@proteinjs/conversation 2.5.0 → 2.7.0

package/src/OpenAiStreamProcessor.ts

@@ -90,10 +90,10 @@ export class OpenAiStreamProcessor {
             this.outputStreamTerminated = true;
           } else if (chunk.usage) {
             this.usageDataAccumulator.addTokenUsage({
-              promptTokens: chunk.usage.prompt_tokens,
+              inputTokens: chunk.usage.prompt_tokens,
+              cachedInputTokens: chunk.usage.prompt_tokens_details?.cached_tokens ?? 0,
               reasoningTokens: chunk.usage.completion_tokens_details?.reasoning_tokens ?? 0,
-              cachedPromptTokens: chunk.usage.prompt_tokens_details?.cached_tokens ?? 0,
-              completionTokens: chunk.usage.completion_tokens,
+              outputTokens: chunk.usage.completion_tokens,
               totalTokens: chunk.usage.total_tokens,
             });
             if (finishedProcessingToolCallStream) {

package/src/UsageData.ts

@@ -1,13 +1,30 @@
 import { TiktokenModel } from 'tiktoken';
 
 export type TokenUsage = {
-  promptTokens: number;
+  inputTokens: number;
+  cachedInputTokens: number;
   reasoningTokens: number;
-  cachedPromptTokens: number;
-  completionTokens: number;
+  outputTokens: number;
   totalTokens: number;
 };
 
+export type ModelApiCost = {
+  /** USD per 1M input tokens */
+  inputUsdPer1M: number;
+  /** USD per 1M cached input tokens (if supported) */
+  cachedInputUsdPer1M?: number;
+  /** USD per 1M output tokens */
+  outputUsdPer1M: number;
+};
+
+export type UsageCostUsd = {
+  inputUsd: number;
+  cachedInputUsd: number;
+  reasoningUsd: number;
+  outputUsd: number;
+  totalUsd: number;
+};
+
 /**
  * Usage data accumulated throughout the lifecycle of a single call to
  * `OpenAi.generateResponse` or `OpenAi.generateStreamingResponse`.
@@ -17,8 +34,12 @@ export type UsageData = {
   model: TiktokenModel;
   /** The token usage of the initial request sent to the assistant */
   initialRequestTokenUsage: TokenUsage;
+  /** The USD cost of the initial request */
+  initialRequestCostUsd: UsageCostUsd;
   /** The total token usage of all requests sent to the assistant (ie. initial request + all subsequent tool call requests) */
   totalTokenUsage: TokenUsage;
+  /** The total USD cost of all requests sent to the assistant */
+  totalCostUsd: UsageCostUsd;
   /** The number of requests sent to the assistant */
   totalRequestsToAssistant: number;
   /** The number of times each tool was called by the assistant */
@@ -39,36 +60,71 @@ export class UsageDataAccumulator {
     this.usageData = {
       model,
       initialRequestTokenUsage: {
-        promptTokens: 0,
+        inputTokens: 0,
         reasoningTokens: 0,
-        cachedPromptTokens: 0,
-        completionTokens: 0,
+        cachedInputTokens: 0,
+        outputTokens: 0,
         totalTokens: 0,
       },
+      initialRequestCostUsd: {
+        inputUsd: 0,
+        cachedInputUsd: 0,
+        reasoningUsd: 0,
+        outputUsd: 0,
+        totalUsd: 0,
+      },
       totalTokenUsage: {
-        promptTokens: 0,
+        inputTokens: 0,
+        cachedInputTokens: 0,
         reasoningTokens: 0,
-        cachedPromptTokens: 0,
-        completionTokens: 0,
+        outputTokens: 0,
         totalTokens: 0,
       },
+      totalCostUsd: {
+        inputUsd: 0,
+        cachedInputUsd: 0,
+        reasoningUsd: 0,
+        outputUsd: 0,
+        totalUsd: 0,
+      },
       totalRequestsToAssistant: 0,
       callsPerTool: {},
       totalToolCalls: 0,
     };
   }
 
