[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6F7PV4MGXET42MVAFYM9N3Y":3,"skill-same:01M6F7PV4MGXET42MVAFYM9N3Y":87},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":28,"publishedAt":55,"linkOnly":56,"markdownContent":57,"markdownBody":58,"frontMatter":59,"refName":64,"commitSha":65,"skillDirectory":66,"forksCount":67,"repositoryHtmlUrl":68,"repositoryHomepageUrl":6,"ownerHtmlUrl":69,"ownerBlogUrl":70,"ownerDescription":71,"updatedAt":72,"install":73,"resources":86},"01M6F7PV4MGXET42MVAFYM9N3Y","langfuse-cost-tuning",null,"Monitor and optimize LLM costs using Langfuse analytics and dashboards.\nUse when tracking LLM spending, identifying cost anomalies,\nor implementing cost controls for AI applications.\nTrigger with phrases like \"langfuse costs\", \"LLM spending\",\n\"track AI costs\", \"langfuse token usage\", \"optimize LLM budget\".","Dicklesworthstone\u002Fpi_agent_rust","Dicklesworthstone","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F35050222?v=4","01M6CG26AR942CSVMN1M5E3DSE","Jeff Emanuel","01M6CG26EPP71PSHW1VDS71121",1992,"MIT","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Flangfuse-pack\u002Fskills\u002Flangfuse-cost-tuning\u002FSKILL.md",[18,23],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000013","分析与可视化","SCENARIO","使用场景",{"categoryNo":24,"categoryName":25,"taxonomyType":26,"taxonomyTypeDesc":27},"01JBSEEDCAT000000000000033","数据分析师","PROFESSION","适用职业",[29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54],"langfuse","tuning","langfuse cost tuning","cost","optimize","when","track","for","implementing","spending","usage","with","controls","trigger","io","token","analytics","use","tracking","edit","write","read","dashboards","applications","Budget","monitor","2026-11-10 10:12:30",false,"---\nname: langfuse-cost-tuning\ndescription: |\n  Monitor and optimize LLM costs using Langfuse analytics and dashboards.\n  Use when tracking LLM spending, identifying cost anomalies,\n  or implementing cost controls for AI applications.\n  Trigger with phrases like \"langfuse costs\", \"LLM spending\",\n  \"track AI costs\", \"langfuse token usage\", \"optimize LLM budget\".\nallowed-tools: Read, Write, Edit\nversion: 1.0.0\nlicense: MIT\nauthor: Jeremy Longshore \u003Cjeremy@intentsolutions.io>\n---\n\n# Langfuse Cost Tuning\n\n## Overview\nTrack, analyze, and optimize LLM costs using Langfuse observability data.\n\n## Prerequisites\n- Langfuse tracing with token usage\n- Understanding of LLM pricing models\n- Access to Langfuse analytics dashboard\n\n## LLM Cost Reference\n\n| Model | Input (per 1M) | Output (per 1M) |\n|-------|----------------|-----------------|\n| GPT-4 Turbo | $10.00 | $30.00 |\n| GPT-4o | $5.00 | $15.00 |\n| GPT-4o-mini | $0.15 | $0.60 |\n| GPT-3.5 Turbo | $0.50 | $1.50 |\n| Claude 3 Opus | $15.00 | $75.00 |\n| Claude 3 Sonnet | $3.00 | $15.00 |\n| Claude 3 Haiku | $0.25 | $1.25 |\n\n## Instructions\n\n### Step 1: Track Token Usage in Generations\n\n```typescript\nimport { Langfuse } from \"langfuse\";\n\nconst langfuse = new Langfuse();\n\n\u002F\u002F Model pricing configuration\nconst MODEL_PRICING: Record\u003C\n  string,\n  { input: number; output: number }\n> = {\n  \"gpt-4-turbo\": { input: 10.0, output: 30.0 },\n  \"gpt-4o\": { input: 5.0, output: 15.0 },\n  \"gpt-4o-mini\": { input: 0.15, output: 0.6 },\n  \"gpt-3.5-turbo\": { input: 0.5, output: 1.5 },\n  \"claude-3-opus\": { input: 15.0, output: 75.0 },\n  \"claude-3-sonnet\": { input: 3.0, output: 15.0 },\n  \"claude-3-haiku\": { input: 0.25, output: 1.25 },\n};\n\nfunction calculateCost(\n  model: string,\n  promptTokens: number,\n  completionTokens: number\n): number {\n  const pricing = MODEL_PRICING[model] || { input: 0, output: 0 };\n  const inputCost = (promptTokens \u002F 1_000_000) * pricing.input;\n  const outputCost = (completionTokens \u002F 1_000_000) * pricing.output;\n  return inputCost + outputCost;\n}\n\n\u002F\u002F Track with cost metadata\nasync function tracedLLMCall(\n  trace: ReturnType\u003Ctypeof langfuse.trace>,\n  model: string,\n  messages: any[]\n) {\n  const generation = trace.generation({\n    name: \"llm-call\",\n    model,\n    input: messages,\n  });\n\n  const response = await openai.chat.completions.create({\n    model,\n    messages,\n  });\n\n  const usage = response.usage!;\n  const cost = calculateCost(model, usage.prompt_tokens, usage.completion_tokens);\n\n  generation.end({\n    output: response.choices[0].message,\n    usage: {\n      promptTokens: usage.prompt_tokens,\n      completionTokens: usage.completion_tokens,\n      totalTokens: usage.total_tokens,\n    },\n    metadata: {\n      cost_usd: cost,\n      model_version: model,\n    },\n  });\n\n  return response;\n}\n```\n\n### Step 2: Create Cost Dashboard Queries\n\n```typescript\n\u002F\u002F Fetch cost data from Langfuse\nasync function getCostAnalytics(days: number = 30) {\n  const langfuse = new Langfuse();\n\n  const fromDate = new Date();\n  fromDate.setDate(fromDate.getDate() - days);\n\n  const generations = await langfuse.fetchGenerations({\n    fromTimestamp: fromDate,\n  });\n\n  \u002F\u002F Aggregate costs\n  const costByModel: Record\u003Cstring, number> = {};\n  const costByDay: Record\u003Cstring, number> = {};\n  const tokensByModel: Record\u003Cstring, { prompt: number; completion: number }> = {};\n\n  for (const gen of generations.data) {\n    const model = gen.model || \"unknown\";\n    const date = new Date(gen.startTime).toISOString().split(\"T\")[0];\n\n    \u002F\u002F Get cost from metadata or calculate\n    const cost = gen.metadata?.cost_usd || calculateCost(\n      model,\n      gen.usage?.promptTokens || 0,\n      gen.usage?.completionTokens || 0\n    );\n\n    \u002F\u002F Aggregate by model\n    costByModel[model] = (costByModel[model] || 0) + cost;\n\n    \u002F\u002F Aggregate by day\n    costByDay[date] = (costByDay[date] || 0) + cost;\n\n    \u002F\u002F Token usage by model\n    if (!tokensByModel[model]) {\n      tokensByModel[model] = { prompt: 0, completion: 0 };\n    }\n    tokensByModel[model].prompt += gen.usage?.promptTokens || 0;\n    tokensByModel[model].completion += gen.usage?.completionTokens || 0;\n  }\n\n  return {\n    totalCost: Object.values(costByModel).reduce((a, b) => a + b, 0),\n    costByModel,\n    costByDay,\n    tokensByModel,\n    generationCount: generations.data.length,\n  };\n}\n```\n\n### Step 3: Implement Cost Alerts\n\n```typescript\ninterface CostAlert {\n  type: \"daily\" | \"hourly\" | \"per-request\";\n  threshold: number;\n  action: \"warn\" | \"block\" | \"notify\";\n}\n\nconst COST_ALERTS: CostAlert[] = [\n  { type: \"daily\", threshold: 100, action: \"warn\" },\n  { type: \"daily\", threshold: 500, action: \"notify\" },\n  { type: \"per-request\", threshold: 1, action: \"warn\" },\n];\n\nclass CostMonitor {\n  private hourlySpend: Map\u003Cstring, number> = new Map();\n  private dailySpend: Map\u003Cstring, number> = new Map();\n\n  trackCost(cost: number) {\n    const hourKey = new Date().toISOString().slice(0, 13);\n    const dayKey = new Date().toISOString().slice(0, 10);\n\n    this.hourlySpend.set(\n      hourKey,\n      (this.hourlySpend.get(hourKey) || 0) + cost\n    );\n    this.dailySpend.set(\n      dayKey,\n      (this.dailySpend.get(dayKey) || 0) + cost\n    );\n\n    this.checkAlerts(cost);\n  }\n\n  private checkAlerts(requestCost: number) {\n    const dayKey = new Date().toISOString().slice(0, 10);\n    const dailyTotal = this.dailySpend.get(dayKey) || 0;\n\n    for (const alert of COST_ALERTS) {\n      let currentValue: number;\n\n      switch (alert.type) {\n        case \"daily\":\n          currentValue = dailyTotal;\n          break;\n        case \"per-request\":\n          currentValue = requestCost;\n          break;\n        default:\n          continue;\n      }\n\n      if (currentValue >= alert.threshold) {\n        this.triggerAlert(alert, currentValue);\n      }\n    }\n  }\n\n  private triggerAlert(alert: CostAlert, value: number) {\n    const message = `Cost alert: ${alert.type} spend $${value.toFixed(2)} exceeded threshold $${alert.threshold}`;\n\n    switch (alert.action) {\n      case \"warn\":\n        console.warn(message);\n        break;\n      case \"notify\":\n        this.sendNotification(message);\n        break;\n      case \"block\":\n        throw new Error(`Request blocked: ${message}`);\n    }\n  }\n\n  private async sendNotification(message: string) {\n    \u002F\u002F Send to Slack, PagerDuty, etc.