memorix 1.2.2 → 1.2.3

package/src/orchestrate/cost-tracker.ts

@@ -1,219 +1,219 @@
-/**
- * Cost Tracker — Phase 7, Step 8: Token → USD conversion and budget enforcement.
- *
- * Converts accumulated token usage (per model) into USD using a configurable
- * price table. Supports budget limits that abort the pipeline when exceeded.
- *
- * Design principle: if the price table doesn't have a model, report tokens only
- * (skip USD calculation). Never crash because of a missing price entry.
- */
-
-import type { TokenUsage } from './adapters/types.js';
-
-// ── Types ──────────────────────────────────────────────────────────
-
-export interface ModelPrice {
-  /** Cost per 1M input tokens in USD */
-  inputPer1M: number;
-  /** Cost per 1M output tokens in USD */
-  outputPer1M: number;
-  /** Cost per 1M cache read tokens in USD (default: 0) */
-  cacheReadPer1M?: number;
-  /** Cost per 1M cache write tokens in USD (default: 0) */
-  cacheWritePer1M?: number;
-}
-
-export interface CostSummary {
-  /** Total cost in USD (null if no price data available) */
-  totalUSD: number | null;
-  /** Per-model breakdown */
-  models: {
-    model: string;
-    inputTokens: number;
-    outputTokens: number;
-    cacheReadTokens: number;
-    cacheWriteTokens: number;
-    costUSD: number | null;
-  }[];
-  /** Whether the budget has been exceeded */
-  budgetExceeded: boolean;
-  /** Configured budget (null if no budget) */
-  budgetUSD: number | null;
-}
-
-// ── Default Price Table ────────────────────────────────────────────
-
-const DEFAULT_PRICES: Record<string, ModelPrice> = {
-  // Claude models — verified 2026-04-13 via Anthropic docs
-  'claude-opus-4-6': { inputPer1M: 5, outputPer1M: 25, cacheReadPer1M: 0.50, cacheWritePer1M: 6.25 },
-  'claude-sonnet-4-6': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
-  'claude-haiku-4-5': { inputPer1M: 1, outputPer1M: 5, cacheReadPer1M: 0.10, cacheWritePer1M: 1.25 },
-  'claude-haiku-4-5-20251001': { inputPer1M: 1, outputPer1M: 5, cacheReadPer1M: 0.10, cacheWritePer1M: 1.25 },
-  // Legacy Claude models (still in use by some adapters)
-  'claude-sonnet-4-20250514': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
-  'claude-sonnet-4-0-20250514': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
-  'claude-opus-4-20250514': { inputPer1M: 15, outputPer1M: 75, cacheReadPer1M: 1.50, cacheWritePer1M: 18.75 },
-  'claude-3-5-sonnet': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
-  'claude-3-5-haiku': { inputPer1M: 0.80, outputPer1M: 4, cacheReadPer1M: 0.08, cacheWritePer1M: 1 },
-
-  // OpenAI / Codex models — verified 2026-04-13
-  'gpt-5': { inputPer1M: 1.25, outputPer1M: 10 },
-  'gpt-5-mini': { inputPer1M: 0.25, outputPer1M: 2 },
-  'gpt-5-nano': { inputPer1M: 0.05, outputPer1M: 0.40 },
-  'gpt-5.2': { inputPer1M: 1.75, outputPer1M: 14 },
-  'o3': { inputPer1M: 2, outputPer1M: 8 },
-  'o3-mini': { inputPer1M: 1.10, outputPer1M: 4.40 },
-  'o4-mini': { inputPer1M: 1.10, outputPer1M: 4.40 },
-  'gpt-4.1': { inputPer1M: 2, outputPer1M: 8, cacheReadPer1M: 0.50, cacheWritePer1M: 2 },
-  'gpt-4.1-mini': { inputPer1M: 0.40, outputPer1M: 1.60, cacheReadPer1M: 0.10, cacheWritePer1M: 0.40 },
-  'gpt-4.1-nano': { inputPer1M: 0.10, outputPer1M: 0.40 },
-  'gpt-4o': { inputPer1M: 2.50, outputPer1M: 10, cacheReadPer1M: 1.25 },
-  'gpt-4o-mini': { inputPer1M: 0.15, outputPer1M: 0.60, cacheReadPer1M: 0.075 },
