aiopt 0.0.1

package/README.md

@@ -0,0 +1,74 @@
+# npx aiopt scan → your LLM cost & savings in 5 seconds
+
+```bash
+npx aiopt init
+npx aiopt scan
+cat ./aiopt-output/report.txt
+```
+
+This is a **serverless local CLI MVP**.
+- No signup, no upload, no dashboard, no server deployment.
+- Reads local JSONL/CSV → writes local outputs.
+- **No LLM calls** (math + deterministic rules only).
+
+## What you get
+After `scan`, you will have:
+1) `./aiopt-output/analysis.json` (top cost by feature/model)
+2) `./aiopt-output/report.txt` (total cost + estimated savings)
+3) `./aiopt-output/cost-policy.json` (policy file)
+
+### Sample `report.txt`
+```
+총비용: $0.23
+절감 가능 금액(Estimated): $0.21
+절감 근거 3줄:
+a) 모델 라우팅 절감(추정): $0.07
+b) 컨텍스트 감축(추정): $0.02 (상위 20% input에 25% 감축 가정)
+c) 재시도/오류 낭비: $0.13 (retries 기반)
+```
+
+## Input (default)
+- Default path: `./aiopt-input/usage.jsonl`
+- Change: `npx aiopt scan --input <path>`
+- Supported: JSONL (1 event per line), CSV
+
+### Required fields (fixed)
+`ts, provider, model, input_tokens, output_tokens, feature_tag, retries, status`
+
+`billed_cost` is optional.
+
+### JSONL example (5 lines)
+```jsonl
+{"ts":"2026-02-07T00:00:01Z","provider":"openai","model":"gpt-5-mini","input_tokens":12000,"output_tokens":1800,"feature_tag":"summarize","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:02Z","provider":"openai","model":"gpt-5.2","input_tokens":35000,"output_tokens":3000,"feature_tag":"coding","retries":1,"status":"ok"}
+{"ts":"2026-02-07T00:00:03Z","provider":"anthropic","model":"claude-sonnet","input_tokens":22000,"output_tokens":2500,"feature_tag":"classify","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:04Z","provider":"gemini","model":"gemini-1.5-flash","input_tokens":9000,"output_tokens":1200,"feature_tag":"translate","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:05Z","provider":"openai","model":"unknown-model-x","input_tokens":8000,"output_tokens":1000,"feature_tag":"summarize","retries":2,"status":"error"}
+```
+
+## Outputs (fixed)
+- `analysis.json`
+  - `total_cost`
+  - `by_model_top` (top 10)
+  - `by_feature_top` (top 10)
+  - `unknown_models`
+  - `rate_table_version`, `rate_table_date`
+- `report.txt`
+  - total cost
+  - estimated savings
+  - 3-line rationale
+- `cost-policy.json`
+  - `version, default_provider, rules, budgets, generated_from`
+
+## Rate table
+- `./rates/rate_table.json`
+- Unknown models are marked as `Estimated` and listed in `unknown_models`.
+
+## Local dev
+```bash
+cd aiopt
+npm i
+npm run build
+node dist/cli.js init
+node dist/cli.js scan --input ./aiopt-input/usage.jsonl
+```

package/dist/cli.js

@@ -0,0 +1,316 @@
+#!/usr/bin/env node
+"use strict";
+var __create = Object.create;
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
+
+// src/cli.ts
+var import_fs3 = __toESM(require("fs"));
+var import_path3 = __toESM(require("path"));
+var import_commander = require("commander");
+
+// src/io.ts
+var import_fs = __toESM(require("fs"));
+var import_path = __toESM(require("path"));
+var import_sync = require("csv-parse/sync");
+function ensureDir(p) {
+  import_fs.default.mkdirSync(p, { recursive: true });
+}
+function readJsonl(filePath) {
+  const raw = import_fs.default.readFileSync(filePath, "utf8");
+  const lines = raw.split(/\r?\n/).filter((l) => l.trim().length > 0);
+  const out = [];
+  for (const line of lines) {
+    const obj = JSON.parse(line);
+    out.push(normalizeEvent(obj));
+  }
+  return out;
+}
+function readCsv(filePath) {
+  const raw = import_fs.default.readFileSync(filePath, "utf8");
+  const records = (0, import_sync.parse)(raw, { columns: true, skip_empty_lines: true, trim: true });
+  return records.map((r) => normalizeEvent(r));
+}
+function toNum(x, def = 0) {
+  const n = Number(x);
+  return Number.isFinite(n) ? n : def;
+}
+function normalizeEvent(x) {
+  return {
+    ts: String(x.ts ?? ""),
+    provider: String(x.provider ?? "").toLowerCase(),
+    model: String(x.model ?? ""),
+    input_tokens: toNum(x.input_tokens),
+    output_tokens: toNum(x.output_tokens),
+    feature_tag: String(x.feature_tag ?? ""),
+    retries: toNum(x.retries),
+    status: String(x.status ?? ""),
+    billed_cost: x.billed_cost === void 0 || x.billed_cost === "" ? void 0 : toNum(x.billed_cost)
+  };
+}
+function isCsvPath(p) {
+  return import_path.default.extname(p).toLowerCase() === ".csv";
+}
+
+// src/scan.ts
+var import_fs2 = __toESM(require("fs"));
+var import_path2 = __toESM(require("path"));
+
+// src/cost.ts
+function getRates(rt, provider, model) {
+  const p = rt.providers[provider];
+  if (!p) return null;
+  const m = p.models[model];
+  if (m) return { kind: "official", input: m.input, output: m.output };
+  return { kind: "estimated", input: p.default_estimated.input, output: p.default_estimated.output };
+}
+function costOfEvent(rt, ev) {
+  if (typeof ev.billed_cost === "number" && Number.isFinite(ev.billed_cost)) {
+    return {
+      cost: ev.billed_cost,
+      used_rate: {
+        kind: "billed_cost",
+        provider: ev.provider,
+        model: ev.model,
+        input_per_m: 0,
+        output_per_m: 0
+      }
+    };
+  }
+  const r = getRates(rt, ev.provider, ev.model);
+  if (!r) {
+    const input_per_m = 1;
