@trazum/core 2.7.4 → 2.7.6
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 +7 -1
- package/dist/anthropic-usage.d.ts.map +1 -1
- package/dist/anthropic-usage.js +17 -1
- package/dist/anthropic-usage.js.map +1 -1
- package/dist/billing.d.ts +10 -0
- package/dist/billing.d.ts.map +1 -1
- package/dist/billing.js +14 -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 +34 -2
- 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/otel.d.ts +16 -4
- package/dist/otel.d.ts.map +1 -1
- package/dist/otel.js +23 -5
- package/dist/otel.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.map +1 -1
- package/dist/pricing.js +38 -0
- 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 +34 -0
- package/dist/usage.d.ts.map +1 -1
- package/dist/usage.js +34 -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 +24 -2
- package/src/billing.ts +15 -0
- package/src/claude-code.ts +16 -1
- package/src/connector.ts +56 -3
- package/src/csv.ts +11 -0
- package/src/gateway.ts +16 -0
- package/src/otel.ts +21 -5
- package/src/pricing-overlay.ts +15 -0
- package/src/pricing.ts +39 -0
- 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 +68 -3
package/src/store.ts
CHANGED
|
@@ -56,6 +56,11 @@ export interface StoreRecord {
|
|
|
56
56
|
/** False when the source reported writes without saying which TTL. */
|
|
57
57
|
ttlKnown: boolean;
|
|
58
58
|
output: number;
|
|
59
|
+
/**
|
|
60
|
+
* Web searches, billed per search on top of the tokens. Absent is none, so
|
|
61
|
+
* a store written before the field existed reads the same as it did.
|
|
62
|
+
*/
|
|
63
|
+
webSearch?: number;
|
|
59
64
|
/**
|
|
60
65
|
* The account's own opaque identifiers — workspace, key, tier — as the
|
|
61
66
|
* provider served them.
|
|
@@ -179,6 +184,7 @@ export function recordsFromBuckets(
|
|
|
179
184
|
write1h: bucket.cacheWrite1hTokens,
|
|
180
185
|
ttlKnown: bucket.writeTtlKnown,
|
|
181
186
|
output: bucket.outputTokens,
|
|
187
|
+
...((bucket.webSearchRequests ?? 0) > 0 ? { webSearch: bucket.webSearchRequests } : {}),
|
|
182
188
|
group: bucket.group,
|
|
183
189
|
pulledAtMs,
|
|
184
190
|
}));
|
|
@@ -197,6 +203,9 @@ export function bucketsFromRecords(records: readonly StoreRecord[]): UsageBucket
|
|
|
197
203
|
cacheWrite1hTokens: record.write1h,
|
|
198
204
|
writeTtlKnown: record.ttlKnown,
|
|
199
205
|
outputTokens: record.output,
|
|
206
|
+
...(typeof record.webSearch === 'number' && Number.isFinite(record.webSearch) && record.webSearch > 0
|
|
207
|
+
? { webSearchRequests: record.webSearch }
|
|
208
|
+
: {}),
|
|
200
209
|
group: record.group,
|
|
201
210
|
}));
|
|
202
211
|
}
|
package/src/truncation-retry.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { effectivePricing, multipliersFor } from './pricing.js';
|
|
2
|
-
import { billedAt, promptSizeOf } from './billing.js';
|
|
2
|
+
import { billedAt, promptSizeOf, webSearchUsdOf } from './billing.js';
|
|
3
3
|
import { UNLABELLED } from './usage.js';
|
|
4
4
|
import type { PricingCatalogue } from './pricing.js';
|
|
5
5
|
import type { UsageRecord } from './usage.js';
|
|
@@ -100,7 +100,8 @@ export function createTruncationRetryTracker(options: TruncationRetryOptions): T
|
|
|
100
100
|
per(record.cacheReadTokens, inputPerMTok * rates.cacheRead) +
|
|
101
101
|
per(record.cacheWrite5mTokens, inputPerMTok * rates.cacheWrite5m) +
|
|
102
102
|
per(record.cacheWrite1hTokens, inputPerMTok * rates.cacheWrite1h) +
|
|
103
|
-
per(record.outputTokens, outputPerMTok)
|
|
103
|
+
per(record.outputTokens, outputPerMTok) +
|
|
104
|
+
webSearchUsdOf(record, model)
|
|
104
105
|
);
|
|
105
106
|
};
|
|
106
107
|
|
package/src/types.ts
CHANGED
|
@@ -414,6 +414,19 @@ export interface ModelPricing {
|
|
|
414
414
|
* here: a row without the field prices every geography at its own rate.