\n    await fetch(process.env.SLACK_WEBHOOK_URL!, {\n      method: \"POST\",\n      body: JSON.stringify({ text: message }),\n    });\n  }\n\n  getStats() {\n    const dayKey = new Date().toISOString().slice(0, 10);\n    return {\n      dailySpend: this.dailySpend.get(dayKey) || 0,\n      dailyBudgetRemaining: 100 - (this.dailySpend.get(dayKey) || 0),\n    };\n  }\n}\n\nconst costMonitor = new CostMonitor();\n```\n\n### Step 4: Implement Cost Optimization Strategies\n\n```typescript\n\u002F\u002F Model selection based on task complexity\ninterface ModelSelector {\n  selectModel(task: string, inputLength: number): string;\n}\n\nclass CostOptimizedModelSelector implements ModelSelector {\n  selectModel(task: string, inputLength: number): string {\n    \u002F\u002F Simple tasks -> cheaper model\n    const simpleTasks = [\"summarize\", \"classify\", \"extract\"];\n    if (simpleTasks.some((t) => task.toLowerCase().includes(t))) {\n      return \"gpt-4o-mini\";\n    }\n\n    \u002F\u002F Short inputs -> cheaper model\n    if (inputLength \u003C 500) {\n      return \"gpt-4o-mini\";\n    }\n\n    \u002F\u002F Complex tasks -> more capable model\n    const complexTasks = [\"analyze\", \"reason\", \"code\", \"math\"];\n    if (complexTasks.some((t) => task.toLowerCase().includes(t))) {\n      return \"gpt-4o\";\n    }\n\n    \u002F\u002F Default to mid-tier\n    return \"gpt-4o-mini\";\n  }\n}\n\n\u002F\u002F Prompt optimization to reduce tokens\nfunction optimizePrompt(prompt: string): string {\n  \u002F\u002F Remove excessive whitespace\n  let optimized = prompt.replace(\u002F\\s+\u002Fg, \" \").trim();\n\n  \u002F\u002F Remove redundant instructions\n  optimized = optimized.replace(\u002Fplease |kindly |could you \u002Fgi, \"\");\n\n  return optimized;\n}\n\n\u002F\u002F Caching for repeated queries\nconst responseCache = new Map\u003Cstring, { response: string; timestamp: Date }>();\n\nasync function cachedLLMCall(\n  prompt: string,\n  model: string,\n  ttlMs: number = 3600000\n): Promise\u003Cstring> {\n  const cacheKey = `${model}:${prompt}`;\n  const cached = responseCache.get(cacheKey);\n\n  if (cached && Date.now() - cached.timestamp.getTime() \u003C ttlMs) {\n    console.log(\"Cache hit - saved API call\");\n    return cached.response;\n  }\n\n  const response = await callLLM(prompt, model);\n  responseCache.set(cacheKey, { response, timestamp: new Date() });\n\n  return response;\n}\n```\n\n### Step 5: Generate Cost Reports\n\n```typescript\nasync function generateCostReport(period: \"daily\" | \"weekly\" | \"monthly\") {\n  const days = period === \"daily\" ? 1 : period === \"weekly\" ? 7 : 30;\n  const analytics = await getCostAnalytics(days);\n\n  const report = `\n# LLM Cost Report - ${period.charAt(0).toUpperCase() + period.slice(1)}\nGenerated: ${new Date().toISOString()}\n\n## Summary\n- Total Cost: $${analytics.totalCost.toFixed(2)}\n- Total Generations: ${analytics.generationCount}\n- Average Cost per Generation: $${(analytics.totalCost \u002F analytics.generationCount).toFixed(4)}\n\n## Cost by Model\n${Object.entries(analytics.costByModel)\n  .sort(([, a], [, b]) => b - a)\n  .map(([model, cost]) => `- ${model}: $${cost.toFixed(2)}`)\n  .join(\"\\n\")}\n\n## Token Usage by Model\n${Object.entries(analytics.tokensByModel)\n  .map(\n    ([model, tokens]) =>\n      `- ${model}: ${tokens.prompt.toLocaleString()} prompt, ${tokens.completion.toLocaleString()} completion`\n  )\n  .join(\"\\n\")}\n\n## Recommendations\n${generateRecommendations(analytics)}\n`;\n\n  return report;\n}\n\nfunction generateRecommendations(analytics: any): string {\n  const recommendations: string[] = [];\n\n  \u002F\u002F Check for expensive model overuse\n  const gpt4Cost = analytics.costByModel[\"gpt-4-turbo\"] || 0;\n  const totalCost = analytics.totalCost;\n\n  if (gpt4Cost > totalCost * 0.5) {\n    recommendations.push(\n      \"- Consider using GPT-4o