+    const output_per_m = 4;
+    const cost2 = ev.input_tokens / 1e6 * input_per_m + ev.output_tokens / 1e6 * output_per_m;
+    return {
+      cost: cost2,
+      used_rate: { kind: "estimated", provider: ev.provider, model: ev.model, input_per_m, output_per_m }
+    };
+  }
+  const cost = ev.input_tokens / 1e6 * r.input + ev.output_tokens / 1e6 * r.output;
+  return {
+    cost,
+    used_rate: {
+      kind: r.kind,
+      provider: ev.provider,
+      model: ev.model,
+      input_per_m: r.input,
+      output_per_m: r.output
+    }
+  };
+}
+
+// src/scan.ts
+var ROUTE_TO_CHEAP_FEATURES = /* @__PURE__ */ new Set(["summarize", "classify", "translate"]);
+function topN(map, n) {
+  return [...map.entries()].map(([key, v]) => ({ key, cost: v.cost, events: v.events })).sort((a, b) => b.cost - a.cost).slice(0, n);
+}
+function analyze(rt, events) {
+  const byModel = /* @__PURE__ */ new Map();
+  const byFeature = /* @__PURE__ */ new Map();
+  const unknownModels = [];
+  const perEventCosts = [];
+  let total = 0;
+  for (const ev of events) {
+    const cr = costOfEvent(rt, ev);
+    total += cr.cost;
+    perEventCosts.push({ ev, cost: cr.cost });
+    const mk = `${ev.provider}:${ev.model}`;
+    const fk = ev.feature_tag || "(none)";
+    const mv = byModel.get(mk) || { cost: 0, events: 0 };
+    mv.cost += cr.cost;
+    mv.events += 1;
+    byModel.set(mk, mv);
+    const fv = byFeature.get(fk) || { cost: 0, events: 0 };
+    fv.cost += cr.cost;
+    fv.events += 1;
+    byFeature.set(fk, fv);
+    const rr = getRates(rt, ev.provider, ev.model);
+    if (!rr) {
+      unknownModels.push({ provider: ev.provider, model: ev.model, reason: "unknown provider (estimated)" });
+    } else if (rr.kind === "estimated") {
+      unknownModels.push({ provider: ev.provider, model: ev.model, reason: "unknown model (estimated)" });
+    }
+  }
+  let routingSavings = 0;
+  for (const { ev } of perEventCosts) {
+    if (!ROUTE_TO_CHEAP_FEATURES.has(String(ev.feature_tag || "").toLowerCase())) continue;
+    const provider = ev.provider;
+    const p = rt.providers[provider];
+    if (!p) continue;
+    const entries = Object.entries(p.models);
+    if (entries.length === 0) continue;
+    const cheapest = entries.map(([name, r]) => ({ name, score: (r.input + r.output) / 2, r })).sort((a, b) => a.score - b.score)[0];
+    const currentRate = getRates(rt, provider, ev.model);
+    if (!currentRate) continue;
+    if (currentRate.kind === "estimated") continue;
+    const currentCost = ev.input_tokens / 1e6 * currentRate.input + ev.output_tokens / 1e6 * currentRate.output;
+    const cheapCost = ev.input_tokens / 1e6 * cheapest.r.input + ev.output_tokens / 1e6 * cheapest.r.output;
+    const diff = currentCost - cheapCost;
+    if (diff > 0) routingSavings += diff;
+  }
+  const sortedByInput = [...events].sort((a, b) => (b.input_tokens || 0) - (a.input_tokens || 0));
+  const k = Math.max(1, Math.floor(sortedByInput.length * 0.2));
+  const contextTargets = sortedByInput.slice(0, k);
+  let contextSavings = 0;
+  for (const ev of contextTargets) {
+    const r = getRates(rt, ev.provider, ev.model);
+    if (!r) continue;
+    const inputPerM = r.input;
+    const saveTokens = (ev.input_tokens || 0) * 0.25;
+    contextSavings += saveTokens / 1e6 * inputPerM;
+  }
+  let retryWaste = 0;
+  for (const ev of events) {
+    const retries = Number(ev.retries || 0);
+    if (retries <= 0) continue;
+    const base = costOfEvent(rt, { ...ev, retries: 0 }).cost;
+    retryWaste += base * retries;
+  }
+  const estimatedSavingsTotal = routingSavings + contextSavings + retryWaste;
+  const analysis = {
+    total_cost: round2(total),
+    by_model_top: topN(byModel, 10).map((x) => ({ ...x, cost: round2(x.cost) })),
+    by_feature_top: topN(byFeature, 10).map((x) => ({ ...x, cost: round2(x.cost) })),
+    unknown_models: uniqUnknown(unknownModels),
+    rate_table_version: rt.version,
+    rate_table_date: rt.date
+  };
+  const savings = {
+    estimated_savings_total: round2(estimatedSavingsTotal),
+    routing_savings: round2(routingSavings),
+    context_savings: round2(contextSavings),
+    retry_waste: round2(retryWaste),
+    notes: [
+      `a) \uBAA8\uB378 \uB77C\uC6B0\uD305 \uC808\uAC10(\uCD94\uC815): $${round2(routingSavings)}`,
+      `b) \uCEE8\uD14D\uC2A4\uD2B8 \uAC10\uCD95(\uCD94\uC815): $${round2(contextSavings)} (\uC0C1\uC704 20% input\uC5D0 25% \uAC10\uCD95 \uAC00\uC815)`,
+      `c) \uC7AC\uC2DC\uB3C4/\uC624\uB958 \uB0AD\uBE44: $${round2(retryWaste)} (retries \uAE30\uBC18)`
+    ]
+  };
+  const policy = buildPolicy(rt, events);
+  return { analysis, savings, policy };
+}
+function buildPolicy(rt, events) {
+  const freq = /* @__PURE__ */ new Map();
+  for (const ev of events) freq.set(ev.provider, (freq.get(ev.provider) || 0) + 1);
+  const defaultProvider = [...freq.entries()].sort((a, b) => b[1] - a[1])[0]?.[0] || "openai";
+  const rules = [];
+  for (const provider of Object.keys(rt.providers)) {
+    const p = rt.providers[provider];
+    const entries = Object.entries(p.models);
+    if (entries.length === 0) continue;
+    const cheapest = entries.map(([name, r]) => ({ name, score: (r.input + r.output) / 2, r })).sort((a, b) => a.score - b.score)[0];
