nexus-agents 3.7.0 → 3.8.0

package/dist/chunk-V7EFU7LB.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/learning/usage-log.ts"],"sourcesContent":["/**\n * usage-log — append-only per-call usage events with cost.\n *\n * Source: Issue #2469 (epic #2467 child).\n *\n * For operators running against metered API gateways, per-call cost +\n * tokens + latency is the data they need to manage spend. This module\n * provides three things:\n *\n *   1. `recordUsageEvent(event)` — append a per-call record to a JSONL\n *      log under <NEXUS_DATA_DIR>/usage/usage-YYYY-MM.jsonl.\n *   2. `loadUsageEvents({...})` — read events for a window, filtered\n *      by model / category.\n *   3. `computeCostDetail(modelId, inputTokens, outputTokens)` (and its\n *      thin wrapper `computeCostUSD`) — compute cost from the FULL model\n *      registry pricing chain (manifest > in-tree > models.dev > generated\n *      LiteLLM catalog), including the #4164 normalized/identity resolution\n *      tier for decorated gateway model ids. A model with no pricing\n *      anywhere in the chain is reported as `priced: false` so consumers\n *      can treat the missing cost as UNMEASURED (#3855), never a real $0.\n *\n * The `usage` CLI command (cli/usage-command.ts) consumes this for the\n * operator dashboard. Existing OutcomeStore is intentionally untouched\n * — its schema is for routing/learning signals, not billing.\n */\n\nimport { appendFileSync, existsSync, mkdirSync, readFileSync, readdirSync } from 'node:fs';\nimport { dirname, join } from 'node:path';\n\nimport { getNexusDataDir } from '../config/nexus-data-dir.js';\n// NOTE: only the FUNCTION is imported here — `getDefaultRegistry()` must never\n// be CALLED at module scope (#3185 bootstrap hazard: first construction reads\n// the manifest overlay/snapshot from disk). All calls happen at invocation\n// time inside computeCostDetail.\nimport { getDefaultRegistry, type MatchedVia } from '../config/model-registry.js';\n\nexport interface UsageEvent {\n  /** ISO 8601 timestamp of the call. */\n  readonly timestamp: string;\n  /** Model identifier (e.g., 'claude-sonnet-4', 'gpt-4o'). */\n  readonly modelId: string;\n  /** Provider/adapter (e.g., 'anthropic', 'openai', 'openai-compat'). */\n  readonly providerId: string;\n  /** Token counts. */\n  readonly inputTokens: number;\n  readonly outputTokens: number;\n  /**\n   * Cost in USD. Computed at write time from the full model-registry pricing\n   * chain (see `computeCostDetail`). 0 with `priced: false` means the model\n   * was UNPRICED (cost unknown), not a real $0 — check `priced` before\n   * treating this as a measured figure.\n   */\n  readonly usdCost: number;\n  /** Wall-clock latency in milliseconds. */\n  readonly latencyMs: number;\n  /** Whether the call succeeded. */\n  readonly success: boolean;\n  /**\n   * Optional task category — populated when the call was made on behalf of\n   * a routed task (so aggregation can roll up by category).