or GPT-4o-mini for simpler tasks to reduce costs\"\n    );\n  }\n\n  \u002F\u002F Check for high output token ratio\n  for (const [model, tokens] of Object.entries(analytics.tokensByModel)) {\n    const { prompt, completion } = tokens as any;\n    if (completion > prompt * 2) {\n      recommendations.push(\n        `- ${model}: High output ratio. Consider limiting max_tokens or response length`\n      );\n    }\n  }\n\n  return recommendations.length > 0\n    ? recommendations.join(\"\\n\")\n    : \"- No immediate optimization opportunities identified\";\n}\n```\n\n## Output\n- Token usage tracking in generations\n- Cost analytics dashboard\n- Real-time cost alerts\n- Model selection optimization\n- Caching and prompt optimization\n- Automated cost reports\n\n## Cost Optimization Strategies\n\n| Strategy | Potential Savings | Implementation Effort |\n|----------|-------------------|----------------------|\n| Model downgrade | 50-90% | Low |\n| Prompt optimization | 10-30% | Low |\n| Response caching | 20-60% | Medium |\n| Batch processing | 10-20% | Medium |\n| Sampling | Variable | Medium |\n\n## Error Handling\n| Issue | Cause | Solution |\n|-------|-------|----------|\n| Missing usage data | SDK not capturing | Verify generation.end() includes usage |\n| Inaccurate costs | Wrong pricing | Update MODEL_PRICING regularly |\n| Budget exceeded | No alerts | Implement cost alerts |\n| Report failures | API limits | Add pagination to fetchGenerations |\n\n## Resources\n- [Langfuse Analytics](https:\u002F\u002Flangfuse.com\u002Fdocs\u002Fanalytics)\n- [OpenAI Pricing](https:\u002F\u002Fopenai.com\u002Fpricing)\n- [Anthropic Pricing](https:\u002F\u002Fwww.anthropic.com\u002Fpricing)\n\n## Next Steps\nFor reference architecture, see `langfuse-reference-architecture`.\n","# Langfuse Cost Tuning\n\n## Overview\nTrack, analyze, and optimize LLM costs using Langfuse observability data.\n\n## Prerequisites\n- Langfuse tracing with token usage\n- Understanding of LLM pricing models\n- Access to Langfuse analytics dashboard\n\n## LLM Cost Reference\n\n| Model | Input (per 1M) | Output (per 1M) |\n|-------|----------------|-----------------|\n| GPT-4 Turbo | $10.00 | $30.00 |\n| GPT-4o | $5.00 | $15.00 |\n| GPT-4o-mini | $0.15 | $0.60 |\n| GPT-3.5 Turbo | $0.50 | $1.50 |\n| Claude 3 Opus | $15.00 | $75.00 |\n| Claude 3 Sonnet | $3.00 | $15.00 |\n| Claude 3 Haiku | $0.25 | $1.25 |\n\n## Instructions\n\n### Step 1: Track Token Usage in Generations\n\n```typescript\nimport { Langfuse } from \"langfuse\";\n\nconst langfuse = new Langfuse();\n\n\u002F\u002F Model pricing configuration\nconst MODEL_PRICING: Record\u003C\n  string,\n  { input: number; output: number }\n> = {\n  \"gpt-4-turbo\": { input: 10.0, output: 30.0 },\n  \"gpt-4o\": { input: 5.0, output: 15.0 },\n  \"gpt-4o-mini\": { input: 0.15, output: 0.6 },\n  \"gpt-3.5-turbo\": { input: 0.5, output: 1.5 },\n  \"claude-3-opus\": { input: 15.0, output: 75.0 },\n  \"claude-3-sonnet\": { input: 3.0, output: 15.0 },\n  \"claude-3-haiku\": { input: 0.25, output: 1.25 },\n};\n\nfunction calculateCost(\n  model: string,\n  promptTokens: number,\n  completionTokens: number\n): number {\n  const pricing = MODEL_PRICING[model] || { input: 0, output: 0 };\n  const inputCost = (promptTokens \u002F 1_000_000) * pricing.input;\n  const outputCost = (completionTokens \u002F 1_000_000) * pricing.output;\n  return inputCost + outputCost;\n}\n\n\u002F\u002F Track with cost metadata\nasync function tracedLLMCall(\n  trace: ReturnType\u003Ctypeof langfuse.trace>,\n  model: string,\n  messages: any[]\n) {\n  const generation = trace.generation({\n    name: \"llm-call\",\n    model,\n    input: messages,\n  });\n\n  const response = await openai.chat.completions.create({\n    model,\n    messages,\n  });\n\n  const usage = response.usage!;\n  const cost = calculateCost(model, usage.prompt_tokens, usage.completion_tokens);\n\n  generation.end({\n    output: response.choices[0].message,\n    usage: {\n      promptTokens: usage.prompt_tokens,\n      completionTokens: usage.completion_tokens,\n      totalTokens: usage.total_tokens,\n    },\n    