-  addTokenUsage(tokenUsage: TokenUsage) {
+  addTokenUsage(tokenUsage: TokenUsage, opts?: { serviceTier?: string }) {
     this.usageData.totalRequestsToAssistant++;
+
+    const cost = calculateUsageCostUsd(this.usageData.model, tokenUsage, { serviceTier: opts?.serviceTier });
+
     if (!this.processedInitialRequest) {
       this.usageData.initialRequestTokenUsage = tokenUsage;
+      this.usageData.initialRequestCostUsd = cost;
       this.processedInitialRequest = true;
     }
+
+    if (cost) {
+      if (!this.usageData.totalCostUsd) {
+        this.usageData.totalCostUsd = { ...cost };
+      } else {
+        this.usageData.totalCostUsd = {
+          inputUsd: this.usageData.totalCostUsd.inputUsd + cost.inputUsd,
+          cachedInputUsd: this.usageData.totalCostUsd.cachedInputUsd + cost.cachedInputUsd,
+          reasoningUsd: this.usageData.totalCostUsd.reasoningUsd + cost.reasoningUsd,
+          outputUsd: this.usageData.totalCostUsd.outputUsd + cost.outputUsd,
+          totalUsd: this.usageData.totalCostUsd.totalUsd + cost.totalUsd,
+        };
+      }
+
+      this.usageData.totalCostUsd = roundUsageCostUsdToCents(this.usageData.totalCostUsd);
+    }
+
     this.usageData.totalTokenUsage = {
-      promptTokens: this.usageData.totalTokenUsage.promptTokens + tokenUsage.promptTokens,
+      inputTokens: this.usageData.totalTokenUsage.inputTokens + tokenUsage.inputTokens,
+      cachedInputTokens: this.usageData.totalTokenUsage.cachedInputTokens + tokenUsage.cachedInputTokens,
       reasoningTokens: this.usageData.totalTokenUsage.reasoningTokens + tokenUsage.reasoningTokens,
-      cachedPromptTokens: this.usageData.totalTokenUsage.promptTokens + tokenUsage.cachedPromptTokens,
-      completionTokens: this.usageData.totalTokenUsage.completionTokens + tokenUsage.completionTokens,
+      outputTokens: this.usageData.totalTokenUsage.outputTokens + tokenUsage.outputTokens,
       totalTokens: this.usageData.totalTokenUsage.totalTokens + tokenUsage.totalTokens,
     };
   }
@@ -82,3 +138,310 @@ export class UsageDataAccumulator {
     this.usageData.totalToolCalls++;
   }
 }
+
+/**
+ * Aggregate multiple UsageData objects into a single UsageData.
+ */
+export function aggregateUsageData(list: UsageData[]): UsageData | undefined {
+  if (!Array.isArray(list) || list.length === 0) {
+    return undefined;
+  }
+
+  const first = list[0];
+
+  const out: UsageData = {
+    model: first.model,
+    initialRequestTokenUsage: { ...first.initialRequestTokenUsage },
+    totalTokenUsage: { ...first.totalTokenUsage },
+    totalRequestsToAssistant: first.totalRequestsToAssistant,
+    callsPerTool: { ...first.callsPerTool },
+    totalToolCalls: first.totalToolCalls,
+    initialRequestCostUsd: { ...first.initialRequestCostUsd },
+    totalCostUsd: { ...first.totalCostUsd },
+  };
+
+  for (const u of list.slice(1)) {
+    out.totalTokenUsage.inputTokens += u.totalTokenUsage.inputTokens;
+    out.totalTokenUsage.cachedInputTokens += u.totalTokenUsage.cachedInputTokens;
+    out.totalTokenUsage.reasoningTokens += u.totalTokenUsage.reasoningTokens;