|
|
415
415
|
*/
|
|
416
416
|
inferenceGeoMultipliers?: Partial<Record<InferenceGeo, number>>;
|
|
417
|
+
/**
|
|
418
|
+
* What one thousand web searches cost, in US dollars, billed per search on
|
|
419
|
+
* top of the tokens.
|
|
420
|
+
*
|
|
421
|
+
* Anthropic charges $10 per 1,000 searches on every model and counts them in
|
|
422
|
+
* `usage.server_tool_use.web_search_requests`. The fee is flat: the Batch
|
|
423
|
+
* API prices searches "the same as those in regular Messages API requests",
|
|
424
|
+
* and US-only inference and fast mode multiply token rates, which a search
|
|
425
|
+
* is not. Absent on a row whose provider publishes no such fee, and then a
|
|
426
|
+
* search the log counts is named as unpriced rather than billed at a rate
|
|
427
|
+
* borrowed from somewhere else.
|
|
428
|
+
*/
|
|
429
|
+
webSearchPerThousandUsd?: number;
|
|
417
430
|
/**
|
|
418
431
|
* Other ids that bill at exactly this rate.
|
|
419
432
|
*
|
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';
|
|
@@ -209,6 +209,14 @@ export interface UsageRecord {
|
|
|
209
209
|
* report converters write for a batch row. Absent is not batch.
|
|
210
210
|
*/
|
|
211
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;
|
|
212
220
|
/**
|
|
213
221
|
* Set when an OpenAI-shaped line split its cache writes by lifetime and the
|
|
214
222
|
* parts did not add up to the total. The lifetime is then unknown even on a
|
|
@@ -310,6 +318,30 @@ export interface UsageBreakdown {
|
|
|
310
318
|
tierCeilingCalls: number;
|
|
311
319
|
tierCeilingUsd: number;
|
|
312
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;
|
|
313
345
|
}
|
|
314
346
|
|
|
315
347
|
/**
|
|
@@ -646,6 +678,8 @@ export interface UsageShares {
|
|
|
646
678
|
cacheRead: number;
|
|
647
679
|
cacheWrite: number;
|
|
648
680
|
output: number;
|
|
681
|
+
/** Web searches, billed per search. Zero on a bill that ran none. */
|
|
682
|
+
webSearch: number;
|
|
649
683
|
}
|
|
650
684
|
|
|
651
685
|
const EMPTY = (): UsageBreakdown => ({
|
|
@@ -671,6 +705,10 @@ const EMPTY = (): UsageBreakdown => ({
|
|
|
671
705
|
tierCeilingCalls: 0,
|
|
672
706
|
tierCeilingUsd: 0,
|
|
673
707
|
tierFloorUsd: 0,
|
|
708
|
+
webSearchRequests: 0,
|
|
709
|
+
webSearchUsd: 0,
|
|
710
|
+
webSearchUnpricedRequests: 0,
|
|
711
|
+
batchUsd: 0,
|
|
674
712
|
});
|
|
675
713
|
|
|
676
714
|
/**
|
|
@@ -866,6 +904,19 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
866
904
|
const geminiThoughts = geminiMeta ? readCount(geminiMeta.thoughtsTokenCount) : ({ kind: 'absent' } as Count);
|
|
867
905
|
const geminiToolPrompt = geminiMeta ? readCount(geminiMeta.toolUsePromptTokenCount) : ({ kind: 'absent' } as Count);
|
|
868
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
|
+
|
|
869
920
|
const counts: Record<string, Count> = {
|
|
870
921
|
input: readCount(
|
|
871
922
|
usage.input_tokens,
|
|
@@ -890,6 +941,7 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
890
941
|
geminiToolPrompt,
|
|
891
942
|
write5m,
|
|
892
943
|
write1h,
|
|
944
|
+
webSearch,
|
|
893
945
|
};
|
|
894
946
|
|
|
895
947
|
// Any field present and unreadable rejects the line. See `readCount`.