metadata: {\n      cost_usd: cost,\n      model_version: model,\n    },\n  });\n\n  return response;\n}\n```\n\n### Step 2: Create Cost Dashboard Queries\n\n```typescript\n\u002F\u002F Fetch cost data from Langfuse\nasync function getCostAnalytics(days: number = 30) {\n  const langfuse = new Langfuse();\n\n  const fromDate = new Date();\n  fromDate.setDate(fromDate.getDate() - days);\n\n  const generations = await langfuse.fetchGenerations({\n    fromTimestamp: fromDate,\n  });\n\n  \u002F\u002F Aggregate costs\n  const costByModel: Record\u003Cstring, number> = {};\n  const costByDay: Record\u003Cstring, number> = {};\n  const tokensByModel: Record\u003Cstring, { prompt: number; completion: number }> = {};\n\n  for (const gen of generations.data) {\n    const model = gen.model || \"unknown\";\n    const date = new Date(gen.startTime).toISOString().split(\"T\")[0];\n\n    \u002F\u002F Get cost from metadata or calculate\n    const cost = gen.metadata?.cost_usd || calculateCost(\n      model,\n      gen.usage?.promptTokens || 0,\n      gen.usage?.completionTokens || 0\n    );\n\n    \u002F\u002F Aggregate by model\n    costByModel[model] = (costByModel[model] || 0) + cost;\n\n    \u002F\u002F Aggregate by day\n    costByDay[date] = (costByDay[date] || 0) + cost;\n\n    \u002F\u002F Token usage by model\n    if (!tokensByModel[model]) {\n      tokensByModel[model] = { prompt: 0, completion: 0 };\n    }\n    tokensByModel[model].prompt += gen.usage?.promptTokens || 0;\n    tokensByModel[model].completion += gen.usage?.completionTokens || 0;\n  }\n\n  return {\n    totalCost: Object.values(costByModel).reduce((a, b) => a + b, 0),\n    costByModel,\n    costByDay,\n    tokensByModel,\n    generationCount: generations.data.length,\n  };\n}\n```\n\n### Step 3: Implement Cost Alerts\n\n```typescript\ninterface CostAlert {\n  type: \"daily\" | \"hourly\" | \"per-request\";\n  threshold: number;\n  action: \"warn\" | \"block\" | \"notify\";\n}\n\nconst COST_ALERTS: CostAlert[] = [\n  { type: \"daily\", threshold: 100, action: \"warn\" },\n  { type: \"daily\", threshold: 500, action: \"notify\" },\n  { type: \"per-request\", threshold: 1, action: \"warn\" },\n];\n\nclass CostMonitor {\n  private hourlySpend: Map\u003Cstring, number> = new Map();\n  private dailySpend: Map\u003Cstring, number> = new Map();\n\n  trackCost(cost: number) {\n    const hourKey = new Date().toISOString().slice(0, 13);\n    const dayKey = new Date().toISOString().slice(0, 10);\n\n    this.hourlySpend.set(\n      hourKey,\n      (this.hourlySpend.get(hourKey) || 0) + cost\n    );\n    this.dailySpend.set(\n      dayKey,\n      (this.dailySpend.get(dayKey) || 0) + cost\n    );\n\n    this.checkAlerts(cost);\n  }\n\n  private checkAlerts(requestCost: number) {\n    const dayKey = new Date().toISOString().slice(0, 10);\n    const dailyTotal = this.dailySpend.get(dayKey) || 0;\n\n    for (const alert of COST_ALERTS) {\n      let currentValue: number;\n\n      switch (alert.type) {\n        case \"daily\":\n          currentValue = dailyTotal;\n          break;\n        case \"per-request\":\n          currentValue = requestCost;\n          break;\n        default:\n          continue;\n      }\n\n      if (currentValue >= alert.threshold) {\n        this.triggerAlert(alert, currentValue);\n      }\n    }\n  }\n\n  private triggerAlert(alert: CostAlert, value: number) {\n    const message = `Cost alert: ${alert.type} spend $${value.toFixed(2)} exceeded threshold $${alert.threshold}`;\n\n    switch (alert.action) {\n      case \"warn\":\n        console.warn(message);\n        break;\n      case \"notify\":\n        this.sendNotification(message);\n        break;\n      case \"block\":\n        throw new Error(`Request blocked: ${message}`);\n    }\n  }\n\n  private async sendNotification(message: string) {\n    \u002F\u002F Send to Slack, PagerDuty, etc.