+    rules.push({
+      match: { provider, feature_tag_in: ["summarize", "classify", "translate"] },
+      action: { recommend_model: cheapest.name, reason: "cheap-feature routing" }
+    });
+  }
+  rules.push({ match: { model_unknown: true }, action: { keep: true, reason: "unknown model -> no policy applied" } });
+  return {
+    version: 1,
+    default_provider: defaultProvider,
+    rules,
+    budgets: { currency: rt.currency, notes: "MVP: budgets not enforced" },
+    generated_from: { rate_table_version: rt.version, input: "./aiopt-input/usage.jsonl" }
+  };
+}
+function uniqUnknown(list) {
+  const seen = /* @__PURE__ */ new Set();
+  const out = [];
+  for (const x of list) {
+    const k = `${x.provider}:${x.model}:${x.reason}`;
+    if (seen.has(k)) continue;
+    seen.add(k);
+    out.push(x);
+  }
+  return out;
+}
+function round2(n) {
+  return Math.round(n * 100) / 100;
+}
+function writeOutputs(outDir, analysis, savings, policy) {
+  import_fs2.default.mkdirSync(outDir, { recursive: true });
+  import_fs2.default.writeFileSync(import_path2.default.join(outDir, "analysis.json"), JSON.stringify(analysis, null, 2));
+  const report = [
+    `\uCD1D\uBE44\uC6A9: $${analysis.total_cost}`,
+    `\uC808\uAC10 \uAC00\uB2A5 \uAE08\uC561(Estimated): $${savings.estimated_savings_total}`,
+    `\uC808\uAC10 \uADFC\uAC70 3\uC904:`,
+    savings.notes[0],
+    savings.notes[1],
+    savings.notes[2],
+    ""
+  ].join("\n");
+  import_fs2.default.writeFileSync(import_path2.default.join(outDir, "report.txt"), report);
+  import_fs2.default.writeFileSync(import_path2.default.join(outDir, "cost-policy.json"), JSON.stringify(policy, null, 2));
+}
+
+// src/cli.ts
+var program = new import_commander.Command();
+var DEFAULT_INPUT = "./aiopt-input/usage.jsonl";
+var DEFAULT_OUTPUT_DIR = "./aiopt-output";
+function loadRateTable() {
+  const p = import_path3.default.join(__dirname, "..", "rates", "rate_table.json");
+  return JSON.parse(import_fs3.default.readFileSync(p, "utf8"));
+}
+program.name("aiopt").description("AI \uBE44\uC6A9 \uC790\uB3D9 \uC808\uAC10 \uC778\uD504\uB77C \u2014 \uC11C\uBC84 \uC5C6\uB294 \uB85C\uCEEC CLI MVP").version("0.0.1");
+program.command("init").description("aiopt-input/ \uBC0F \uC0D8\uD50C usage.jsonl, aiopt-output/ \uC0DD\uC131").action(() => {
+  ensureDir("./aiopt-input");
+  ensureDir("./aiopt-output");
+  const sampleSrc = import_path3.default.join(__dirname, "..", "samples", "sample_usage.jsonl");
+  const dst = import_path3.default.join("./aiopt-input", "usage.jsonl");
+  if (!import_fs3.default.existsSync(dst)) {
+    import_fs3.default.copyFileSync(sampleSrc, dst);
+    console.log("Created ./aiopt-input/usage.jsonl (sample)");
+  } else {
+    console.log("Exists ./aiopt-input/usage.jsonl (skip)");
+  }
+  console.log("Ready: ./aiopt-output/");
+});
+program.command("scan").description("\uC785\uB825 \uB85C\uADF8(JSONL/CSV)\uB97C \uBD84\uC11D\uD558\uACE0 3\uAC1C \uC0B0\uCD9C\uBB3C \uC0DD\uC131").option("--input <path>", "input file path (default: ./aiopt-input/usage.jsonl)", DEFAULT_INPUT).option("--out <dir>", "output dir (default: ./aiopt-output)", DEFAULT_OUTPUT_DIR).action((opts) => {
+  const inputPath = String(opts.input);
+  const outDir = String(opts.out);
+  if (!import_fs3.default.existsSync(inputPath)) {
+    console.error(`Input not found: ${inputPath}`);
+    process.exit(1);
+  }
+  const rt = loadRateTable();
+  const events = isCsvPath(inputPath) ? readCsv(inputPath) : readJsonl(inputPath);
+  const { analysis, savings, policy } = analyze(rt, events);
+  policy.generated_from.input = inputPath;
+  writeOutputs(outDir, analysis, savings, policy);
+  console.log(`OK: ${outDir}/analysis.json`);
+  console.log(`OK: ${outDir}/report.txt`);
+  console.log(`OK: ${outDir}/cost-policy.json`);
+});
+program.command("policy").description("\uB9C8\uC9C0\uB9C9 scan \uACB0\uACFC \uAE30\uBC18\uC73C\uB85C cost-policy.json\uB9CC \uC7AC\uC0DD\uC131 (MVP: scan\uACFC \uB3D9\uC77C \uB85C\uC9C1)").option("--input <path>", "input file path (default: ./aiopt-input/usage.jsonl)", DEFAULT_INPUT).option("--out <dir>", "output dir (default: ./aiopt-output)", DEFAULT_OUTPUT_DIR).action((opts) => {
+  const inputPath = String(opts.input);
+  const outDir = String(opts.out);
+  const rt = loadRateTable();
+  const events = isCsvPath(inputPath) ? readCsv(inputPath) : readJsonl(inputPath);
+  const { policy } = analyze(rt, events);
+  policy.generated_from.input = inputPath;
+  ensureDir(outDir);
+  import_fs3.default.writeFileSync(import_path3.default.join(outDir, "cost-policy.json"), JSON.stringify(policy, null, 2));
+  console.log(`OK: ${outDir}/cost-policy.json`);
+});
+program.parse(process.argv);
+//# sourceMappingURL=cli.js.map