\n   */\n  readonly category?: string;\n  /** Optional failure code when success === false. */\n  readonly errorCode?: string;\n  /**\n   * Whether pricing data existed for the model at write time (#4165), so\n   * audit can distinguish a real $0 from an unpriced model. Absent on lines\n   * written before this field existed.\n   */\n  readonly priced?: boolean;\n  /**\n   * Pricing provenance: the canonical registry id whose pricing was applied\n   * (the entry's `resolvedFrom` when the #4164 fuzzy tier matched a decorated\n   * id, else the caller's model id). Present only when `priced` is true.\n   */\n  readonly priceSource?: string;\n}\n\n/** Result of a pricing-aware cost computation (#4165). */\nexport interface CostDetail {\n  /** Cost in USD. Always 0 when `priced` is false — an unknown, not a $0. */\n  readonly costUsd: number;\n  /** Whether the resolved registry entry carried pricing data. */\n  readonly priced: boolean;\n  /**\n   * Canonical id the pricing/metadata came from: the entry's `resolvedFrom`\n   * when the #4164 fuzzy tier matched a decorated id, else the caller's id.\n   */\n  readonly resolvedId: string;\n  /** Fuzzy-resolution provenance (#4164), passed through from the entry. */\n  readonly matchedVia?: MatchedVia;\n}\n\n/**\n * Compute cost in USD given a model and token counts, with pricing\n * provenance. Reads the FULL registry chain via\n * `getDefaultRegistry().getEntry()` — manifest > in-tree > models.dev >\n * generated LiteLLM catalog, plus the #4164 normalized/identity tier that\n * resolves decorated model names from OpenAI-compatible gateways to their\n * canonical entry's pricing.\n *\n * `priced: false` (with `costUsd: 0`) means NO pricing existed anywhere in\n * the chain — consumers must treat the cost as UNMEASURED (#3855), never a\n * real $0. Operators can add pricing for gateway-only models via the\n * models-manifest overlay.\n */\nexport function computeCostDetail(\n  modelId: string,\n  inputTokens: number,\n  outputTokens: number\n): CostDetail {\n  // Registry read happens at INVOCATION time, never module load — first\n  // construction touches the filesystem (#3185 bootstrap hazard).\n  const entry = getDefaultRegistry().getEntry(modelId);\n  const resolvedId = entry.resolvedFrom ?? modelId;\n  const provenance = entry.matchedVia !== undefined ? { matchedVia: entry.matchedVia } : {};\n  if (entry.pricing === undefined) {\n    return { costUsd: 0, priced: false, resolvedId, ...provenance };\n  }\n  // Multiply token counts by per-million rate then divide. Use Math.round\n  // at micro-USD precision so JSONL files don't drift to floating-point\n  // noise on small calls.\n  const microUsd = Math.round(\n    inputTokens * entry.pricing.inputPer1M + outputTokens * entry.pricing.outputPer1M\n  );\n  return { costUsd: microUsd / 1_000_000, priced: true, resolvedId, ...provenance };\n}\n\n/**\n * Thin wrapper over {@link computeCostDetail} returning only the USD figure.