime(),
+      );
+      const mid = Math.floor(sortedByDate.length / 2);
+      const oldHalfBest = Math.max(...sortedByDate.slice(0, mid).map((s) => s.weight));
+      const newHalfBest = Math.max(...sortedByDate.slice(mid).map((s) => s.weight));
+      // No improvement in the new half compared to the old half
+      if (newHalfBest <= oldHalfBest) stagnant = true;
+    }
+
+    exercises.push({
+      name: group.name,
+      type: group.type,
+      totalSets: groupSets.length,
+      distinctWorkouts,
+      daysSinceLast,
+      lastWeight: mostRecent.weight ?? null,
+      lastReps: mostRecent.reps ?? null,
+      bestWeight,
+      estimated1RM,
+      avgRpe,
+      stagnant,
+    });
+  }
+
+  // Sort exercises by total volume (sets) descending so the most
+  // important context is first if the model truncates.
+  exercises.sort((a, b) => b.totalSets - a.totalSets);
+
+  // Primary types by aggregate sets
+  const typeVolume = new Map<string, number>();
+  for (const ex of exercises) {
+    typeVolume.set(ex.type, (typeVolume.get(ex.type) ?? 0) + ex.totalSets);
+  }
+  const primaryTypes = Array.from(typeVolume.entries())
+    .sort((a, b) => b[1] - a[1])
+    .map(([t]) => t);
+
+  return {
+    windowDays,
+    totalWorkouts,
+    workoutsPerWeek,
+    primaryTypes,
+    exercises,
+  };
+}
+
+/**
+ * Format a HistorySummary as a compact string the LLM can actually
+ * use. Aims for <2KB of text even for heavy users.
+ */
+export function formatHistoryContext(summary: HistorySummary): string {
+  if (summary.totalWorkouts === 0) {
+    return `\nUSER HISTORY: no workouts logged in the last ${summary.windowDays} days.`;
+  }
+  const lines: string[] = [];
+  lines.push(
+    `\nUSER HISTORY (last ${summary.windowDays} days):`,
+    `  ${summary.totalWorkouts} workouts (~${summary.workoutsPerWeek}/week)`,
+    `  Primary work: ${summary.primaryTypes.slice(0, 4).join(', ')}`,
+    '',
+    `  Per-exercise activity (descending by volume; weights in user's logged unit):`,
+  );
+  // Cap at top 30 exercises
+  const top = summary.exercises.slice(0, 30);
+  for (const ex of top) {
+    const bits: string[] = [
+      `${ex.totalSets}s/${ex.distinctWorkouts}w`,
+      `${ex.daysSinceLast}d ago`,
+    ];
+    if (ex.bestWeight != null && ex.lastReps != null)
+      bits.push(`best ${ex.bestWeight}×${ex.lastReps}`);
+    if (ex.estimated1RM != null) bits.push(`~${ex.estimated1RM} 1RM`);
+    if (ex.avgRpe != null) bits.push(`avg RPE ${ex.avgRpe}`);
+    if (ex.stagnant) bits.push('STAGNANT');
+    lines.push(`    - ${ex.name} (${ex.type}): ${bits.join(' · ')}`);
+  }
+  if (summary.exercises.length > top.length) {
+    lines.push(
+      `    ...and ${summary.exercises.length - top.length} more exercises with lower volume`,
+    );
+  }
+  lines.push(
+    '',
+    `  When designing the program, weight recent activity heavily. Address STAGNANT exercises if relevant. Don't propose deload-week-heavy work for someone training infrequently, and don't propose 6-day splits for someone averaging <3 sessions/week.`,
+  );
+  return lines.join('\n');
+}