package/dist/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/cli.ts","../src/io.ts","../src/scan.ts","../src/cost.ts"],"sourcesContent":["#!/usr/bin/env node\n\nimport fs from 'fs';\nimport path from 'path';\nimport { Command } from 'commander';\nimport { ensureDir, isCsvPath, readCsv, readJsonl } from './io';\nimport { RateTable } from './types';\nimport { analyze, writeOutputs } from './scan';\n\nconst program = new Command();\n\nconst DEFAULT_INPUT = './aiopt-input/usage.jsonl';\nconst DEFAULT_OUTPUT_DIR = './aiopt-output';\n\nfunction loadRateTable(): RateTable {\n  const p = path.join(__dirname, '..', 'rates', 'rate_table.json');\n  return JSON.parse(fs.readFileSync(p, 'utf8'));\n}\n\nprogram\n  .name('aiopt')\n  .description('AI 비용 자동 절감 인프라 — 서버 없는 로컬 CLI MVP')\n  .version('0.0.1');\n\nprogram\n  .command('init')\n  .description('aiopt-input/ 및 샘플 usage.jsonl, aiopt-output/ 생성')\n  .action(() => {\n    ensureDir('./aiopt-input');\n    ensureDir('./aiopt-output');\n\n    const sampleSrc = path.join(__dirname, '..', 'samples', 'sample_usage.jsonl');\n    const dst = path.join('./aiopt-input', 'usage.jsonl');\n\n    if (!fs.existsSync(dst)) {\n      fs.copyFileSync(sampleSrc, dst);\n      console.log('Created ./aiopt-input/usage.jsonl (sample)');\n    } else {\n      console.log('Exists ./aiopt-input/usage.jsonl (skip)');\n    }\n\n    console.log('Ready: ./aiopt-output/');\n  });\n\nprogram\n  .command('scan')\n  .description('입력 로그(JSONL/CSV)를 분석하고 3개 산출물 생성')\n  .option('--input <path>', 'input file path (default: ./aiopt-input/usage.jsonl)', DEFAULT_INPUT)\n  .option('--out <dir>', 'output dir (default: ./aiopt-output)', DEFAULT_OUTPUT_DIR)\n  .action((opts) => {\n    const inputPath = String(opts.input);\n    const outDir = String(opts.out);\n\n    if (!fs.existsSync(inputPath)) {\n      console.error(`Input not found: ${inputPath}`);\n      process.exit(1);\n    }\n\n    const rt = loadRateTable();\n    const events = isCsvPath(inputPath) ? readCsv(inputPath) : readJsonl(inputPath);\n\n    const { analysis, savings, policy } = analyze(rt, events);\n    // For reproducibility: embed input path & rate table meta\n    policy.generated_from.input = inputPath;\n\n    writeOutputs(outDir, analysis, savings, policy);\n\n    console.log(`OK: ${outDir}/analysis.json`);\n    console.log(`OK: ${outDir}/report.txt`);\n    console.log(`OK: ${outDir}/cost-policy.json`);\n  });\n\nprogram\n  .command('policy')\n  .description('마지막 scan 결과 기반으로 cost-policy.json만 재생성 (MVP: scan과 동일 로직)')\n  .option('--input <path>', 'input file path (default: ./aiopt-input/usage.jsonl)', DEFAULT_INPUT)\n  .option('--out <dir>', 'output dir (default: ./aiopt-output)', DEFAULT_OUTPUT_DIR)\n  .action((opts) => {\n    const inputPath = String(opts.input);\n    const outDir = String(opts.out);\n    const rt = loadRateTable();\n    const events = isCsvPath(inputPath) ? readCsv(inputPath) : readJsonl(inputPath);\n    const { policy } = analyze(rt, events);\n    policy.generated_from.input = inputPath;\n\n    ensureDir(outDir);\n    fs.writeFileSync(path.join(outDir, 'cost-policy.json'), JSON.stringify(policy, null, 2));\n    console.log(`OK: ${outDir}/cost-policy.json`);\n  });\n\nprogram.parse(process.argv);\n","import fs from 'fs';\nimport path from 'path';\nimport { parse as parseCsv } from 'csv-parse/sync';\nimport { UsageEvent } from './types';\n\nexport function ensureDir(p: string) {\n  fs.mkdirSync(p, { recursive: true });\n}\n\nexport function readJsonl(filePath: string): UsageEvent[] {\n  const raw = fs.readFileSync(filePath, 'utf8');\n  const lines = raw.split(/\\r?\\n/).filter(l => l.trim().length > 0);\n  const out: UsageEvent[] = [];\n  for (const line of lines) {\n    const obj = JSON.parse(line);\n    out.push(normalizeEvent(obj));\n  }\n  return out;\n}\n\nexport function readCsv(filePath: string): UsageEvent[] {\n  const raw = fs.readFileSync(filePath, 'utf8');\n  const records = parseCsv(raw, { columns: true, skip_empty_lines: true, trim: true });\n  return records.map((r: any) => normalizeEvent(r));\n}\n\nfunction toNum(x: any, def = 0): number {\n  const n = Number(x);\n  return Number.isFinite(n) ? n : def;\n}\n\nfunction normalizeEvent(x: any): UsageEvent {\n  return {\n    ts: String(x.ts ?? ''),\n    provider: String(x.provider ?? '').toLowerCase(),\n    model: String(x.model ?? ''),\n    input_tokens: toNum(x.input_tokens),\n    output_tokens: toNum(x.output_tokens),\n    feature_tag: String(x.feature_tag ?? ''),\n    retries: toNum(x.retries),\n    status: String(x.status ?? ''),\n    billed_cost: x.billed_cost === undefined || x.billed_cost === '' ? undefined : toNum(x.billed_cost)\n  };\n}\n\nexport function isCsvPath(p: string) {\n  return path.extname(p).toLowerCase() === '.csv';\n}\n","import fs from 'fs';\nimport path from 'path';\nimport { RateTable, UsageEvent } from './types';\nimport { costOfEvent, getRates } from './cost';\n\nexport type AnalysisJson = {\n  total_cost: number;\n  by_model_top: Array<{ key: string; cost: number; events: number }>;\n  by_feature_top: Array<{ key: string; cost: number; events: number }>;\n  unknown_models: Array<{ provider: string; model: string; reason: string }>;\n  rate_table_version: string;\n  rate_table_date: string;\n};\n\nexport type Savings = {\n  estimated_savings_total: number;\n  routing_savings: number;\n  context_savings: number;\n  retry_waste: number;\n  notes: [string, string, string];\n};\n\nexport type PolicyJson = {\n  version: number;\n  default_provider: string;\n  rules: Array<any>;\n  budgets: { currency: string; notes?: string };\n  generated_from: { rate_table_version: string; input: string };\n};\n\nconst ROUTE_TO_CHEAP_FEATURES = new