+    out.totalTokenUsage.outputTokens += u.totalTokenUsage.outputTokens;
+    out.totalTokenUsage.totalTokens += u.totalTokenUsage.totalTokens;
+
+    out.totalRequestsToAssistant += u.totalRequestsToAssistant;
+    out.totalToolCalls += u.totalToolCalls;
+
+    for (const [k, v] of Object.entries(u.callsPerTool)) {
+      out.callsPerTool[k] = (out.callsPerTool[k] ?? 0) + v;
+    }
+
+    out.totalCostUsd.inputUsd += u.totalCostUsd.inputUsd;
+    out.totalCostUsd.cachedInputUsd += u.totalCostUsd.cachedInputUsd;
+    out.totalCostUsd.reasoningUsd += u.totalCostUsd.reasoningUsd;
+    out.totalCostUsd.outputUsd += u.totalCostUsd.outputUsd;
+    out.totalCostUsd.totalUsd += u.totalCostUsd.totalUsd;
+
+    out.totalCostUsd = roundUsageCostUsdToCents(out.totalCostUsd);
+  }
+
+  return out;
+}
+
+/**
+ * Standard tier costs (USD per 1M tokens) based on the pricing on OpenAI's website.
+ */
+export const MODEL_API_COST_USD_PER_1M_TOKENS_STANDARD: Record<string, ModelApiCost> = {
+  'gpt-5.2': { inputUsdPer1M: 1.75, cachedInputUsdPer1M: 0.175, outputUsdPer1M: 14.0 },
+  'gpt-5.1': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-5': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-5-mini': { inputUsdPer1M: 0.25, cachedInputUsdPer1M: 0.025, outputUsdPer1M: 2.0 },
+  'gpt-5-nano': { inputUsdPer1M: 0.05, cachedInputUsdPer1M: 0.005, outputUsdPer1M: 0.4 },
+
+  'gpt-5.2-chat-latest': { inputUsdPer1M: 1.75, cachedInputUsdPer1M: 0.175, outputUsdPer1M: 14.0 },
+  'gpt-5.1-chat-latest': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-5-chat-latest': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+
+  'gpt-5.2-codex': { inputUsdPer1M: 1.75, cachedInputUsdPer1M: 0.175, outputUsdPer1M: 14.0 },
+  'gpt-5.1-codex-max': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-5.1-codex': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-5-codex': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+
+  'gpt-5.2-pro': { inputUsdPer1M: 21.0, outputUsdPer1M: 168.0 },
+  'gpt-5-pro': { inputUsdPer1M: 15.0, outputUsdPer1M: 120.0 },
+
+  'gpt-4.1': { inputUsdPer1M: 2.0, cachedInputUsdPer1M: 0.5, outputUsdPer1M: 8.0 },
+  'gpt-4.1-mini': { inputUsdPer1M: 0.4, cachedInputUsdPer1M: 0.1, outputUsdPer1M: 1.6 },
+  'gpt-4.1-nano': { inputUsdPer1M: 0.1, cachedInputUsdPer1M: 0.025, outputUsdPer1M: 0.4 },
+
+  'gpt-4o': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 1.25, outputUsdPer1M: 10.0 },
+  'gpt-4o-2024-05-13': { inputUsdPer1M: 5.0, outputUsdPer1M: 15.0 },
+  'gpt-4o-mini': { inputUsdPer1M: 0.15, cachedInputUsdPer1M: 0.075, outputUsdPer1M: 0.6 },
+
+  'gpt-realtime': { inputUsdPer1M: 4.0, cachedInputUsdPer1M: 0.4, outputUsdPer1M: 16.0 },
+  'gpt-realtime-mini': { inputUsdPer1M: 0.6, cachedInputUsdPer1M: 0.06, outputUsdPer1M: 2.4 },
+
+  'gpt-4o-realtime-preview': { inputUsdPer1M: 5.0, cachedInputUsdPer1M: 2.5, outputUsdPer1M: 20.0 },
+  'gpt-4o-mini-realtime-preview': { inputUsdPer1M: 0.6, cachedInputUsdPer1M: 0.3, outputUsdPer1M: 2.4 },
+
+  'gpt-audio': { inputUsdPer1M: 2.5, outputUsdPer1M: 10.0 },
+  'gpt-audio-mini': { inputUsdPer1M: 0.6, outputUsdPer1M: 2.4 },