-
-  // Google Gemini models — verified 2026-04-13 via Google AI pricing
-  'gemini-2.5-flash': { inputPer1M: 0.30, outputPer1M: 2.50 },
-  'gemini-2.5-flash-preview-05-20': { inputPer1M: 0.30, outputPer1M: 2.50 },
-  'gemini-2.5-flash-lite': { inputPer1M: 0.10, outputPer1M: 0.40 },
-  'gemini-2.5-pro': { inputPer1M: 1.25, outputPer1M: 10 },
-  'gemini-2.0-flash': { inputPer1M: 0.10, outputPer1M: 0.40 },
-  'gemini-2.0-flash-lite': { inputPer1M: 0.075, outputPer1M: 0.30 },
-  // Gemini 3.x generation
-  'gemini-3.1-pro': { inputPer1M: 2, outputPer1M: 12 },
-  'gemini-3-flash': { inputPer1M: 0.50, outputPer1M: 3 },
-
-  // OpenCode / open models — verified 2026-04-13
-  'deepseek-coder': { inputPer1M: 0.28, outputPer1M: 0.42 },
-  'deepseek-chat': { inputPer1M: 0.28, outputPer1M: 0.42 },
-};
-
-// ── Core ───────────────────────────────────────────────────────────
-
-/**
- * Calculate cost for a single model's token usage.
- * Returns null if the model is not in the price table.
- */
-export function calculateModelCost(
-  model: string,
-  usage: TokenUsage,
-  customPrices?: Record<string, ModelPrice>,
-): number | null {
-  const prices = { ...DEFAULT_PRICES, ...customPrices };
-  const price = findPrice(model, prices);
-  if (!price) return null;
-
-  const inputCost = (usage.inputTokens / 1_000_000) * price.inputPer1M;
-  const outputCost = (usage.outputTokens / 1_000_000) * price.outputPer1M;
-  const cacheReadCost = (usage.cacheReadTokens / 1_000_000) * (price.cacheReadPer1M ?? 0);
-  const cacheWriteCost = (usage.cacheWriteTokens / 1_000_000) * (price.cacheWritePer1M ?? 0);
-
-  return inputCost + outputCost + cacheReadCost + cacheWriteCost;
-}
-
-/**
- * Calculate total pipeline cost from accumulated token usage.
- */
-export function calculatePipelineCost(
-  tokenUsage: Record<string, TokenUsage>,
-  budgetUSD?: number,
-  customPrices?: Record<string, ModelPrice>,
-): CostSummary {
-  const models: CostSummary['models'] = [];
-  let totalUSD: number | null = 0;
-  let hasAnyPrice = false;
-
-  for (const [model, usage] of Object.entries(tokenUsage)) {
-    const cost = calculateModelCost(model, usage, customPrices);
-    if (cost !== null) {
-      hasAnyPrice = true;
-      totalUSD = (totalUSD ?? 0) + cost;
-    }
-
-    models.push({
-      model,
-      inputTokens: usage.inputTokens,
-      outputTokens: usage.outputTokens,
-      cacheReadTokens: usage.cacheReadTokens,
-      cacheWriteTokens: usage.cacheWriteTokens,
-      costUSD: cost !== null ? Math.round(cost * 10000) / 10000 : null, // Round to 4 decimals
-    });
-  }
-
-  if (!hasAnyPrice) totalUSD = null;
-  else totalUSD = Math.round(totalUSD! * 10000) / 10000;
-
-  return {
-    totalUSD,
-    models,
-    budgetExceeded: budgetUSD != null && totalUSD != null && totalUSD > budgetUSD,
-    budgetUSD: budgetUSD ?? null,
-  };
-}
-
-/**
- * Check if the current cost exceeds the budget.
- * Returns false if no budget is set or cost cannot be calculated.
- */
-export function isBudgetExceeded(
-  tokenUsage: Record<string, TokenUsage>,
-  budgetUSD?: number,
-  customPrices?: Record<string, ModelPrice>,
-): boolean {
-  if (budgetUSD == null) return false;
-  const summary = calculatePipelineCost(tokenUsage, budgetUSD, customPrices);
-  return summary.budgetExceeded;
-}
-
-/**
- * Format cost summary for CLI display.