Set(['summarize', 'classify', 'translate']);\n\nfunction topN(map: Map<string, { cost: number; events: number }>, n: number) {\n  return [...map.entries()]\n    .map(([key, v]) => ({ key, cost: v.cost, events: v.events }))\n    .sort((a, b) => b.cost - a.cost)\n    .slice(0, n);\n}\n\nexport function analyze(rt: RateTable, events: UsageEvent[]): { analysis: AnalysisJson; savings: Savings; policy: PolicyJson } {\n  const byModel = new Map<string, { cost: number; events: number }>();\n  const byFeature = new Map<string, { cost: number; events: number }>();\n  const unknownModels: AnalysisJson['unknown_models'] = [];\n\n  const perEventCosts: Array<{ ev: UsageEvent; cost: number }> = [];\n\n  let total = 0;\n  for (const ev of events) {\n    const cr = costOfEvent(rt, ev);\n    total += cr.cost;\n    perEventCosts.push({ ev, cost: cr.cost });\n\n    const mk = `${ev.provider}:${ev.model}`;\n    const fk = ev.feature_tag || '(none)';\n\n    const mv = byModel.get(mk) || { cost: 0, events: 0 };\n    mv.cost += cr.cost; mv.events += 1;\n    byModel.set(mk, mv);\n\n    const fv = byFeature.get(fk) || { cost: 0, events: 0 };\n    fv.cost += cr.cost; fv.events += 1;\n    byFeature.set(fk, fv);\n\n    const rr = getRates(rt, ev.provider, ev.model);\n    if (!rr) {\n      unknownModels.push({ provider: ev.provider, model: ev.model, reason: 'unknown provider (estimated)' });\n    } else if (rr.kind === 'estimated') {\n      unknownModels.push({ provider: ev.provider, model: ev.model, reason: 'unknown model (estimated)' });\n    }\n  }\n\n  // --- Savings lever 1: routing (cheap features -> cheapest known model per provider)\n  let routingSavings = 0;\n  for (const { ev } of perEventCosts) {\n    if (!ROUTE_TO_CHEAP_FEATURES.has(String(ev.feature_tag || '').toLowerCase())) continue;\n\n    const provider = ev.provider;\n    const p = rt.providers[provider];\n    if (!p) continue;\n\n    // Choose cheapest model by (input+output) average.\n    const entries = Object.entries(p.models);\n    if (entries.length === 0) continue;\n    const cheapest = entries\n      .map(([name, r]) => ({ name, score: (r.input + r.output) / 2, r }))\n      .sort((a, b) => a.score - b.score)[0];\n\n    // current cost vs cheapest cost\n    const currentRate = getRates(rt, provider, ev.model);\n    if (!currentRate) continue;\n    if (currentRate.kind === 'estimated') continue; // unknown model: policy not applied\n\n    const currentCost = (ev.input_tokens / 1e6) * currentRate.input + (ev.output_tokens / 1e6) * currentRate.output;\n    const cheapCost = (ev.input_tokens / 1e6) * cheapest.r.input + (ev.output_tokens / 1e6) * cheapest.r.output;\n    const diff = currentCost - cheapCost;\n    if (diff > 0) routingSavings += diff;\n  }\n\n  // --- Savings lever 2: context reduction estimate\n  // Deterministic: top 20% by input_tokens -> reduce input_tokens by 25%\n  const sortedByInput = [...events].sort((a, b) => (b.input_tokens || 0) - (a.input_tokens || 0));\n  const k = Math.max(1, Math.floor(sortedByInput.length * 0.2));\n  const contextTargets = sortedByInput.slice(0, k);\n  let contextSavings = 0;\n  for (const ev of contextTargets) {\n    const r = getRates(rt, ev.provider, ev.model);\n    if (!r) continue;\n    const inputPerM = r.input;\n    const saveTokens = (ev.input_tokens || 0) * 0.25;\n    contextSavings += (saveTokens / 1e6) * inputPerM;\n  }\n\n  // --- Savings lever 3: retry waste\n  // retries>=1 -> wasted cost = base cost per call * retries\n  let retryWaste = 0;\n  for (const ev of events) {\n    const retries = Number(ev.retries || 0);\n    if (retries <= 0) continue;\n    const base = costOfEvent(rt, { ...ev, retries: 0 }).cost;\n    retryWaste += base * retries;\n  }\n\n  const estimatedSavingsTotal = routingSavings + contextSavings + retryWaste;\n\n  const analysis: AnalysisJson = {\n    total_cost: round2(total),\n    by_model_top: topN(byModel, 10).map(x => ({ ...x, cost: round2(x.cost) })),\n    by_feature_top: topN(byFeature, 10).map(x => ({ ...x, cost: round2(x.cost) })),\n    unknown_models: uniqUnknown(unknownModels),\n    rate_table_version: rt.version,\n    rate_table_date: rt.date\n  };\n\n  const savings: Savings = {\n    estimated_savings_total: round2(estimatedSavingsTotal),\n    routing_savings: round2(routingSavings),\n    context_savings: round2(contextSavings),\n    retry_waste: round2(retryWaste),\n    notes: [\n      `a) 모델 라우팅 절감(추정): $${round2(routingSavings)}`,\n      `b) 컨텍스트 감축(추정): $${round2(contextSavings)} (상위 20% input에 25% 감축 가정)`,\n      `c) 재시도/오류 낭비: $${round2(retryWaste)} (retries 기반)`\n    ]\n  };\n\n  const policy: PolicyJson = buildPolicy(rt, events);\n\n  return { analysis, savings, policy };\n}\n\nfunction buildPolicy(rt: RateTable, events: UsageEvent[]): PolicyJson {\n  // Default provider: most frequent\n  const freq = new Map<string, number>();\n  for (const ev of events) freq.set(ev.provider, (freq.get(ev.provider) || 0) + 1);\n  const defaultProvider = [...freq.entries()].sort((a, b) => b[1] - a[1])[0]?.[0] || 'openai';\n\n  // For cheap features: recommend cheapest known model per provider.