@trazum/core 2.7.3 → 2.7.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/anthropic-cost.d.ts +22 -5
- package/dist/anthropic-cost.d.ts.map +1 -1
- package/dist/anthropic-cost.js +33 -10
- package/dist/anthropic-cost.js.map +1 -1
- package/dist/anthropic-usage.d.ts +15 -2
- package/dist/anthropic-usage.d.ts.map +1 -1
- package/dist/anthropic-usage.js +41 -5
- package/dist/anthropic-usage.js.map +1 -1
- package/dist/batch-results.d.ts +77 -0
- package/dist/batch-results.d.ts.map +1 -0
- package/dist/batch-results.js +145 -0
- package/dist/batch-results.js.map +1 -0
- package/dist/billing.d.ts +10 -0
- package/dist/billing.d.ts.map +1 -1
- package/dist/billing.js +15 -0
- package/dist/billing.js.map +1 -1
- package/dist/claude-code.d.ts +4 -0
- package/dist/claude-code.d.ts.map +1 -1
- package/dist/claude-code.js +14 -1
- package/dist/claude-code.js.map +1 -1
- package/dist/connector.d.ts +19 -1
- package/dist/connector.d.ts.map +1 -1
- package/dist/connector.js +46 -6
- package/dist/connector.js.map +1 -1
- package/dist/csv.d.ts +1 -1
- package/dist/csv.d.ts.map +1 -1
- package/dist/csv.js +11 -0
- package/dist/csv.js.map +1 -1
- package/dist/gateway.d.ts +2 -0
- package/dist/gateway.d.ts.map +1 -1
- package/dist/gateway.js +16 -0
- package/dist/gateway.js.map +1 -1
- package/dist/index.d.ts +2 -0
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1 -0
- package/dist/index.js.map +1 -1
- package/dist/openai-usage.d.ts +5 -1
- package/dist/openai-usage.d.ts.map +1 -1
- package/dist/openai-usage.js +20 -4
- package/dist/openai-usage.js.map +1 -1
- package/dist/pricing-overlay.d.ts +1 -1
- package/dist/pricing-overlay.d.ts.map +1 -1
- package/dist/pricing-overlay.js +11 -0
- package/dist/pricing-overlay.js.map +1 -1
- package/dist/pricing.d.ts +8 -0
- package/dist/pricing.d.ts.map +1 -1
- package/dist/pricing.js +74 -17
- package/dist/pricing.js.map +1 -1
- package/dist/receipt.d.ts +38 -4
- package/dist/receipt.d.ts.map +1 -1
- package/dist/receipt.js +19 -1
- package/dist/receipt.js.map +1 -1
- package/dist/repeats.d.ts.map +1 -1
- package/dist/repeats.js +3 -2
- package/dist/repeats.js.map +1 -1
- package/dist/rollup.d.ts.map +1 -1
- package/dist/rollup.js +4 -0
- package/dist/rollup.js.map +1 -1
- package/dist/session-cost.d.ts.map +1 -1
- package/dist/session-cost.js +4 -2
- package/dist/session-cost.js.map +1 -1
- package/dist/store.d.ts +5 -0
- package/dist/store.d.ts.map +1 -1
- package/dist/store.js +4 -0
- package/dist/store.js.map +1 -1
- package/dist/truncation-retry.d.ts.map +1 -1
- package/dist/truncation-retry.js +3 -2
- package/dist/truncation-retry.js.map +1 -1
- package/dist/types.d.ts +13 -0
- package/dist/types.d.ts.map +1 -1
- package/dist/usage.d.ts +42 -0
- package/dist/usage.d.ts.map +1 -1
- package/dist/usage.js +42 -3
- package/dist/usage.js.map +1 -1
- package/package.json +1 -1
- package/src/anthropic-cost.ts +46 -12
- package/src/anthropic-usage.ts +54 -7
- package/src/batch-results.ts +174 -0
- package/src/billing.ts +16 -0
- package/src/claude-code.ts +16 -1
- package/src/connector.ts +68 -8
- package/src/csv.ts +11 -0
- package/src/gateway.ts +16 -0
- package/src/index.ts +2 -0
- package/src/openai-usage.ts +24 -5
- package/src/pricing-overlay.ts +15 -0
- package/src/pricing.ts +83 -17
- package/src/receipt.ts +53 -6
- package/src/repeats.ts +3 -2
- package/src/rollup.ts +4 -0
- package/src/session-cost.ts +4 -2
- package/src/store.ts +9 -0
- package/src/truncation-retry.ts +3 -2
- package/src/types.ts +13 -0
- package/src/usage.ts +84 -3
package/src/usage.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { cacheTtlOf, effectivePricing, multipliersFor } from './pricing.js';
|
|
2
|
-
import { billedAt, promptSizeOf } from './billing.js';
|
|
2
|
+
import { billedAt, promptSizeOf, webSearchUsdOf } from './billing.js';
|
|
3
3
|
import { createConversationTracker } from './conversation.js';
|
|
4
4
|
import { createOutputShapeTracker } from './output-shape.js';
|
|
5
5
|
import { createInputShapeTracker } from './input-shape.js';
|
|
@@ -201,6 +201,22 @@ export interface UsageRecord {
|
|
|
201
201
|
* not-available are the row's own price.