- */
-export function formatCostSummary(summary: CostSummary): string {
-  const lines: string[] = [];
-
-  for (const m of summary.models) {
-    const tokens = `in=${fmtNum(m.inputTokens)} out=${fmtNum(m.outputTokens)}`;
-    const cache = m.cacheReadTokens > 0 || m.cacheWriteTokens > 0
-      ? ` cache_r=${fmtNum(m.cacheReadTokens)} cache_w=${fmtNum(m.cacheWriteTokens)}`
-      : '';
-    const cost = m.costUSD !== null ? ` ($${m.costUSD.toFixed(4)})` : '';
-    lines.push(`  ${m.model}: ${tokens}${cache}${cost}`);
-  }
-
-  if (summary.totalUSD !== null) {
-    lines.push(`  Total: $${summary.totalUSD.toFixed(4)}`);
-  }
-
-  if (summary.budgetUSD !== null) {
-    const status = summary.budgetExceeded ? ' [WARN] EXCEEDED' : '';
-    lines.push(`  Budget: $${summary.budgetUSD.toFixed(2)}${status}`);
-  }
-
-  return lines.join('\n');
-}
-
-// ── Helpers ────────────────────────────────────────────────────────
-
-/**
- * Find price entry by model name. Supports fuzzy matching:
- * "claude-sonnet-4-20250514" matches "claude-sonnet-4-20250514" exactly,
- * but also tries prefix matching for versioned models.
- */
-function findPrice(model: string, prices: Record<string, ModelPrice>): ModelPrice | null {
-  // Exact match
-  if (prices[model]) return prices[model];
-
-  // Prefix match: "claude-sonnet-4-20250514" → try "claude-sonnet-4"
-  const normalized = model.toLowerCase();
-  for (const [key, price] of Object.entries(prices)) {
-    if (normalized.startsWith(key.toLowerCase()) || key.toLowerCase().startsWith(normalized)) {
-      return price;
-    }
-  }
-
-  return null;
-}
-
-function fmtNum(n: number): string {
-  if (n >= 1_000_000) return `${(n / 1_000_000).toFixed(1)}M`;
-  if (n >= 1_000) return `${(n / 1_000).toFixed(1)}K`;
-  return String(n);
-}
+/**
+ * Cost Tracker — Phase 7, Step 8: Token → USD conversion and budget enforcement.
+ *
+ * Converts accumulated token usage (per model) into USD using a configurable
+ * price table. Supports budget limits that abort the pipeline when exceeded.
+ *
+ * Design principle: if the price table doesn't have a model, report tokens only
+ * (skip USD calculation). Never crash because of a missing price entry.
+ */
+
+import type { TokenUsage } from './adapters/types.js';
+
+// ── Types ──────────────────────────────────────────────────────────
+
+export interface ModelPrice {
+  /** Cost per 1M input tokens in USD */
+  inputPer1M: number;
+  /** Cost per 1M output tokens in USD */
+  outputPer1M: number;
+  /** Cost per 1M cache read tokens in USD (default: 0) */
+  cacheReadPer1M?: number;
+  /** Cost per 1M cache write tokens in USD (default: 0) */
+  cacheWritePer1M?: number;
+}
+
+export interface CostSummary {
+  /** Total cost in USD (null if no price data available) */
+  totalUSD: number | null;
+  /** Per-model breakdown */
+  models: {
+    model: string;
+    inputTokens: number;
+    outputTokens: number;
+    cacheReadTokens: number;
+    cacheWriteTokens: number;
+    costUSD: number | null;
+  }[];
+  /** Whether the budget has been exceeded */
+  budgetExceeded: boolean;
+  /** Configured budget (null if no budget) */
+  budgetUSD: number | null;
+}
+
+// ── Default Price Table ────────────────────────────────────────────
+
+const DEFAULT_PRICES: Record<string, ModelPrice> = {
+  // Claude models — verified 2026-04-13 via Anthropic docs
+  'claude-opus-4-6': { inputPer1M: 5, outputPer1M: 25, cacheReadPer1M: 0.50, cacheWritePer1M: 6.25 },