+  'gpt-4o-audio-preview': { inputUsdPer1M: 2.5, outputUsdPer1M: 10.0 },
+  'gpt-4o-mini-audio-preview': { inputUsdPer1M: 0.15, outputUsdPer1M: 0.6 },
+
+  o1: { inputUsdPer1M: 15.0, cachedInputUsdPer1M: 7.5, outputUsdPer1M: 60.0 },
+  'o1-pro': { inputUsdPer1M: 150.0, outputUsdPer1M: 600.0 },
+  'o1-mini': { inputUsdPer1M: 1.1, cachedInputUsdPer1M: 0.55, outputUsdPer1M: 4.4 },
+
+  o3: { inputUsdPer1M: 2.0, cachedInputUsdPer1M: 0.5, outputUsdPer1M: 8.0 },
+  'o3-pro': { inputUsdPer1M: 20.0, outputUsdPer1M: 80.0 },
+  'o3-mini': { inputUsdPer1M: 1.1, cachedInputUsdPer1M: 0.55, outputUsdPer1M: 4.4 },
+  'o3-deep-research': { inputUsdPer1M: 10.0, cachedInputUsdPer1M: 2.5, outputUsdPer1M: 40.0 },
+
+  'o4-mini': { inputUsdPer1M: 1.1, cachedInputUsdPer1M: 0.275, outputUsdPer1M: 4.4 },
+  'o4-mini-deep-research': { inputUsdPer1M: 2.0, cachedInputUsdPer1M: 0.5, outputUsdPer1M: 8.0 },
+
+  'gpt-5-search-api': { inputUsdPer1M: 1.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 10.0 },
+  'gpt-4o-mini-search-preview': { inputUsdPer1M: 0.15, outputUsdPer1M: 0.6 },
+  'gpt-4o-search-preview': { inputUsdPer1M: 2.5, outputUsdPer1M: 10.0 },
+  'computer-use-preview': { inputUsdPer1M: 3.0, outputUsdPer1M: 12.0 },
+
+  'gpt-5.1-codex-mini': { inputUsdPer1M: 0.25, cachedInputUsdPer1M: 0.025, outputUsdPer1M: 2.0 },
+  'codex-mini-latest': { inputUsdPer1M: 1.5, cachedInputUsdPer1M: 0.375, outputUsdPer1M: 6.0 },
+};
+
+export const MODEL_API_COST_USD_PER_1M_TOKENS_BATCH: Record<string, ModelApiCost> = {
+  'gpt-5.2': { inputUsdPer1M: 0.875, cachedInputUsdPer1M: 0.0875, outputUsdPer1M: 7.0 },
+  'gpt-5.1': { inputUsdPer1M: 0.625, cachedInputUsdPer1M: 0.0625, outputUsdPer1M: 5.0 },
+  'gpt-5': { inputUsdPer1M: 0.625, cachedInputUsdPer1M: 0.0625, outputUsdPer1M: 5.0 },
+  'gpt-5-mini': { inputUsdPer1M: 0.125, cachedInputUsdPer1M: 0.0125, outputUsdPer1M: 1.0 },
+  'gpt-5-nano': { inputUsdPer1M: 0.025, cachedInputUsdPer1M: 0.0025, outputUsdPer1M: 0.2 },
+
+  'gpt-5.2-pro': { inputUsdPer1M: 10.5, outputUsdPer1M: 84.0 },
+  'gpt-5-pro': { inputUsdPer1M: 7.5, outputUsdPer1M: 60.0 },
+
+  'gpt-4.1': { inputUsdPer1M: 1.0, outputUsdPer1M: 4.0 },
+  'gpt-4.1-mini': { inputUsdPer1M: 0.2, outputUsdPer1M: 0.8 },
+  'gpt-4.1-nano': { inputUsdPer1M: 0.05, outputUsdPer1M: 0.2 },
+
+  'gpt-4o': { inputUsdPer1M: 1.25, outputUsdPer1M: 5.0 },
+  'gpt-4o-2024-05-13': { inputUsdPer1M: 2.5, outputUsdPer1M: 7.5 },
+  'gpt-4o-mini': { inputUsdPer1M: 0.075, outputUsdPer1M: 0.3 },
+
+  o1: { inputUsdPer1M: 7.5, outputUsdPer1M: 30.0 },
+  'o1-pro': { inputUsdPer1M: 75.0, outputUsdPer1M: 300.0 },
+
+  'o3-pro': { inputUsdPer1M: 10.0, outputUsdPer1M: 40.0 },
+  o3: { inputUsdPer1M: 1.0, outputUsdPer1M: 4.0 },
+  'o3-deep-research': { inputUsdPer1M: 5.0, outputUsdPer1M: 20.0 },
+
+  'o4-mini': { inputUsdPer1M: 0.55, outputUsdPer1M: 2.2 },
+  'o4-mini-deep-research': { inputUsdPer1M: 1.0, outputUsdPer1M: 4.0 },
+  'o3-mini': { inputUsdPer1M: 0.55, outputUsdPer1M: 2.2 },
+  'o1-mini': { inputUsdPer1M: 0.55, outputUsdPer1M: 2.2 },
+
+  'computer-use-preview': { inputUsdPer1M: 1.5, outputUsdPer1M: 6.0 },
+};
+