|
|
202
202
|
*/
|
|
203
203
|
inferenceGeo?: string;
|
|
204
|
+
/**
|
|
205
|
+
* True when the call ran on the provider's Batch API, which bills it at the
|
|
206
|
+
* row's batch ratio (half the standard rate at Anthropic and OpenAI). Read
|
|
207
|
+
* from Anthropic's `usage.service_tier: "batch"`, which every Messages
|
|
208
|
+
* response carries, or from a record-level `batch: true`, which the usage
|
|
209
|
+
* report converters write for a batch row. Absent is not batch.
|
|
210
|
+
*/
|
|
211
|
+
batch?: true;
|
|
212
|
+
/**
|
|
213
|
+
* Web searches the call ran, when the log counted any: Anthropic's
|
|
214
|
+
* `usage.server_tool_use.web_search_requests`, which every Messages
|
|
215
|
+
* response and every Claude Code log line carries once a search has run.
|
|
216
|
+
* Billed per search at the row's `webSearchPerThousandUsd`, flat, on top of
|
|
217
|
+
* the tokens. Absent is none.
|
|
218
|
+
*/
|
|
219
|
+
webSearchRequests?: number;
|
|
204
220
|
/**
|
|
205
221
|
* Set when an OpenAI-shaped line split its cache writes by lifetime and the
|
|
206
222
|
* parts did not add up to the total. The lifetime is then unknown even on a
|
|
@@ -302,6 +318,30 @@ export interface UsageBreakdown {
|
|
|
302
318
|
tierCeilingCalls: number;
|
|
303
319
|
tierCeilingUsd: number;
|
|
304
320
|
tierFloorUsd: number;
|
|
321
|
+
/**
|
|
322
|
+
* Web searches, and what they cost.
|
|
323
|
+
*
|
|
324
|
+
* Billed per search, not per token, so none of the token lines above holds
|
|
325
|
+
* them: `webSearchUsd` is its own part of `totalUsd`, at each model's
|
|
326
|
+
* `webSearchPerThousandUsd`. It is flat, the same on the Batch API, in
|
|
327
|
+
* fast mode and on US-only inference, because the provider says so for
|
|
328
|
+
* the first and states the other two for tokens only.
|
|
329
|
+
*
|
|
330
|
+
* `webSearchUnpricedRequests` are searches counted on a model whose row
|
|
331
|
+
* publishes no per-search fee. They are left out of `totalUsd` and named,
|
|
332
|
+
* the same treatment an unknown model gets, rather than billed at a rate
|
|
333
|
+
* taken from another provider.
|
|
334
|
+
*/
|
|
335
|
+
webSearchRequests: number;
|
|
336
|
+
webSearchUsd: number;
|
|
337
|
+
webSearchUnpricedRequests: number;
|
|
338
|
+
/**
|
|
339
|
+
* What the calls priced at the Batch API rate cost in tokens: already inside
|
|
340
|
+
* the four token lines above, and said apart so a provider's own batch
|
|
341
|
+
* charge can be set against it (`reconcile`) rather than against the whole
|
|
342
|
+
* difference. Their searches are in `webSearchUsd`, at the same fee as any.