\n    await fetch(process.env.SLACK_WEBHOOK_URL!, {\n      method: \"POST\",\n      body: JSON.stringify({ text: message }),\n    });\n  }\n\n  getStats() {\n    const dayKey = new Date().toISOString().slice(0, 10);\n    return {\n      dailySpend: this.dailySpend.get(dayKey) || 0,\n      dailyBudgetRemaining: 100 - (this.dailySpend.get(dayKey) || 0),\n    };\n  }\n}\n\nconst costMonitor = new CostMonitor();\n```\n\n### Step 4: Implement Cost Optimization Strategies\n\n```typescript\n\u002F\u002F Model selection based on task complexity\ninterface ModelSelector {\n  selectModel(task: string, inputLength: number): string;\n}\n\nclass CostOptimizedModelSelector implements ModelSelector {\n  selectModel(task: string, inputLength: number): string {\n    \u002F\u002F Simple tasks -> cheaper model\n    const simpleTasks = [\"summarize\", \"classify\", \"extract\"];\n    if (simpleTasks.some((t) => task.toLowerCase().includes(t))) {\n      return \"gpt-4o-mini\";\n    }\n\n    \u002F\u002F Short inputs -> cheaper model\n    if (inputLength \u003C 500) {\n      return \"gpt-4o-mini\";\n    }\n\n    \u002F\u002F Complex tasks -> more capable model\n    const complexTasks = [\"analyze\", \"reason\", \"code\", \"math\"];\n    if (complexTasks.some((t) => task.toLowerCase().includes(t))) {\n      return \"gpt-4o\";\n    }\n\n    \u002F\u002F Default to mid-tier\n    return \"gpt-4o-mini\";\n  }\n}\n\n\u002F\u002F Prompt optimization to reduce tokens\nfunction optimizePrompt(prompt: string): string {\n  \u002F\u002F Remove excessive whitespace\n  let optimized = prompt.replace(\u002F\\s+\u002Fg, \" \").trim();\n\n  \u002F\u002F Remove redundant instructions\n  optimized = optimized.replace(\u002Fplease |kindly |could you \u002Fgi, \"\");\n\n  return optimized;\n}\n\n\u002F\u002F Caching for repeated queries\nconst responseCache = new Map\u003Cstring, { response: string; timestamp: Date }>();\n\nasync function cachedLLMCall(\n  prompt: string,\n  model: string,\n  ttlMs: number = 3600000\n): Promise\u003Cstring> {\n  const cacheKey = `${model}:${prompt}`;\n  const cached = responseCache.get(cacheKey);\n\n  if (cached && Date.now() - cached.timestamp.getTime() \u003C ttlMs) {\n    console.log(\"Cache hit - saved API call\");\n    return cached.response;\n  }\n\n  const response = await callLLM(prompt, model);\n  responseCache.set(cacheKey, { response, timestamp: new Date() });\n\n  return response;\n}\n```\n\n### Step 5: Generate Cost Reports\n\n```typescript\nasync function generateCostReport(period: \"daily\" | \"weekly\" | \"monthly\") {\n  const days = period === \"daily\" ? 1 : period === \"weekly\" ? 7 : 30;\n  const analytics = await getCostAnalytics(days);\n\n  const report = `\n# LLM Cost Report - ${period.charAt(0).toUpperCase() + period.slice(1)}\nGenerated: ${new Date().toISOString()}\n\n## Summary\n- Total Cost: $${analytics.totalCost.toFixed(2)}\n- Total Generations: ${analytics.generationCount}\n- Average Cost per Generation: $${(analytics.totalCost \u002F analytics.generationCount).toFixed(4)}\n\n## Cost by Model\n${Object.entries(analytics.costByModel)\n  .sort(([, a], [, b]) => b - a)\n  .map(([model, cost]) => `- ${model}: $${cost.toFixed(2)}`)\n  .join(\"\\n\")}\n\n## Token Usage by Model\n${Object.entries(analytics.tokensByModel)\n  .map(\n    ([model, tokens]) =>\n      `- ${model}: ${tokens.prompt.toLocaleString()} prompt, ${tokens.completion.toLocaleString()} completion`\n  )\n  .join(\"\\n\")}\n\n## Recommendations\n${generateRecommendations(analytics)}\n`;\n\n  return report;\n}\n\nfunction generateRecommendations(analytics: any): string {\n  const recommendations: string[] = [];\n\n  \u002F\u002F Check for expensive model overuse\n  const gpt4Cost = analytics.costByModel[\"gpt-4-turbo\"] || 0;\n  const totalCost = analytics.totalCost;\n\n  if (gpt4Cost > totalCost * 0.5) {\n    recommendations.push(\n      \"- Consider using GPT-4o or GPT-4o-mini for simpler tasks to reduce costs\"\n    );\n  }\n\n  \u002F\u002F Check for high output token ratio\n  for (const [model, tokens] of Object.entries(analytics.tokensByModel)) {\n    const { prompt, completion } = tokens as any;\n    if (completion > prompt * 2) {\n      recommendations.push(\n        `- ${model}: High output ratio. Consider limiting max_tokens or response length`\n      );\n    }\n  }\n\n  return recommendations.length > 0\n    ? recommendations.join(\"\\n\")\n    : \"- No immediate optimization opportunities identified\";\n}\n```\n\n## Output\n- Token usage tracking in generations\n- Cost analytics dashboard\n- Real-time cost alerts\n- Model selection optimization\n- Caching and prompt optimization\n- Automated cost reports\n\n## Cost Optimization Strategies\n\n| Strategy | Potential Savings | Implementation Effort |\n|----------|-------------------|----------------------|\n| Model downgrade | 50-90% | Low |\n| Prompt optimization | 10-30% | Low |\n| Response caching | 20-60% | Medium |\n| Batch processing | 10-20% | Medium |\n| Sampling | Variable | Medium |\n\n## Error Handling\n| Issue | Cause | Solution |\n|-------|-------|----------|\n| Missing usage data | SDK not capturing | Verify generation.end() includes usage |\n| Inaccurate costs | Wrong pricing | Update MODEL_PRICING regularly |\n| Budget exceeded | No alerts | Implement cost alerts |\n| Report failures | API limits | Add pagination to fetchGenerations |\n\n## Resources\n- [Langfuse Analytics](https:\u002F\u002Flangfuse.com\u002Fdocs\u002Fanalytics)\n- [OpenAI Pricing](https:\u002F\u002Fopenai.com\u002Fpricing)\n- [Anthropic Pricing](https:\u002F\u002Fwww.anthropic.com\u002Fpricing)\n\n## Next Steps\nFor reference architecture, see `langfuse-reference-architecture`.\n",{"name":5,"author":60,"license":15,"version":61,"description":62,"allowed-tools":63},"Jeremy Longshore \u003Cjeremy@intentsolutions.io>","1.0.0","Monitor and optimize LLM costs using Langfuse analytics and dashboards.\nUse when tracking LLM spending, identifying cost anomalies,\nor implementing cost controls for AI applications.\nTrigger with phrases like \"langfuse costs\", \"LLM spending\",\n\"track AI costs\", \"langfuse token usage\", \"optimize LLM budget\".\n","Read, Write, Edit","main","90b50ed0778745a80fd5bfb13b1843d45b9e82f7","tests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Flangfuse-pack\u002Fskills\u002Flangfuse-cost-tuning",233,"https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust","https:\u002F\u002Fgithub.com\u002FDicklesworthstone","https:\u002F\u002Fwww.jeffreyemanuel.com\u002F","Building in NY","2026-11-15 01:54:15",{"source":8,"skillName":5,"commands":74,"prompt":84,"zipAvailable":85},[75,78,81],{"tool":76,"command":77},"npx","npx skills add Dicklesworthstone\u002Fpi_agent_rust --skill langfuse-cost-tuning",{"tool":79,"command":80},"bunx","bunx skills add Dicklesworthstone\u002Fpi_agent_rust --skill langfuse-cost-tuning",{"tool":82,"command":83},"pnpm","pnpm dlx skills add Dicklesworthstone\u002Fpi_agent_rust --skill langfuse-cost-tuning","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Fplugins-community\u002Fplugins\u002Fsaas-packs\u002Flangfuse-pack\u002Fskills\u002Flangfuse-cost-tuning\u002FSKILL.md",true,[],[88,119,153,175,201,228],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":92,"licenseFileUrl":93,"sourceHtmlUrl":94,"categories":95,"keywords":99,"publishedAt":118},"01M6FCR7G5A005P364Z6ZWCJDP","theme-factory","Toolkit for styling artifacts with a theme. These artifacts can be slides, docs, reportings, HTML landing pages, etc. There are 10 pre-set themes with colors\u002Ffonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fcreative-design\u002Ftheme-factory\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fcreative-design\u002Ftheme-factory\u002FSKILL.md",[96],{"categoryNo":97,"categoryName":98,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000015","视觉设计",[100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,40,116,117],"theme factory","factory","theme","docs","toolkit","generate","artifacts","SET","fly","slides","landing","html","styling","artifact","pages","themes","you","fonts","2026-11-11 15:11:14",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":92,"licenseFileUrl":123,"sourceHtmlUrl":124,"categories":125,"keywords":129,"publishedAt":55},"01M6FEGPPK0ZTKCCM0R3E3KS33","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Futilities\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Futilities\u002Fweb-artifacts-builder\u002FSKILL.md",[126],{"categoryNo":127,"categoryName":128,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000009","前端开发",[130,131,106,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,46,111,147,148,149,150,36,151,152],"web