+export const MODEL_API_COST_USD_PER_1M_TOKENS_FLEX: Record<string, ModelApiCost> = {
+  'gpt-5.2': { inputUsdPer1M: 0.875, cachedInputUsdPer1M: 0.0875, outputUsdPer1M: 7.0 },
+  'gpt-5.1': { inputUsdPer1M: 0.625, cachedInputUsdPer1M: 0.0625, outputUsdPer1M: 5.0 },
+  'gpt-5': { inputUsdPer1M: 0.625, cachedInputUsdPer1M: 0.0625, outputUsdPer1M: 5.0 },
+  'gpt-5-mini': { inputUsdPer1M: 0.125, cachedInputUsdPer1M: 0.0125, outputUsdPer1M: 1.0 },
+  'gpt-5-nano': { inputUsdPer1M: 0.025, cachedInputUsdPer1M: 0.0025, outputUsdPer1M: 0.2 },
+
+  o3: { inputUsdPer1M: 1.0, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 4.0 },
+  'o4-mini': { inputUsdPer1M: 0.55, cachedInputUsdPer1M: 0.138, outputUsdPer1M: 2.2 },
+};
+
+export const MODEL_API_COST_USD_PER_1M_TOKENS_PRIORITY: Record<string, ModelApiCost> = {
+  'gpt-5.2': { inputUsdPer1M: 3.5, cachedInputUsdPer1M: 0.35, outputUsdPer1M: 28.0 },
+  'gpt-5.1': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 20.0 },
+  'gpt-5': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 20.0 },
+  'gpt-5-mini': { inputUsdPer1M: 0.45, cachedInputUsdPer1M: 0.045, outputUsdPer1M: 3.6 },
+
+  'gpt-5.2-codex': { inputUsdPer1M: 3.5, cachedInputUsdPer1M: 0.35, outputUsdPer1M: 28.0 },
+  'gpt-5.1-codex-max': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 20.0 },
+  'gpt-5.1-codex': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 20.0 },
+  'gpt-5-codex': { inputUsdPer1M: 2.5, cachedInputUsdPer1M: 0.25, outputUsdPer1M: 20.0 },
+
+  'gpt-4.1': { inputUsdPer1M: 3.5, cachedInputUsdPer1M: 0.875, outputUsdPer1M: 14.0 },
+  'gpt-4.1-mini': { inputUsdPer1M: 0.7, cachedInputUsdPer1M: 0.175, outputUsdPer1M: 2.8 },
+  'gpt-4.1-nano': { inputUsdPer1M: 0.2, cachedInputUsdPer1M: 0.05, outputUsdPer1M: 0.8 },
+
+  'gpt-4o': { inputUsdPer1M: 4.25, cachedInputUsdPer1M: 2.125, outputUsdPer1M: 17.0 },
+  'gpt-4o-2024-05-13': { inputUsdPer1M: 8.75, outputUsdPer1M: 26.25 },
+  'gpt-4o-mini': { inputUsdPer1M: 0.25, cachedInputUsdPer1M: 0.125, outputUsdPer1M: 1.0 },
+
+  o3: { inputUsdPer1M: 3.5, cachedInputUsdPer1M: 0.875, outputUsdPer1M: 14.0 },
+  'o4-mini': { inputUsdPer1M: 2.0, cachedInputUsdPer1M: 0.5, outputUsdPer1M: 8.0 },
+};
+
+const TOKENS_PER_1M = 1_000_000;
+
+const normalizeModelIdForPricing = (model: string): string => {
+  const raw = String(model ?? '').trim();
+  if (!raw) {
+    return '';
+  }
+
+  // handle e.g. "openai:gpt-4o" or "openai/gpt-4o"
+  const afterColon = raw.includes(':') ? raw.split(':').pop() ?? raw : raw;
+  const afterSlash = afterColon.includes('/') ? afterColon.split('/').pop() ?? afterColon : afterColon;
+  return afterSlash;
+};
+
+type UsagePricingTier = 'standard' | 'batch' | 'flex' | 'priority';
+
+const normalizeServiceTierForPricing = (serviceTier?: string): UsagePricingTier => {
+  const v = String(serviceTier ?? '')
+    .trim()
+    .toLowerCase();
+  if (v === 'priority') {
+    return 'priority';
+  }
+  if (v === 'flex') {
+    return 'flex';
+  }
+  if (v === 'batch') {
+    return 'batch';
+  }
+  return 'standard';
+};
+
+const resolveModelApiCost = (model: string, tier?: UsagePricingTier): ModelApiCost | undefined => {
