Langfuse Cost Tuning

// Automatic: observeOpenAI captures everything
import { observeOpenAI } from "@langfuse/openai";
const openai = observeOpenAI(new OpenAI());
// Tokens, model, latency, and cost are all auto-tracked

// Manual: include usage in generation observations
import { startActiveObservation, updateActiveObservation } from "@langfuse/tracing";

await startActiveObservation(
  { name: "llm-call", asType: "generation" },
  async () => {
    updateActiveObservation({ model: "gpt-4o" }); // Model required for cost calc

    const response = await openai.chat.completions.create({
      model: "gpt-4o",
      messages: [{ role: "user", content: prompt }],
    });

    updateActiveObservation({
      output: response.choices[0].message.content,
      usage: {
        promptTokens: response.usage?.prompt_tokens,
        completionTokens: response.usage?.completion_tokens,
        totalTokens: response.usage?.total_tokens,
      },
      // Optional: override inferred cost (in USD)
      // costInUsd: 0.0015,
    });
  }
);

import { LangfuseClient } from "@langfuse/client";

const langfuse = new LangfuseClient();

// Fetch aggregated cost metrics
async function getCostReport(days: number) {
  const fromTimestamp = new Date(Date.now() - days * 86400000).toISOString();

  // Use the API to list traces with cost data
  const traces = await langfuse.api.traces.list({
    fromTimestamp,
    limit: 1000,
    orderBy: "timestamp",
  });

  const costByModel = new Map<string, { cost: number; tokens: number; count: number }>();

  for (const trace of traces.data) {
    const observations = await langfuse.api.observations.list({
      traceId: trace.id,
      type: "GENERATION",
    });

    for (const obs of observations.data) {
      const model = obs.model || "unknown";
      const existing = costByModel.get(model) || { cost: 0, tokens: 0, count: 0 };
      existing.cost += obs.calculatedTotalCost || 0;
      existing.tokens += obs.totalTokens || 0;
      existing.count += 1;
      costByModel.set(model, existing);
    }
  }

  console.log("\n=== LLM Cost Report ===");
  console.log(`Period: Last ${days} days\n`);

  let totalCost = 0;
  for (const [model, data] of costByModel.entries()) {
    console.log(`${model}:`);
    console.log(`  Calls: ${data.count}`);
    console.log(`  Tokens: ${data.tokens.toLocaleString()}`);
    console.log(`  Cost: $${data.cost.toFixed(4)}`);
    totalCost += data.cost;
  }
  console.log(`\nTotal: $${totalCost.toFixed(4)}`);
}

getCostReport(7);

import { observe, updateActiveObservation } from "@langfuse/tracing";

interface ModelConfig {
  model: string;
  costPer1MInput: number;
  costPer1MOutput: number;
  maxComplexity: "simple" | "moderate" | "complex";
}

const MODELS: ModelConfig[] = [
  { model: "gpt-4o-mini", costPer1MInput: 0.15, costPer1MOutput: 0.60, maxComplexity: "simple" },
  { model: "gpt-4o", costPer1MInput: 2.50, costPer1MOutput: 10.00, maxComplexity: "moderate" },
  { model: "claude-sonnet-4-20250514", costPer1MInput: 3.00, costPer1MOutput: 15.00, maxComplexity: "complex" },
];

function selectModel(task: string, inputLength: number): ModelConfig {
  const simpleTasks = ["classify", "extract", "summarize-short", "translate"];
  const isSimple = simpleTasks.some((t) => task.includes(t));
  const isShort = inputLength < 500;

  if (isSimple && isShort) return MODELS[0]; // gpt-4o-mini
  if (isSimple || inputLength < 2000) return MODELS[1]; // gpt-4o
  return MODELS[2]; // claude-sonnet-4
}

const costOptimizedLLM = observe(
  { name: "cost-optimized-llm", asType: "generation" },
  async (task: string, input: string) => {
    const config = selectModel(task, input.length);

    updateActiveObservation({
      model: config.model,
      metadata: {
        task,
        selectedReason: `${config.maxComplexity} tier`,
        estimatedCostPer1M: config.costPer1MInput,
      },
    });

    const response = await callModel(config.model, input);
    updateActiveObservation({
      output: response.content,
      usage: response.usage,
    });

    return response;
  }
);

// scripts/cost-alert.ts -- run as cron job
import { LangfuseClient } from "@langfuse/client";

const langfuse = new LangfuseClient();

const ALERT_THRESHOLDS = {
  dailyWarn: 50,    // $50/day warning
  dailyCritical: 200, // $200/day critical
  perRequestWarn: 1,  // $1/request warning
};

async function checkCostAlerts() {
  const since = new Date(Date.now() - 86400000).toISOString(); // Last 24h

  const traces = await langfuse.api.traces.list({
    fromTimestamp: since,
    limit: 500,
  });

  let dailyCost = 0;
  let maxRequestCost = 0;

  for (const trace of traces.data) {
    const observations = await langfuse.api.observations.list({
      traceId: trace.id,
      type: "GENERATION",
    });

    const traceCost = observations.data.reduce(
      (sum, obs) => sum + (obs.calculatedTotalCost || 0), 0
    );

    dailyCost += traceCost;
    maxRequestCost = Math.max(maxRequestCost, traceCost);
  }

  console.log(`Daily cost: $${dailyCost.toFixed(2)}`);
  console.log(`Max request cost: $${maxRequestCost.toFixed(4)}`);

  if (dailyCost > ALERT_THRESHOLDS.dailyCritical) {
    await sendAlert("CRITICAL", `Daily LLM cost: $${dailyCost.toFixed(2)}`);
  } else if (dailyCost > ALERT_THRESHOLDS.dailyWarn) {
    await sendAlert("WARNING", `Daily LLM cost: $${dailyCost.toFixed(2)}`);
  }
}

checkCostAlerts();