\n  const rules: any[] = [];\n  for (const provider of Object.keys(rt.providers)) {\n    const p = rt.providers[provider];\n    const entries = Object.entries(p.models);\n    if (entries.length === 0) continue;\n    const cheapest = entries\n      .map(([name, r]) => ({ name, score: (r.input + r.output) / 2, r }))\n      .sort((a, b) => a.score - b.score)[0];\n\n    rules.push({\n      match: { provider, feature_tag_in: ['summarize', 'classify', 'translate'] },\n      action: { recommend_model: cheapest.name, reason: 'cheap-feature routing' }\n    });\n  }\n\n  // Unknown models: keep (no policy)\n  rules.push({ match: { model_unknown: true }, action: { keep: true, reason: 'unknown model -> no policy applied' } });\n\n  return {\n    version: 1,\n    default_provider: defaultProvider,\n    rules,\n    budgets: { currency: rt.currency, notes: 'MVP: budgets not enforced' },\n    generated_from: { rate_table_version: rt.version, input: './aiopt-input/usage.jsonl' }\n  };\n}\n\nfunction uniqUnknown(list: AnalysisJson['unknown_models']) {\n  const seen = new Set<string>();\n  const out: AnalysisJson['unknown_models'] = [];\n  for (const x of list) {\n    const k = `${x.provider}:${x.model}:${x.reason}`;\n    if (seen.has(k)) continue;\n    seen.add(k);\n    out.push(x);\n  }\n  return out;\n}\n\nfunction round2(n: number) {\n  return Math.round(n * 100) / 100;\n}\n\nexport function writeOutputs(outDir: string, analysis: AnalysisJson, savings: Savings, policy: PolicyJson) {\n  fs.mkdirSync(outDir, { recursive: true });\n  fs.writeFileSync(path.join(outDir, 'analysis.json'), JSON.stringify(analysis, null, 2));\n\n  const report = [\n    `총비용: $${analysis.total_cost}`,\n    `절감 가능 금액(Estimated): $${savings.estimated_savings_total}`,\n    `절감 근거 3줄:`,\n    savings.notes[0],\n    savings.notes[1],\n    savings.notes[2],\n    ''\n  ].join('\\n');\n  fs.writeFileSync(path.join(outDir, 'report.txt'), report);\n\n  fs.writeFileSync(path.join(outDir, 'cost-policy.json'), JSON.stringify(policy, null, 2));\n}\n","import { RateTable, UsageEvent } from './types';\n\nexport type CostResult = {\n  cost: number;\n  used_rate: {\n    kind: 'billed_cost' | 'official' | 'estimated';\n    provider: string;\n    model: string;\n    input_per_m: number;\n    output_per_m: number;\n  };\n};\n\nexport function getRates(rt: RateTable, provider: string, model: string) {\n  const p = rt.providers[provider];\n  if (!p) return null;\n  const m = p.models[model];\n  if (m) return { kind: 'official' as const, input: m.input, output: m.output };\n  return { kind: 'estimated' as const, input: p.default_estimated.input, output: p.default_estimated.output };\n}\n\nexport function costOfEvent(rt: RateTable, ev: UsageEvent): CostResult {\n  if (typeof ev.billed_cost === 'number' && Number.isFinite(ev.billed_cost)) {\n    return {\n      cost: ev.billed_cost,\n      used_rate: {\n        kind: 'billed_cost',\n        provider: ev.provider,\n        model: ev.model,\n        input_per_m: 0,\n        output_per_m: 0\n      }\n    };\n  }\n\n  const r = getRates(rt, ev.provider, ev.model);\n  if (!r) {\n    // Unknown provider: deterministic fallback estimate\n    const input_per_m = 1.0;\n    const output_per_m = 4.0;\n    const cost = (ev.input_tokens / 1e6) * input_per_m + (ev.output_tokens / 1e6) * output_per_m;\n    return {\n      cost,\n      used_rate: { kind: 'estimated', provider: ev.provider, model: ev.model, input_per_m, output_per_m }\n    };\n  }\n\n  const cost = (ev.input_tokens / 1e6) * r.input + (ev.output_tokens / 1e6) * r.output;\n  return {\n    cost,\n    used_rate: {\n      kind: r.kind,\n      provider: ev.provider,\n      model: ev.model,\n      input_per_m: r.input,\n      output_per_m: r.output\n    }\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAEA,IAAAA,aAAe;AACf,IAAAC,eAAiB;AACjB,uBAAwB;;;ACJxB,gBAAe;AACf,kBAAiB;AACjB,kBAAkC;AAG3B,SAAS,UAAU,GAAW;AACnC,YAAAC,QAAG,UAAU,GAAG,EAAE,WAAW,KAAK,CAAC;AACrC;AAEO,SAAS,UAAU,UAAgC;AACxD,QAAM,MAAM,UAAAA,QAAG,aAAa,UAAU,MAAM;AAC5C,QAAM,QAAQ,IAAI,MAAM,OAAO,EAAE,OAAO,OAAK,EAAE,KAAK,EAAE,SAAS,CAAC;AAChE,QAAM,MAAoB,CAAC;AAC3B,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAM,KAAK,MAAM,IAAI;AAC3B,QAAI,KAAK,eAAe,GAAG,CAAC;AAAA,EAC9B;AACA,SAAO;AACT;AAEO,SAAS,QAAQ,UAAgC;AACtD,QAAM,MAAM,UAAAA,QAAG,aAAa,UAAU,MAAM;AAC5C,QAAM,cAAU,YAAAC,OAAS,KAAK,EAAE,SAAS,MAAM,kBAAkB,MAAM,MAAM,KAAK,CAAC;AACnF,SAAO,QAAQ,IAAI,CAAC,MAAW,eAAe,CAAC,CAAC;AAClD;AAEA,SAAS,MAAM,GAAQ,MAAM,GAAW;AACtC,QAAM,IAAI,OAAO,CAAC;AAClB,SAAO,OAAO,SAAS,CAAC,IAAI,IAAI;AAClC;AAEA,SAAS,eAAe,GAAoB;AAC1C,SAAO;AAAA,IACL,IAAI,OAAO,EAAE,MAAM,EAAE;AAAA,IACrB,UAAU,OAAO,EAAE,YAAY,EAAE,EAAE,YAAY;AAAA,IAC/C,OAAO,OAAO,EAAE,SAAS,EAAE;AAAA,IAC3B,cAAc,MAAM,EAAE,YAAY;AAAA,IAClC,eAAe,MAAM,EAAE,aAAa;AAAA,IACpC,aAAa,OAAO,EAAE,eAAe,EAAE;AAAA,IACvC,SAAS,MAAM,EAAE,OAAO;AAAA,IACxB,QAAQ,OAAO,EAAE,UAAU,EAAE;AAAA,IAC7B,aAAa,EAAE,gBAAgB,UAAa,EAAE,gBAAgB,KAAK,SAAY,MAAM,EAAE,WAAW;AAAA,EACpG;AACF;AAEO,SAAS,UAAU,GAAW;AACnC,SAAO,YAAAC,QAAK,QAAQ,CAAC,EAAE,YAAY,MAAM;AAC3C;;;AC/CA,IAAAC,aAAe;AACf,IAAAC,eAAiB;;;ACYV,SAAS,SAAS,IAAe,UAAkB,OAAe;AACvE,QAAM,IAAI,GAAG,UAAU,QAAQ;AAC/B,MAAI,CAAC,EAAG,QAAO;AACf,QAAM,IAAI,EAAE,OAAO,KAAK;AACxB,MAAI,EAAG,QAAO,EAAE,MAAM,YAAqB,OAAO,EAAE,OAAO,QAAQ,EAAE,OAAO;AAC5E,SAAO,EAAE,MAAM,aAAsB,OAAO,EAAE,kBAAkB,OAAO,QAAQ,EAAE,kBAAkB,OAAO;AAC5G;AAEO,SAAS,YAAY,IAAe,IAA4B;AACrE,MAAI,OAAO,GAAG,gBAAgB,YAAY,OAAO,SAAS,GAAG,WAAW,GAAG;AACzE,WAAO;AAAA,MACL,MAAM,GAAG;AAAA,MACT,WAAW;AAAA,QACT,MAAM;AAAA,QACN,UAAU,GAAG;AAAA,QACb,OAAO,GAAG;AAAA,QACV,aAAa;AAAA,QACb,cAAc;AAAA,MAChB;AAAA,IACF;AAAA,EACF;AAEA,QAAM,IAAI,SAAS,IAAI,GAAG,UAAU,GAAG,KAAK;AAC5C,MAAI,CAAC,GAAG;AAEN,UAAM,cAAc;AACpB,UAAM,eAAe;AACrB,UAAMC,QAAQ,GAAG,eAAe,MAAO,cAAe,GAAG,gBAAgB,MAAO;AAChF,WAAO;AAAA,MACL,MAAAA;AAAA,MACA,WAAW,EAAE,MAAM,aAAa,UAAU,GAAG,UAAU,OAAO,GAAG,OAAO,aAAa,aAAa;AAAA,IACpG;AAAA,EACF;AAEA,QAAM,OAAQ,GAAG,eAAe,MAAO,EAAE,QAAS,GAAG,gBAAgB,MAAO,EAAE;AAC9E,SAAO;AAAA,IACL;AAAA,IACA,WAAW;AAAA,MACT,MAAM,EAAE;AAAA,MACR,UAAU,GAAG;AAAA,MACb,OAAO,GAAG;AAAA,MACV,aAAa,EAAE;AAAA,MACf,cAAc,EAAE;AAAA,IAClB;AAAA,EACF;AACF;;;AD5BA,IAAM,0BAA0B,oBAAI,IAAI,CAAC,aAAa,YAAY,WAAW,CAAC;AAE9E,SAAS,KAAK,KAAoD,GAAW;AAC3E,SAAO,CAAC,GAAG,IAAI,QAAQ,CAAC,EACrB,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,MAAM,EAAE,MAAM,QAAQ,EAAE,OAAO,EAAE,EAC3D,KAAK,CAAC,GAAG,MAAM,EAAE,OAAO,EAAE,IAAI,EAC9B,MAAM,GAAG,CAAC;AACf;AAEO,SAAS,QAAQ,IAAe,QAAwF;AAC7H,QAAM,UAAU,oBAAI,IAA8C;AAClE,QAAM,YAAY,oBAAI,IAA8C;AACpE,QAAM,gBAAgD,CAAC;AAEvD,QAAM,gBAAyD,CAAC;AAEhE,MAAI,QAAQ;AACZ,aAAW,MAAM,QAAQ;AACvB,UAAM,KAAK,YAAY,IAAI,EAAE;AAC7B,aAAS,GAAG;AACZ,kBAAc,KAAK,EAAE,IAAI,MAAM,GAAG,KAAK,CAAC;AAExC,UAAM,KAAK,GAAG,GAAG,QAAQ,IAAI,GAAG,KAAK;AACrC,UAAM,KAAK,GAAG,eAAe;AAE7B,UAAM,KAAK,QAAQ,IAAI,EAAE,KAAK,EAAE,MAAM,GAAG,QAAQ,EAAE;AACnD,OAAG,QAAQ,GAAG;AAAM,OAAG,UAAU;AACjC,YAAQ,IAAI,IAAI,EAAE;AAElB,UAAM,KAAK,UAAU,IAAI,EAAE,KAAK,EAAE,MAAM,GAAG,QAAQ,EAAE;AACrD,OAAG,QAAQ,GAAG;AAAM,OAAG,UAAU;AACjC,cAAU,IAAI,IAAI,EAAE;AAEpB,UAAM,KAAK,SAAS,IAAI,GAAG,UAAU,GAAG,KAAK;AAC7C,QAAI,CAAC,IAAI;AACP,oBAAc,KAAK,EAAE,UAAU,GAAG,UAAU,OAAO,GAAG,OAAO,QAAQ,+BAA+B,CAAC;AAAA,IACvG,WAAW,GAAG,SAAS,aAAa;AAClC,oBAAc,KAAK,EAAE,UAAU,GAAG,UAAU,OAAO,GAAG,OAAO,QAAQ,4BAA4B,CAAC;AAAA,IACpG;AAAA,EACF;AAGA,MAAI,iBAAiB;AACrB,aAAW,EAAE,GAAG,KAAK,eAAe;AAClC,QAAI,CAAC,wBAAwB,IAAI,OAAO,GAAG,eAAe,EAAE,EAAE,YAAY,CAAC,EAAG;AAE9E,UAAM,WAAW,GAAG;AACpB,UAAM,IAAI,GAAG,UAAU,QAAQ;AAC/B,QAAI,CAAC,EAAG;AAGR,UAAM,UAAU,OAAO,QAAQ,EAAE,MAAM;AACvC,QAAI,QAAQ,WAAW,EAAG;AAC1B,UAAM,WAAW,QACd,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,QAAQ,EAAE,QAAQ,EAAE,UAAU,GAAG,EAAE,EAAE,EACjE,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;AAGtC,UAAM,cAAc,SAAS,IAAI,UAAU,GAAG,KAAK;AACnD,QAAI,CAAC,YAAa;AAClB,QAAI,YAAY,SAAS,YAAa;AAEtC,UAAM,cAAe,GAAG,eAAe,MAAO,YAAY,QAAS,GAAG,gBAAgB,MAAO,YAAY;AACzG,UAAM,YAAa,GAAG,eAAe,MAAO,SAAS,EAAE,QAAS,GAAG,gBAAgB,MAAO,SAAS,EAAE;AACrG,UAAM,OAAO,cAAc;AAC3B,QAAI,OAAO,EAAG,mBAAkB;AAAA,EAClC;AAIA,QAAM,gBAAgB,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,OAAO,EAAE,gBAAgB,MAAM,EAAE,gBAAgB,EAAE;AAC9F,QAAM,IAAI,KAAK,IAAI,GAAG,KAAK,MAAM,cAAc,SAAS,GAAG,CAAC;AAC5D,QAAM,iBAAiB,cAAc,MAAM,GAAG,CAAC;AAC/C,MAAI,iBAAiB;AACrB,aAAW,MAAM,gBAAgB;AAC/B,UAAM,IAAI,SAAS,IAAI,GAAG,UAAU,GAAG,KAAK;AAC5C,QAAI,CAAC,EAAG;AACR,UAAM,YAAY,EAAE;AACpB,UAAM,cAAc,GAAG,gBAAgB,KAAK;AAC5C,sBAAmB,aAAa,MAAO;AAAA,EACzC;AAIA,MAAI,aAAa;AACjB,aAAW,MAAM,QAAQ;AACvB,UAAM,UAAU,OAAO,GAAG,WAAW,CAAC;AACtC,QAAI,WAAW,EAAG;AAClB,UAAM,OAAO,YAAY,IAAI,EAAE,GAAG,IAAI,SAAS,EAAE,CAAC,EAAE;AACpD,kBAAc,OAAO;AAAA,EACvB;AAEA,QAAM,wBAAwB,iBAAiB,iBAAiB;AAEhE,QAAM,WAAyB;AAAA,IAC7B,YAAY,OAAO,KAAK;AAAA,IACxB,cAAc,KAAK,SAAS,EAAE,EAAE,IAAI,QAAM,EAAE,GAAG,GAAG,MAAM,OAAO,EAAE,IAAI,EAAE,EAAE;AAAA,IACzE,gBAAgB,KAAK,WAAW,EAAE,EAAE,IAAI,QAAM,EAAE,GAAG,GAAG,MAAM,OAAO,EAAE,IAAI,EAAE,EAAE;AAAA,IAC7E,gBAAgB,YAAY,aAAa;AAAA,IACzC,oBAAoB,GAAG;AAAA,IACvB,iBAAiB,GAAG;AAAA,EACtB;AAEA,QAAM,UAAmB;AAAA,IACvB,yBAAyB,OAAO,qBAAqB;AAAA,IACrD,iBAAiB,OAAO,cAAc;AAAA,IACtC,iBAAiB,OAAO,cAAc;AAAA,IACtC,aAAa,OAAO,UAAU;AAAA,IAC9B,OAAO;AAAA,MACL,mEAAsB,OAAO,cAAc,CAAC;AAAA,MAC5C,4DAAoB,OAAO,cAAc,CAAC;AAAA,MAC1C,qDAAkB,OAAO,UAAU,CAAC;AAAA,IACtC;AAAA,EACF;AAEA,QAAM,SAAqB,YAAY,IAAI,MAAM;AAEjD,SAAO,EAAE,UAAU,SAAS,OAAO;AACrC;AAEA,SAAS,YAAY,IAAe,QAAkC;AAEpE,QAAM,OAAO,oBAAI,IAAoB;AACrC,aAAW,MAAM,OAAQ,MAAK,IAAI,GAAG,WAAW,KAAK,IAAI,GAAG,QAAQ,KAAK,KAAK,CAAC;AAC/E,QAAM,kBAAkB,CAAC,GAAG,KAAK,QAAQ,CAAC,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,KAAK;AAGnF,QAAM,QAAe,CAAC;AACtB,aAAW,YAAY,OAAO,KAAK,GAAG,SAAS,GAAG;AAChD,UAAM,IAAI,GAAG,UAAU,QAAQ;AAC/B,UAAM,UAAU,OAAO,QAAQ,EAAE,MAAM;AACvC,QAAI,QAAQ,WAAW,EAAG;AAC1B,UAAM,WAAW,QACd,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,QAAQ,EAAE,QAAQ,EAAE,UAAU,GAAG,EAAE,EAAE,EACjE,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;AAEtC,UAAM,KAAK;AAAA,MACT,OAAO,EAAE,UAAU,gBAAgB,CAAC,aAAa,YAAY,WAAW,EAAE;AAAA,MAC1E,QAAQ,EAAE,iBAAiB,SAAS,MAAM,QAAQ,wBAAwB;AAAA,IAC5E,CAAC;AAAA,EACH;AAGA,QAAM,KAAK,EAAE,OAAO,EAAE,eAAe,KAAK,GAAG,QAAQ,EAAE,MAAM,MAAM,QAAQ,qCAAqC,EAAE,CAAC;AAEnH,SAAO;AAAA,IACL,SAAS;AAAA,IACT,kBAAkB;AAAA,IAClB;AAAA,IACA,SAAS,EAAE,UAAU,GAAG,UAAU,OAAO,4BAA4B;AAAA,IACrE,gBAAgB,EAAE,oBAAoB,GAAG,SAAS,OAAO,4BAA4B;AAAA,EACvF;AACF;AAEA,SAAS,YAAY,MAAsC;AACzD,QAAM,OAAO,oBAAI,IAAY;AAC7B,QAAM,MAAsC,CAAC;AAC7C,aAAW,KAAK,MAAM;AACpB,UAAM,IAAI,GAAG,EAAE,QAAQ,IAAI,EAAE,KAAK,IAAI,EAAE,MAAM;AAC9C,QAAI,KAAK,IAAI,CAAC,EAAG;AACjB,SAAK,IAAI,CAAC;AACV,QAAI,KAAK,CAAC;AAAA,EACZ;AACA,SAAO;AACT;AAEA,SAAS,OAAO,GAAW;AACzB,SAAO,KAAK,MAAM,IAAI,GAAG,IAAI;AAC/B;AAEO,SAAS,aAAa,QAAgB,UAAwB,SAAkB,QAAoB;AACzG,aAAAC,QAAG,UAAU,QAAQ,EAAE,WAAW,KAAK,CAAC;AACxC,aAAAA,QAAG,cAAc,aAAAC,QAAK,KAAK,QAAQ,eAAe,GAAG,KAAK,UAAU,UAAU,MAAM,CAAC,CAAC;AAEtF,QAAM,SAAS;AAAA,IACb,wBAAS,SAAS,UAAU;AAAA