+  const m = normalizeModelIdForPricing(model);
+  if (!m) {
+    return undefined;
+  }
+
+  const t: UsagePricingTier = tier ?? 'standard';
+  const table =
+    t === 'priority'
+      ? MODEL_API_COST_USD_PER_1M_TOKENS_PRIORITY
+      : t === 'flex'
+        ? MODEL_API_COST_USD_PER_1M_TOKENS_FLEX
+        : t === 'batch'
+          ? MODEL_API_COST_USD_PER_1M_TOKENS_BATCH
+          : MODEL_API_COST_USD_PER_1M_TOKENS_STANDARD;
+
+  const direct = table[m];
+  if (direct) {
+    return direct;
+  }
+
+  // common suffix normalization (only if the base exists)
+  if (m.endsWith('-latest')) {
+    const base = m.slice(0, -'-latest'.length);
+    const baseCost = table[base];
+    if (baseCost) {
+      return baseCost;
+    }
+  }
+
+  return undefined;
+};
+
+export const calculateUsageCostUsd = (
+  model: string,
+  tokenUsage: TokenUsage,
+  opts?: { serviceTier?: string }
+): UsageCostUsd => {
+  const tier = normalizeServiceTierForPricing(opts?.serviceTier);
+  const pricing = resolveModelApiCost(model, tier);
+  if (!pricing) {
+    return {
+      inputUsd: 0,
+      cachedInputUsd: 0,
+      reasoningUsd: 0,
+      outputUsd: 0,
+      totalUsd: 0,
+    };
+  }
+
+  const input = Number.isFinite(tokenUsage.inputTokens) ? Number(tokenUsage.inputTokens) : 0;
+  const cachedInput = Number.isFinite(tokenUsage.cachedInputTokens) ? Number(tokenUsage.cachedInputTokens) : 0;
+  const reasoning = Number.isFinite(tokenUsage.reasoningTokens) ? Number(tokenUsage.reasoningTokens) : 0;
+  const output = Number.isFinite(tokenUsage.outputTokens) ? Number(tokenUsage.outputTokens) : 0;
+
+  const inputTokens = Math.max(0, input);
+  const cachedInputTokens = Math.max(0, cachedInput);
+  const nonCachedInputTokens = Math.max(0, inputTokens - cachedInputTokens);
+  const reasoningTokens = Math.max(0, reasoning);
+  const outputTokens = Math.max(0, output);
+
+  const cachedRate =
+    typeof pricing.cachedInputUsdPer1M === 'number' ? pricing.cachedInputUsdPer1M : pricing.inputUsdPer1M;
+
+  const inputUsd = (nonCachedInputTokens * pricing.inputUsdPer1M + cachedInputTokens * cachedRate) / TOKENS_PER_1M;
+  const cachedInputUsd = (cachedInputTokens * cachedRate) / TOKENS_PER_1M;
+  const reasoningUsd = (reasoningTokens * pricing.outputUsdPer1M) / TOKENS_PER_1M;
+  const outputUsd = (outputTokens * pricing.outputUsdPer1M) / TOKENS_PER_1M;
+  const totalUsd = inputUsd + outputUsd;
+
+  return roundUsageCostUsdToCents({
+    inputUsd,
+    cachedInputUsd,
+    reasoningUsd,
+    outputUsd,
+    totalUsd,
+  });
+};
+
+function roundToHundredths(value: number): number {
+  return Math.round(value * 100) / 100;
+}
+
+function roundUsageCostUsdToCents(cost: UsageCostUsd): UsageCostUsd {
+  const inputUsd = roundToHundredths(cost.inputUsd);
+  const cachedInputUsd = roundToHundredths(cost.cachedInputUsd);
+  const reasoningUsd = roundToHundredths(cost.reasoningUsd);
+  const outputUsd = roundToHundredths(cost.outputUsd);
+  const totalUsd = roundToHundredths(inputUsd + cachedInputUsd + outputUsd);
+
+  return {
+    inputUsd,
+    cachedInputUsd,
+    reasoningUsd,
+    outputUsd,
+    totalUsd,
+  };
+}