\n * Returns 0 both for a genuinely free model and for a model with no pricing\n * data — callers that must tell those apart (unpriced ⇒ UNMEASURED, #3855)\n * should use `computeCostDetail` and check `priced`.\n */\nexport function computeCostUSD(modelId: string, inputTokens: number, outputTokens: number): number {\n  return computeCostDetail(modelId, inputTokens, outputTokens).costUsd;\n}\n\n/** Resolve the active usage log path for the current month. */\nexport function getUsageLogPath(date: Date = new Date()): string {\n  const year = date.getUTCFullYear();\n  const month = String(date.getUTCMonth() + 1).padStart(2, '0');\n  return join(getNexusDataDir(), 'usage', `usage-${String(year)}-${month}.jsonl`);\n}\n\n/**\n * Append a usage event to the current month's log. Best-effort — failures\n * are silent (we don't want to fail a successful model call because we\n * couldn't write a log line).\n */\nexport function recordUsageEvent(event: UsageEvent): void {\n  try {\n    const path = getUsageLogPath(new Date(event.timestamp));\n    mkdirSync(dirname(path), { recursive: true });\n    appendFileSync(path, `${JSON.stringify(event)}\\n`, 'utf-8');\n  } catch {\n    // Intentionally silent — telemetry must not break user calls.\n  }\n}\n\nexport interface LoadUsageOptions {\n  /** Restrict to events at or after this ISO timestamp. */\n  readonly sinceIso?: string;\n  /** Restrict to events before this ISO timestamp. */\n  readonly untilIso?: string;\n  /** Only events for this model. */\n  readonly modelId?: string;\n  /** Only events for this category. */\n  readonly category?: string;\n}\n\nfunction listUsageFiles(dir: string): readonly string[] {\n  if (!existsSync(dir)) return [];\n  try {\n    return readdirSync(dir).filter((f) => f.startsWith('usage-') && f.endsWith('.jsonl'));\n  } catch {\n    return [];\n  }\n}\n\ninterface LoadFilter {\n  readonly sinceMs: number;\n  readonly untilMs: number;\n  readonly modelId: string | undefined;\n  readonly category: string | undefined;\n}\n\nfunction eventMatches(parsed: UsageEvent, f: LoadFilter): boolean {\n  const ts = Date.parse(parsed.timestamp);\n  if (ts < f.sinceMs || ts >= f.untilMs) return false;\n  if (f.modelId !== undefined && parsed.modelId !== f.modelId) return false;\n  if (f.category !== undefined && parsed.category !== f.category) return false;\n  return true;\n}\n\nfunction parseFileLines(filePath: string, filter: LoadFilter): readonly UsageEvent[] {\n  let content: string;\n  try {\n    content = readFileSync(filePath, 'utf-8');\n  } catch {\n    return [];\n  }\n  const out: UsageEvent[] = [];\n  for (const line of content.split('\\n')) {\n    if (line.trim() === '') continue;\n    try {\n      const parsed = JSON.parse(line) as UsageEvent;\n      if (eventMatches(parsed, filter)) out.push(parsed);\n    } catch {\n      // Skip malformed line; keep reading.