,IAC5B,uDAAyB,QAAQ,uBAAuB;AAAA,IACxD;AAAA,IACA,QAAQ,MAAM,CAAC;AAAA,IACf,QAAQ,MAAM,CAAC;AAAA,IACf,QAAQ,MAAM,CAAC;AAAA,IACf;AAAA,EACF,EAAE,KAAK,IAAI;AACX,aAAAD,QAAG,cAAc,aAAAC,QAAK,KAAK,QAAQ,YAAY,GAAG,MAAM;AAExD,aAAAD,QAAG,cAAc,aAAAC,QAAK,KAAK,QAAQ,kBAAkB,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AACzF;;;AF/MA,IAAM,UAAU,IAAI,yBAAQ;AAE5B,IAAM,gBAAgB;AACtB,IAAM,qBAAqB;AAE3B,SAAS,gBAA2B;AAClC,QAAM,IAAI,aAAAC,QAAK,KAAK,WAAW,MAAM,SAAS,iBAAiB;AAC/D,SAAO,KAAK,MAAM,WAAAC,QAAG,aAAa,GAAG,MAAM,CAAC;AAC9C;AAEA,QACG,KAAK,OAAO,EACZ,YAAY,oHAAoC,EAChD,QAAQ,OAAO;AAElB,QACG,QAAQ,MAAM,EACd,YAAY,0EAAiD,EAC7D,OAAO,MAAM;AACZ,YAAU,eAAe;AACzB,YAAU,gBAAgB;AAE1B,QAAM,YAAY,aAAAD,QAAK,KAAK,WAAW,MAAM,WAAW,oBAAoB;AAC5E,QAAM,MAAM,aAAAA,QAAK,KAAK,iBAAiB,aAAa;AAEpD,MAAI,CAAC,WAAAC,QAAG,WAAW,GAAG,GAAG;AACvB,eAAAA,QAAG,aAAa,WAAW,GAAG;AAC9B,YAAQ,IAAI,4CAA4C;AAAA,EAC1D,OAAO;AACL,YAAQ,IAAI,yCAAyC;AAAA,EACvD;AAEA,UAAQ,IAAI,wBAAwB;AACtC,CAAC;AAEH,QACG,QAAQ,MAAM,EACd,YAAY,6GAAkC,EAC9C,OAAO,kBAAkB,wDAAwD,aAAa,EAC9F,OAAO,eAAe,wCAAwC,kBAAkB,EAChF,OAAO,CAAC,SAAS;AAChB,QAAM,YAAY,OAAO,KAAK,KAAK;AACnC,QAAM,SAAS,OAAO,KAAK,GAAG;AAE9B,MAAI,CAAC,WAAAA,QAAG,WAAW,SAAS,GAAG;AAC7B,YAAQ,MAAM,oBAAoB,SAAS,EAAE;AAC7C,YAAQ,KAAK,CAAC;AAAA,EAChB;AAEA,QAAM,KAAK,cAAc;AACzB,QAAM,SAAS,UAAU,SAAS,IAAI,QAAQ,SAAS,IAAI,UAAU,SAAS;AAE9E,QAAM,EAAE,UAAU,SAAS,OAAO,IAAI,QAAQ,IAAI,MAAM;AAExD,SAAO,eAAe,QAAQ;AAE9B,eAAa,QAAQ,UAAU,SAAS,MAAM;AAE9C,UAAQ,IAAI,OAAO,MAAM,gBAAgB;AACzC,UAAQ,IAAI,OAAO,MAAM,aAAa;AACtC,UAAQ,IAAI,OAAO,MAAM,mBAAmB;AAC9C,CAAC;AAEH,QACG,QAAQ,QAAQ,EAChB,YAAY,qJAA2D,EACvE,OAAO,kBAAkB,wDAAwD,aAAa,EAC9F,OAAO,eAAe,wCAAwC,kBAAkB,EAChF,OAAO,CAAC,SAAS;AAChB,QAAM,YAAY,OAAO,KAAK,KAAK;AACnC,QAAM,SAAS,OAAO,KAAK,GAAG;AAC9B,QAAM,KAAK,cAAc;AACzB,QAAM,SAAS,UAAU,SAAS,IAAI,QAAQ,SAAS,IAAI,UAAU,SAAS;AAC9E,QAAM,EAAE,OAAO,IAAI,QAAQ,IAAI,MAAM;AACrC,SAAO,eAAe,QAAQ;AAE9B,YAAU,MAAM;AAChB,aAAAA,QAAG,cAAc,aAAAD,QAAK,KAAK,QAAQ,kBAAkB,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AACvF,UAAQ,IAAI,OAAO,MAAM,mBAAmB;AAC9C,CAAC;AAEH,QAAQ,MAAM,QAAQ,IAAI;","names":["import_fs","import_path","fs","parseCsv","path","import_fs","import_path","cost","fs","path","path","fs"]}

package/package.json

@@ -0,0 +1,31 @@
+{
+  "name": "aiopt",
+  "version": "0.0.1",
+  "description": "Serverless local CLI MVP for AI API cost analysis & cost-policy generation",
+  "bin": {
+    "aiopt": "dist/cli.js"
+  },
+  "type": "commonjs",
+  "main": "dist/cli.js",
+  "files": [
+    "dist",
+    "rates",
+    "samples",
+    "README.md"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "dev": "node --enable-source-maps dist/cli.js",
+    "prepack": "npm run build",
+    "test:npx": "npm pack --silent && node -e \"const fs=require('fs');const p=fs.readdirSync('.').find(f=>/^aiopt-.*\\.tgz$/.test(f)); if(!p) throw new Error('tgz not found'); console.log('tgz',p);\" && npx --yes ./$(ls -1 aiopt-*.tgz | tail -n 1) init && npx --yes ./$(ls -1 aiopt-*.tgz | tail -n 1) scan --input ./aiopt-input/usage.jsonl && test -f ./aiopt-output/report.txt && echo OK"
+  },
+  "dependencies": {
+    "commander": "^14.0.0",
+    "csv-parse": "^6.1.0"
+  },
+  "devDependencies": {
+    "@types/node": "^24.0.0",
+    "tsup": "^8.5.0",
+    "typescript": "^5.9.2"
+  }
+}

package/rates/rate_table.json

@@ -0,0 +1,30 @@
+{
+  "version": "2026-02-07-a",
+  "date": "2026-02-07",
+  "currency": "USD",
+  "units": "per_1M_tokens",
+  "notes": "MVP subset. Unknown models are treated as estimated.",
+  "providers": {
+    "openai": {
+      "default_estimated": { "input": 1.0, "output": 4.0, "cached_input": 0.1 },
+      "models": {
+        "gpt-5-mini": { "input": 0.25, "output": 2.0, "cached_input": 0.025 },
+        "gpt-5.2": { "input": 1.75, "output": 14.0, "cached_input": 0.175 }
+      }
+    },
+    "anthropic": {
+      "default_estimated": { "input": 3.0, "output": 15.0, "cached_input": 0.0 },
+      "models": {
+        "claude-sonnet": { "input": 3.0, "output": 15.0, "cached_input": 0.0 },
+        "claude-haiku": { "input": 1.0, "output": 5.0, "cached_input": 0.0 }
+      }
+    },
+    "gemini": {
+      "default_estimated": { "input": 0.5, "output": 1.5, "cached_input": 0.0 },
+      "models": {
+        "gemini-1.5-flash": { "input": 0.35, "output": 1.05, "cached_input": 0.0 },
+        "gemini-1.5-pro": { "input": 1.25, "output": 5.0, "cached_input": 0.0 }
+      }
+    }
+  }
+}

package/samples/sample_usage.jsonl

@@ -0,0 +1,5 @@
+{"ts":"2026-02-07T00:00:01Z","provider":"openai","model":"gpt-5-mini","input_tokens":12000,"output_tokens":1800,"feature_tag":"summarize","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:02Z","provider":"openai","model":"gpt-5.2","input_tokens":35000,"output_tokens":3000,"feature_tag":"coding","retries":1,"status":"ok"}
+{"ts":"2026-02-07T00:00:03Z","provider":"anthropic","model":"claude-sonnet","input_tokens":22000,"output_tokens":2500,"feature_tag":"classify","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:04Z","provider":"gemini","model":"gemini-1.5-flash","input_tokens":9000,"output_tokens":1200,"feature_tag":"translate","retries":0,"status":"ok"}
+{"ts":"2026-02-07T00:00:05Z","provider":"openai","model":"unknown-model-x","input_tokens":8000,"output_tokens":1000,"feature_tag":"summarize","retries":2,"status":"error"}