+  'claude-sonnet-4-6': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
+  'claude-haiku-4-5': { inputPer1M: 1, outputPer1M: 5, cacheReadPer1M: 0.10, cacheWritePer1M: 1.25 },
+  'claude-haiku-4-5-20251001': { inputPer1M: 1, outputPer1M: 5, cacheReadPer1M: 0.10, cacheWritePer1M: 1.25 },
+  // Legacy Claude models (still in use by some adapters)
+  'claude-sonnet-4-20250514': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
+  'claude-sonnet-4-0-20250514': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
+  'claude-opus-4-20250514': { inputPer1M: 15, outputPer1M: 75, cacheReadPer1M: 1.50, cacheWritePer1M: 18.75 },
+  'claude-3-5-sonnet': { inputPer1M: 3, outputPer1M: 15, cacheReadPer1M: 0.30, cacheWritePer1M: 3.75 },
+  'claude-3-5-haiku': { inputPer1M: 0.80, outputPer1M: 4, cacheReadPer1M: 0.08, cacheWritePer1M: 1 },
+
+  // OpenAI / Codex models — verified 2026-04-13
+  'gpt-5': { inputPer1M: 1.25, outputPer1M: 10 },
+  'gpt-5-mini': { inputPer1M: 0.25, outputPer1M: 2 },
+  'gpt-5-nano': { inputPer1M: 0.05, outputPer1M: 0.40 },
+  'gpt-5.2': { inputPer1M: 1.75, outputPer1M: 14 },
+  'o3': { inputPer1M: 2, outputPer1M: 8 },
+  'o3-mini': { inputPer1M: 1.10, outputPer1M: 4.40 },
+  'o4-mini': { inputPer1M: 1.10, outputPer1M: 4.40 },
+  'gpt-4.1': { inputPer1M: 2, outputPer1M: 8, cacheReadPer1M: 0.50, cacheWritePer1M: 2 },
+  'gpt-4.1-mini': { inputPer1M: 0.40, outputPer1M: 1.60, cacheReadPer1M: 0.10, cacheWritePer1M: 0.40 },
+  'gpt-4.1-nano': { inputPer1M: 0.10, outputPer1M: 0.40 },
+  'gpt-4o': { inputPer1M: 2.50, outputPer1M: 10, cacheReadPer1M: 1.25 },
+  'gpt-4o-mini': { inputPer1M: 0.15, outputPer1M: 0.60, cacheReadPer1M: 0.075 },
+
+  // Google Gemini models — verified 2026-04-13 via Google AI pricing
+  'gemini-2.5-flash': { inputPer1M: 0.30, outputPer1M: 2.50 },
+  'gemini-2.5-flash-preview-05-20': { inputPer1M: 0.30, outputPer1M: 2.50 },
+  'gemini-2.5-flash-lite': { inputPer1M: 0.10, outputPer1M: 0.40 },
+  'gemini-2.5-pro': { inputPer1M: 1.25, outputPer1M: 10 },
+  'gemini-2.0-flash': { inputPer1M: 0.10, outputPer1M: 0.40 },
+  'gemini-2.0-flash-lite': { inputPer1M: 0.075, outputPer1M: 0.30 },
+  // Gemini 3.x generation
+  'gemini-3.1-pro': { inputPer1M: 2, outputPer1M: 12 },
+  'gemini-3-flash': { inputPer1M: 0.50, outputPer1M: 3 },
+
+  // OpenCode / open models — verified 2026-04-13
+  'deepseek-coder': { inputPer1M: 0.28, outputPer1M: 0.42 },
+  'deepseek-chat': { inputPer1M: 0.28, outputPer1M: 0.42 },
+};
+
+// ── Core ───────────────────────────────────────────────────────────
+
+/**
+ * Calculate cost for a single model's token usage.
+ * Returns null if the model is not in the price table.
+ */
+export function calculateModelCost(
+  model: string,
+  usage: TokenUsage,
+  customPrices?: Record<string, ModelPrice>,
+): number | null {
+  const prices = { ...DEFAULT_PRICES, ...customPrices };
+  const price = findPrice(model, prices);
+  if (!price) return null;
+
+  const inputCost = (usage.inputTokens / 1_000_000) * price.inputPer1M;
+  const outputCost = (usage.outputTokens / 1_000_000) * price.outputPer1M;
+  const cacheReadCost = (usage.cacheReadTokens / 1_000_000) * (price.cacheReadPer1M ?? 0);
+  const cacheWriteCost = (usage.cacheWriteTokens / 1_000_000) * (price.cacheWritePer1M ?? 0);
+
+  return inputCost + outputCost + cacheReadCost + cacheWriteCost;
+}
+
+/**
+ * Calculate total pipeline cost from accumulated token usage.