|
|
343
|
+
*/
|
|
344
|
+
batchUsd: number;
|
|
305
345
|
}
|
|
306
346
|
|
|
307
347
|
/**
|
|
@@ -638,6 +678,8 @@ export interface UsageShares {
|
|
|
638
678
|
cacheRead: number;
|
|
639
679
|
cacheWrite: number;
|
|
640
680
|
output: number;
|
|
681
|
+
/** Web searches, billed per search. Zero on a bill that ran none. */
|
|
682
|
+
webSearch: number;
|
|
641
683
|
}
|
|
642
684
|
|
|
643
685
|
const EMPTY = (): UsageBreakdown => ({
|
|
@@ -663,6 +705,10 @@ const EMPTY = (): UsageBreakdown => ({
|
|
|
663
705
|
tierCeilingCalls: 0,
|
|
664
706
|
tierCeilingUsd: 0,
|
|
665
707
|
tierFloorUsd: 0,
|
|
708
|
+
webSearchRequests: 0,
|
|
709
|
+
webSearchUsd: 0,
|
|
710
|
+
webSearchUnpricedRequests: 0,
|
|
711
|
+
batchUsd: 0,
|
|
666
712
|
});
|
|
667
713
|
|
|
668
714
|
/**
|
|
@@ -858,6 +904,19 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
858
904
|
const geminiThoughts = geminiMeta ? readCount(geminiMeta.thoughtsTokenCount) : ({ kind: 'absent' } as Count);
|
|
859
905
|
const geminiToolPrompt = geminiMeta ? readCount(geminiMeta.toolUsePromptTokenCount) : ({ kind: 'absent' } as Count);
|
|
860
906
|
|
|
907
|
+
/**
|
|
908
|
+
* Anthropic's server tools, counted per use rather than per token:
|
|
909
|
+
* `usage.server_tool_use.web_search_requests` is the searches this call ran
|
|
910
|
+
* and was billed for, $10 per 1,000 on top of the tokens. Read under the
|
|
911
|
+
* same rule as every count here, so a present and unreadable figure rejects
|
|
912
|
+
* the line instead of becoming a free search.
|
|
913
|
+
*/
|
|
914
|
+
const serverTools =
|
|
915
|
+
typeof usage.server_tool_use === 'object' && usage.server_tool_use !== null
|
|
916
|
+
? (usage.server_tool_use as Record<string, unknown>)
|
|
917
|
+
: null;
|
|
918
|
+
const webSearch = serverTools ? readCount(serverTools.web_search_requests) : ({ kind: 'absent' } as Count);
|
|
919
|
+
|
|
861
920
|
const counts: Record<string, Count> = {
|
|
862
921
|
input: readCount(
|
|
863
922
|
usage.input_tokens,
|
|
@@ -882,6 +941,7 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
882
941
|
geminiToolPrompt,
|
|
883
942
|
write5m,
|
|
884
943
|
write1h,
|
|
944
|
+
webSearch,
|
|
885
945
|
};
|
|
886
946
|
|
|
887
947
|
// Any field present and unreadable rejects the line. See `readCount`.
|
|
@@ -999,6 +1059,15 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
999
1059
|
...(typeof usage.inference_geo === 'string' && usage.inference_geo !== ''
|
|
1000
1060
|
? { inferenceGeo: usage.inference_geo }
|
|
1001
1061
|
: {}),
|
|
1062
|
+
/*
|
|
1063
|
+
* The Batch API, billed at half the standard rate by both providers.
|
|
1064
|
+
* Anthropic names it in `usage.service_tier` ("standard", "priority" or
|
|
1065
|
+
* "batch"); a converted usage report row says `batch: true`. OpenAI's
|
|
1066
|
+
* response-level `service_tier` never says "batch", so the two readings
|
|
1067
|
+
* cannot collide on a flat record.