|
|
@@ -1015,6 +1067,7 @@ export function parseUsageLine(line: string): UsageRecord | null {
|
|
|
1015
1067
|
* cannot collide on a flat record.
|
|
1016
1068
|
*/
|
|
1017
1069
|
...(record.batch === true || usage.service_tier === 'batch' ? { batch: true as const } : {}),
|
|
1070
|
+
...(valueOf(webSearch) > 0 ? { webSearchRequests: valueOf(webSearch) } : {}),
|
|
1018
1071
|
...(openAiSplitGiven && !openAiSplitAgrees ? { writeSplitContradicted: true as const } : {}),
|
|
1019
1072
|
};
|
|
1020
1073
|
}
|
|
@@ -1107,6 +1160,7 @@ function countInto(into: UsageBreakdown, record: UsageRecord): void {
|
|
|
1107
1160
|
into.stopReasonCalls += 1;
|
|
1108
1161
|
if (record.truncated) into.truncatedCalls += 1;
|
|
1109
1162
|
}
|
|
1163
|
+
into.webSearchRequests += record.webSearchRequests ?? 0;
|
|
1110
1164
|
}
|
|
1111
1165
|
|
|
1112
1166
|
function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalogue, on: Date): boolean {
|
|
@@ -1154,6 +1208,7 @@ function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalo
|
|
|
1154
1208
|
per(record.cacheWrite5mTokens, input * rates.cacheWrite5m) +
|
|
1155
1209
|
per(record.cacheWrite1hTokens, input * rates.cacheWrite1h) +
|
|
1156
1210
|
per(record.outputTokens, output);
|
|
1211
|
+
if (priced.batch === true) into.batchUsd += recordUsd(inputPerMTok, outputPerMTok);
|
|
1157
1212
|
const floor = priced.tier?.decided === false ? priced.tier.floor : undefined;
|
|
1158
1213
|
if (floor !== undefined) {
|
|
1159
1214
|
into.tierCeilingCalls += 1;
|
|
@@ -1197,8 +1252,17 @@ function add(into: UsageBreakdown, record: UsageRecord, catalogue: PricingCatalo
|
|
|
1197
1252
|
record.cacheReadTokens + record.cacheWrite5mTokens + record.cacheWrite1hTokens,
|
|
1198
1253
|
inputPerMTok,
|
|
1199
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);
|
|
1200
1264
|
into.totalUsd =
|
|
1201
|
-
into.inputUsd + into.cacheReadUsd + into.cacheWriteUsd + into.outputUsd;
|
|
1265
|
+
into.inputUsd + into.cacheReadUsd + into.cacheWriteUsd + into.outputUsd + into.webSearchUsd;
|
|
1202
1266
|
return true;
|
|
1203
1267
|
}
|
|
1204
1268
|
|
|
@@ -1646,12 +1710,13 @@ export function profileUsage(text: string, options: UsageProfileOptions): UsageP
|
|
|
1646
1710
|
*/
|
|
1647
1711
|
export function sharesOf(breakdown: UsageBreakdown): UsageShares {
|
|
1648
1712
|
const { totalUsd } = breakdown;
|
|
1649
|
-
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 };
|
|
1650
1714
|
return {
|
|
1651
1715
|
input: breakdown.inputUsd / totalUsd,
|
|
1652
1716
|
cacheRead: breakdown.cacheReadUsd / totalUsd,
|
|
1653
1717
|
cacheWrite: breakdown.cacheWriteUsd / totalUsd,
|
|
1654
1718
|
output: breakdown.outputUsd / totalUsd,
|
|
1719
|
+
webSearch: breakdown.webSearchUsd / totalUsd,
|
|
1655
1720
|
};
|
|
1656
1721
|
}
|
|
1657
1722
|
|