+ */
+export function calculatePipelineCost(
+  tokenUsage: Record<string, TokenUsage>,
+  budgetUSD?: number,
+  customPrices?: Record<string, ModelPrice>,
+): CostSummary {
+  const models: CostSummary['models'] = [];
+  let totalUSD: number | null = 0;
+  let hasAnyPrice = false;
+
+  for (const [model, usage] of Object.entries(tokenUsage)) {
+    const cost = calculateModelCost(model, usage, customPrices);
+    if (cost !== null) {
+      hasAnyPrice = true;
+      totalUSD = (totalUSD ?? 0) + cost;
+    }
+
+    models.push({
+      model,
+      inputTokens: usage.inputTokens,
+      outputTokens: usage.outputTokens,
+      cacheReadTokens: usage.cacheReadTokens,
+      cacheWriteTokens: usage.cacheWriteTokens,
+      costUSD: cost !== null ? Math.round(cost * 10000) / 10000 : null, // Round to 4 decimals
+    });
+  }
+
+  if (!hasAnyPrice) totalUSD = null;
+  else totalUSD = Math.round(totalUSD! * 10000) / 10000;
+
+  return {
+    totalUSD,
+    models,
+    budgetExceeded: budgetUSD != null && totalUSD != null && totalUSD > budgetUSD,
+    budgetUSD: budgetUSD ?? null,
+  };
+}
+
+/**
+ * Check if the current cost exceeds the budget.
+ * Returns false if no budget is set or cost cannot be calculated.
+ */
+export function isBudgetExceeded(
+  tokenUsage: Record<string, TokenUsage>,
+  budgetUSD?: number,
+  customPrices?: Record<string, ModelPrice>,
+): boolean {
+  if (budgetUSD == null) return false;
+  const summary = calculatePipelineCost(tokenUsage, budgetUSD, customPrices);
+  return summary.budgetExceeded;
+}
+
+/**
+ * Format cost summary for CLI display.
+ */
+export function formatCostSummary(summary: CostSummary): string {
+  const lines: string[] = [];
+
+  for (const m of summary.models) {
+    const tokens = `in=${fmtNum(m.inputTokens)} out=${fmtNum(m.outputTokens)}`;
+    const cache = m.cacheReadTokens > 0 || m.cacheWriteTokens > 0
+      ? ` cache_r=${fmtNum(m.cacheReadTokens)} cache_w=${fmtNum(m.cacheWriteTokens)}`
+      : '';
+    const cost = m.costUSD !== null ? ` ($${m.costUSD.toFixed(4)})` : '';
+    lines.push(`  ${m.model}: ${tokens}${cache}${cost}`);
+  }
+
+  if (summary.totalUSD !== null) {
+    lines.push(`  Total: $${summary.totalUSD.toFixed(4)}`);
+  }
+
+  if (summary.budgetUSD !== null) {
+    const status = summary.budgetExceeded ? ' [WARN] EXCEEDED' : '';
+    lines.push(`  Budget: $${summary.budgetUSD.toFixed(2)}${status}`);
+  }
+
+  return lines.join('\n');
+}
+
+// ── Helpers ────────────────────────────────────────────────────────
+
+/**
+ * Find price entry by model name. Supports fuzzy matching:
+ * "claude-sonnet-4-20250514" matches "claude-sonnet-4-20250514" exactly,
+ * but also tries prefix matching for versioned models.
+ */
+function findPrice(model: string, prices: Record<string, ModelPrice>): ModelPrice | null {
+  // Exact match
+  if (prices[model]) return prices[model];
+
+  // Prefix match: "claude-sonnet-4-20250514" → try "claude-sonnet-4"
+  const normalized = model.toLowerCase();
+  for (const [key, price] of Object.entries(prices)) {
+    if (normalized.startsWith(key.toLowerCase()) || key.toLowerCase().startsWith(normalized)) {
+      return price;
+    }
+  }
+
+  return null;
+}
+
+function fmtNum(n: number): string {
+  if (n >= 1_000_000) return `${(n / 1_000_000).toFixed(1)}M`;
+  if (n >= 1_000) return `${(n / 1_000).toFixed(1)}K`;
+  return String(n);
+}