artifacts builder","web","artifacts builder","builder","frontend","css","components","react","component","routing","state management","state","shadcn","file","ui","Tools","Tailwind","single-file","simple","ui components","multi","jsx","suite",{"skillNo":154,"name":155,"nameZh":6,"description":156,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":92,"licenseFileUrl":157,"sourceHtmlUrl":158,"categories":159,"keywords":164,"publishedAt":55},"01M6FEE9VTK1W4GY21AKREJXRG","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fsecurity\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fsecurity\u002Fwebapp-testing\u002FSKILL.md",[160,163],{"categoryNo":161,"categoryName":162,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":127,"categoryName":128,"taxonomyType":21,"taxonomyTypeDesc":22},[165,166,167,134,168,169,131,170,104,171,144,172,52,173,174,40,36],"testing","webapp testing","webapp","screenshots","playwright","debugging","browser","logs","behavior","local",{"skillNo":176,"name":177,"nameZh":6,"description":178,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":179,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":55},"01M6FE20F7ZTQBFWKF4MPJ2SBX","neuropixels-analysis","Neuropixels neural recording analysis. Load SpikeGLX\u002FOpenEphys data, preprocess, motion correction, Kilosort4 spike sorting, quality metrics, Allen\u002FIBL curation, AI-assisted visual analysis, for Neuropixels 1.0\u002F2.0 extracellular electrophysiology. Use when working with neural recordings, spike sorting, extracellular electrophysiology, or when the user mentions Neuropixels, SpikeGLX, Open Ephys, Kilosort, quality metrics, or unit curation.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fscientific\u002Fneuropixels-analysis\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fscientific\u002Fneuropixels-analysis\u002FSKILL.md",[],[183,184,185,186,187,188,189,190,46,191,192,193,194,195,196,197,40,198,199,200,36,34],"analysis","neuropixels analysis","spike","metrics","data","open","quality","user","recordings","Neural","motion","visual","quality metrics","unit","sorting","curation","recording","load",{"skillNo":202,"name":203,"nameZh":6,"description":204,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":92,"licenseFileUrl":205,"sourceHtmlUrl":206,"categories":207,"keywords":211,"publishedAt":55},"01M6FDP0PDK6F1GGX395C7625Z","skill-creator","Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fskill-creator\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fskill-creator\u002FSKILL.md",[208],{"categoryNo":209,"categoryName":210,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[212,213,214,215,216,217,218,219,220,221,222,223,224,225,40,36,226,34,227],"skill creator","creator","workflows","create","update","capabilities","knowledge","tool","integrations","specialized","guide","new skill","existing skill","users","Existing","new",{"skillNo":229,"name":230,"nameZh":6,"description":231,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":232,"sourceHtmlUrl":233,"categories":234,"keywords":236,"publishedAt":55},"01M6FDM8Y7VK7NYG9SM7QPA3PW","notebooklm","Use this skill to query your Google NotebookLM notebooks directly from Claude Code for source-grounded, citation-backed answers from Gemini. Browser automation, library management, persistent auth. Drastically reduced hallucinations through document-only responses.","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fnotebooklm\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fpi_agent_rust\u002Fblob\u002Fmain\u002Ftests\u002Fext_conformance\u002Fartifacts\u002Ftemplates-davila7\u002Fcli-tool\u002Fcomponents\u002Fskills\u002Fproductivity\u002Fnotebooklm\u002FSKILL.md",[235],{"categoryNo":209,"categoryName":210,"taxonomyType":21,"taxonomyTypeDesc":22},[230,237,238,239,240,241,242,171,243,244,46,245,246,247,248,36,249],"google","browser automation","gemini","automation","auth","code","library","management","only","document","query","Citation","source"]