\n      continue;\n    }\n  }\n  return out;\n}\n\n/**\n * Load all usage events from disk that match the filter. Reads every\n * monthly log file under the data dir; for sub-second filtering at scale\n * a future PR can index by month, but linear scan is fine at the\n * \"operator dashboard\" scale this command targets.\n */\nexport function loadUsageEvents(opts: LoadUsageOptions = {}): readonly UsageEvent[] {\n  const dir = join(getNexusDataDir(), 'usage');\n  const files = listUsageFiles(dir);\n  if (files.length === 0) return [];\n  const filter: LoadFilter = {\n    sinceMs: opts.sinceIso !== undefined ? Date.parse(opts.sinceIso) : Number.NEGATIVE_INFINITY,\n    untilMs: opts.untilIso !== undefined ? Date.parse(opts.untilIso) : Number.POSITIVE_INFINITY,\n    modelId: opts.modelId,\n    category: opts.category,\n  };\n  const events: UsageEvent[] = [];\n  for (const f of files) {\n    events.push(...parseFileLines(join(dir, f), filter));\n  }\n  return events;\n}\n\nexport interface ModelRollup {\n  readonly modelId: string;\n  readonly providerId: string;\n  readonly callCount: number;\n  readonly successCount: number;\n  readonly successRate: number;\n  readonly totalInputTokens: number;\n  readonly totalOutputTokens: number;\n  readonly totalUsdCost: number;\n  readonly avgLatencyMs: number;\n  readonly costPerSuccessUsd: number;\n}\n\n/**\n * Aggregate events into per-model rollups. Sorted by total cost descending\n * — the model burning the most money at top. Useful for \"where is my spend\n * going?\" investigations.\n */\nexport function rollupByModel(events: readonly UsageEvent[]): readonly ModelRollup[] {\n  const groups = new Map<string, UsageEvent[]>();\n  for (const e of events) {\n    const arr = groups.get(e.modelId);\n    if (arr === undefined) groups.set(e.modelId, [e]);\n    else arr.push(e);\n  }\n  const rollups: ModelRollup[] = [];\n  for (const [modelId, group] of groups) {\n    const callCount = group.length;\n    const successCount = group.filter((e) => e.success).length;\n    const totalInputTokens = group.reduce((s, e) => s + e.inputTokens, 0);\n    const totalOutputTokens = group.reduce((s, e) => s + e.outputTokens, 0);\n    const totalUsdCost = group.reduce((s, e) => s + e.usdCost, 0);\n    const totalLatency = group.reduce((s, e) => s + e.latencyMs, 0);\n    const successRate = callCount === 0 ? 0 : successCount / callCount;\n    const avgLatencyMs = callCount === 0 ? 0 : totalLatency / callCount;\n    const costPerSuccessUsd = successCount === 0 ? totalUsdCost : totalUsdCost / successCount;\n    rollups.push({\n      modelId,\n      providerId: group[0]?.providerId ?? 