|
|
1068
|
+
*/
|
|
1069
|
+
...(record.batch === true || usage.service_tier === 'batch' ? { batch: true as const } : {}),
|
|
1070
|
+
...(valueOf(webSearch) > 0 ? { webSearchRequests: valueOf(webSearch) } : {}),
|
|
1002
1071
|
...(openAiSplitGiven && !openAiSplitAgrees ? { writeSplitContradicted: true as const } : {}),
|
|
1003
1072
|
};
|
|
1004
1073
|
}
|
|
@@ -1091,6 +1160,7 @@ function countInto(into: UsageBreakdown, record: UsageRecord): void {
|
|
|
1091
1160
|
into.stopReasonCalls += 1;
|
|
1092
1161
|
if (record.truncated) into.truncatedCalls += 1;
|
|
1093
1162
|
}
|
|
1163
|
+
into.webSearchRequests += record.webSearchRequests ?? 0;
|
|
1094
1164
|
}
|
|
1095
1165
|
|
|
1096
1166
|
function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalogue, on: Date): boolean {
|
|
@@ -1138,6 +1208,7 @@ function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalo
|
|
|
1138
1208
|
per(record.cacheWrite5mTokens, input * rates.cacheWrite5m) +
|
|
1139
1209
|
per(record.cacheWrite1hTokens, input * rates.cacheWrite1h) +
|
|
1140
1210
|
per(record.outputTokens, output);
|
|
1211
|
+
if (priced.batch === true) into.batchUsd += recordUsd(inputPerMTok, outputPerMTok);
|
|
1141
1212
|
const floor = priced.tier?.decided === false ? priced.tier.floor : undefined;
|
|
1142
1213
|
if (floor !== undefined) {
|
|
1143
1214
|
into.tierCeilingCalls += 1;
|
|
@@ -1181,8 +1252,17 @@ function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalo
|
|
|
1181
1252
|
record.cacheReadTokens + record.cacheWrite5mTokens + record.cacheWrite1hTokens,
|
|
1182
1253
|
inputPerMTok,
|
|
1183
1254
|
);
|
|
1255
|
+
/**
|
|
1256
|
+
* Web searches, per search and flat. Read off the row rather than through
|
|
1257
|
+
* `effectivePricing`, which is the token rate with every multiplier the call
|
|
1258
|
+
* earned, and not one of those multipliers reaches a search: the Batch API
|
|
1259
|
+
* prices it "the same", and US-only inference and fast mode scale tokens.
|
|
1260
|
+
*/
|
|
1261
|
+
const searches = record.webSearchRequests ?? 0;
|
|
1262
|
+
if (searches > 0 && model.webSearchPerThousandUsd === undefined) into.webSearchUnpricedRequests += searches;
|
|
1263
|
+
into.webSearchUsd += webSearchUsdOf(record, model);
|
|
1184
1264
|
into.totalUsd =
|
|
1185
|
-
into.inputUsd + into.cacheReadUsd + into.cacheWriteUsd + into.outputUsd;
|
|
1265
|
+
into.inputUsd + into.cacheReadUsd + into.cacheWriteUsd + into.outputUsd + into.webSearchUsd;
|
|
1186
1266
|
return true;
|
|
1187
1267
|
}
|
|
1188
1268
|
|
|
@@ -1630,12 +1710,13 @@ export function profileUsage(text: string, options: UsageProfileOptions): UsageP
|
|
|
1630
1710
|
*/
|
|
1631
1711
|
export function sharesOf(breakdown: UsageBreakdown): UsageShares {
|
|
1632
1712
|
const { totalUsd } = breakdown;
|
|
1633
|
-
if (totalUsd <= 0) return { input: 0, cacheRead: 0, cacheWrite: 0, output: 0 };
|
|
1713
|
+
if (totalUsd <= 0) return { input: 0, cacheRead: 0, cacheWrite: 0, output: 0, webSearch: 0 };
|
|
1634
1714
|
return {
|
|
1635
1715
|
input: breakdown.inputUsd / totalUsd,
|
|
1636
1716
|
cacheRead: breakdown.cacheReadUsd / totalUsd,
|
|
1637
1717
|
cacheWrite: breakdown.cacheWriteUsd / totalUsd,
|
|
1638
1718
|
output: breakdown.outputUsd / totalUsd,
|
|
1719
|
+
webSearch: breakdown.webSearchUsd / totalUsd,
|
|
1639
1720
|
};
|
|
1640
1721
|
}
|
|
1641
1722
|
|