package/src/fs/conversation_fs/ConversationFsModule.ts

@@ -4,6 +4,7 @@ import { searchFilesFunctionName } from '../keyword_to_files_index/KeywordToFile
 import { searchLibrariesFunctionName, searchPackagesFunctionName } from '../package/PackageFunctions';
 import { ConversationFsModerator } from './ConversationFsModerator';
 import {
+  deleteFilesFunction,
   fsFunctions,
   getRecentlyAccessedFilePathsFunction,
   grepFunction,
@@ -46,6 +47,7 @@ export class ConversationFsModule implements ConversationModule {
     return [
       readFilesFunction(this),
       writeFilesFunction(this),
+      deleteFilesFunction(this),
       getRecentlyAccessedFilePathsFunction(this),
       grepFunction(this),
       ...fsFunctions,

package/src/fs/conversation_fs/FsFunctions.ts

@@ -115,6 +115,36 @@ export function writeFilesFunction(fsModule: ConversationFsModule) {
   };
 }
 
+export const deleteFilesFunctionName = 'deleteFiles';
+export function deleteFilesFunction(fsModule: ConversationFsModule) {
+  return {
+    definition: {
+      name: deleteFilesFunctionName,
+      description: 'Delete files from the file system',
+      parameters: {
+        type: 'object',
+        properties: {
+          paths: {
+            type: 'array',
+            description: 'Paths to the files',
+            items: { type: 'string' },
+          },
+        },
+        required: ['paths'],
+      },
+    },
+    call: async (params: { paths: string[] }) => {
+      try {
+        const absPaths = await canonicalizePaths(fsModule, params.paths);
+        return await Fs.deleteFiles(absPaths);
+      } catch (error: any) {
+        return error.message;
+      }
+    },
+    instructions: [`To delete files from the local file system, use the ${deleteFilesFunctionName} function`],
+  };
+}
+
 export const getRecentlyAccessedFilePathsFunctionName = 'getRecentlyAccessedFilePaths';
 export function getRecentlyAccessedFilePathsFunction(fsModule: ConversationFsModule) {
   return {
@@ -131,8 +161,8 @@ export function getRecentlyAccessedFilePathsFunction(fsModule: ConversationFsMod
   };
 }
 
-const createFolderFunctionName = 'createFolder';
-const createFolderFunction: Function = {
+export const createFolderFunctionName = 'createFolder';
+export const createFolderFunction: Function = {
   definition: {
     name: createFolderFunctionName,
     description: 'Create a folder/directory',