'unknown',\n      callCount,\n      successCount,\n      successRate,\n      totalInputTokens,\n      totalOutputTokens,\n      totalUsdCost,\n      avgLatencyMs,\n      costPerSuccessUsd,\n    });\n  }\n  return rollups.sort((a, b) => b.totalUsdCost - a.totalUsdCost);\n}\n"],"mappings":";;;;;;;;AA0BA,SAAS,gBAAgB,YAAY,WAAW,cAAc,mBAAmB;AACjF,SAAS,SAAS,YAAY;AA+EvB,SAAS,kBACd,SACA,aACA,cACY;AAGZ,QAAM,QAAQ,mBAAmB,EAAE,SAAS,OAAO;AACnD,QAAM,aAAa,MAAM,gBAAgB;AACzC,QAAM,aAAa,MAAM,eAAe,SAAY,EAAE,YAAY,MAAM,WAAW,IAAI,CAAC;AACxF,MAAI,MAAM,YAAY,QAAW;AAC/B,WAAO,EAAE,SAAS,GAAG,QAAQ,OAAO,YAAY,GAAG,WAAW;AAAA,EAChE;AAIA,QAAM,WAAW,KAAK;AAAA,IACpB,cAAc,MAAM,QAAQ,aAAa,eAAe,MAAM,QAAQ;AAAA,EACxE;AACA,SAAO,EAAE,SAAS,WAAW,KAAW,QAAQ,MAAM,YAAY,GAAG,WAAW;AAClF;AAaO,SAAS,gBAAgB,OAAa,oBAAI,KAAK,GAAW;AAC/D,QAAM,OAAO,KAAK,eAAe;AACjC,QAAM,QAAQ,OAAO,KAAK,YAAY,IAAI,CAAC,EAAE,SAAS,GAAG,GAAG;AAC5D,SAAO,KAAK,gBAAgB,GAAG,SAAS,SAAS,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;AAChF;AAOO,SAAS,iBAAiB,OAAyB;AACxD,MAAI;AACF,UAAM,OAAO,gBAAgB,IAAI,KAAK,MAAM,SAAS,CAAC;AACtD,cAAU,QAAQ,IAAI,GAAG,EAAE,WAAW,KAAK,CAAC;AAC5C,mBAAe,MAAM,GAAG,KAAK,UAAU,KAAK,CAAC;AAAA,GAAM,OAAO;AAAA,EAC5D,QAAQ;AAAA,EAER;AACF;AAaA,SAAS,eAAe,KAAgC;AACtD,MAAI,CAAC,WAAW,GAAG,EAAG,QAAO,CAAC;AAC9B,MAAI;AACF,WAAO,YAAY,GAAG,EAAE,OAAO,CAAC,MAAM,EAAE,WAAW,QAAQ,KAAK,EAAE,SAAS,QAAQ,CAAC;AAAA,EACtF,QAAQ;AACN,WAAO,CAAC;AAAA,EACV;AACF;AASA,SAAS,aAAa,QAAoB,GAAwB;AAChE,QAAM,KAAK,KAAK,MAAM,OAAO,SAAS;AACtC,MAAI,KAAK,EAAE,WAAW,MAAM,EAAE,QAAS,QAAO;AAC9C,MAAI,EAAE,YAAY,UAAa,OAAO,YAAY,EAAE,QAAS,QAAO;AACpE,MAAI,EAAE,aAAa,UAAa,OAAO,aAAa,EAAE,SAAU,QAAO;AACvE,SAAO;AACT;AAEA,SAAS,eAAe,UAAkB,QAA2C;AACnF,MAAI;AACJ,MAAI;AACF,cAAU,aAAa,UAAU,OAAO;AAAA,EAC1C,QAAQ;AACN,WAAO,CAAC;AAAA,EACV;AACA,QAAM,MAAoB,CAAC;AAC3B,aAAW,QAAQ,QAAQ,MAAM,IAAI,GAAG;AACtC,QAAI,KAAK,KAAK,MAAM,GAAI;AACxB,QAAI;AACF,YAAM,SAAS,KAAK,MAAM,IAAI;AAC9B,UAAI,aAAa,QAAQ,MAAM,EAAG,KAAI,KAAK,MAAM;AAAA,IACnD,QAAQ;AAEN;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAQO,SAAS,gBAAgB,OAAyB,CAAC,GAA0B;AAClF,QAAM,MAAM,KAAK,gBAAgB,GAAG,OAAO;AAC3C,QAAM,QAAQ,eAAe,GAAG;AAChC,MAAI,MAAM,WAAW,EAAG,QAAO,CAAC;AAChC,QAAM,SAAqB;AAAA,IACzB,SAAS,KAAK,aAAa,SAAY,KAAK,MAAM,KAAK,QAAQ,IAAI,OAAO;AAAA,IAC1E,SAAS,KAAK,aAAa,SAAY,KAAK,MAAM,KAAK,QAAQ,IAAI,OAAO;AAAA,IAC1E,SAAS,KAAK;AAAA,IACd,UAAU,KAAK;AAAA,EACjB;AACA,QAAM,SAAuB,CAAC;AAC9B,aAAW,KAAK,OAAO;AACrB,WAAO,KAAK,GAAG,eAAe,KAAK,KAAK,CAAC,GAAG,MAAM,CAAC;AAAA,EACrD;AACA,SAAO;AACT;AAoBO,SAAS,cAAc,QAAuD;AACnF,QAAM,SAAS,oBAAI,IAA0B;AAC7C,aAAW,KAAK,QAAQ;AACtB,UAAM,MAAM,OAAO,IAAI,EAAE,OAAO;AAChC,QAAI,QAAQ,OAAW,QAAO,IAAI,EAAE,SAAS,CAAC,CAAC,CAAC;AAAA,QAC3C,KAAI,KAAK,CAAC;AAAA,EACjB;AACA,QAAM,UAAyB,CAAC;AAChC,aAAW,CAAC,SAAS,KAAK,KAAK,QAAQ;AACrC,UAAM,YAAY,MAAM;AACxB,UAAM,eAAe,MAAM,OAAO,CAAC,MAAM,EAAE,OAAO,EAAE;AACpD,UAAM,mBAAmB,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,aAAa,CAAC;AACpE,UAAM,oBAAoB,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,cAAc,CAAC;AACtE,UAAM,eAAe,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,SAAS,CAAC;AAC5D,UAAM,eAAe,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,EAAE,WAAW,CAAC;AAC9D,UAAM,cAAc,cAAc,IAAI,IAAI,eAAe;AACzD,UAAM,eAAe,cAAc,IAAI,IAAI,eAAe;AAC1D,UAAM,oBAAoB,iBAAiB,IAAI,eAAe,eAAe;AAC7E,YAAQ,KAAK;AAAA,MACX;AAAA,MACA,YAAY,MAAM,CAAC,GAAG,cAAc;AAAA,MACpC;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,CAAC;AAAA,EACH;AACA,SAAO,QAAQ,KAAK,CAAC,GAAG,MAAM,EAAE,eAAe,EAAE,YAAY;AAC/D;","names":[]}