@effect-uai/core 0.11.0 → 0.12.0

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.
Files changed (45) hide show
  1. package/dist/{Items-BWNu6VGp.d.mts → Items-BKQgqJ8A.d.mts} +21 -8
  2. package/dist/Items-BKQgqJ8A.d.mts.map +1 -0
  3. package/dist/{Tool-CXFHCQrT.d.mts → Tool-u5bag8SO.d.mts} +2 -2
  4. package/dist/{Tool-CXFHCQrT.d.mts.map → Tool-u5bag8SO.d.mts.map} +1 -1
  5. package/dist/{ToolEvent-nVijCrvm.d.mts → ToolEvent-BBfOdhF_.d.mts} +2 -2
  6. package/dist/{ToolEvent-nVijCrvm.d.mts.map → ToolEvent-BBfOdhF_.d.mts.map} +1 -1
  7. package/dist/{ToolResult-CQqKkcDa.d.mts → ToolResult-CRpmkBXR.d.mts} +2 -2
  8. package/dist/{ToolResult-CQqKkcDa.d.mts.map → ToolResult-CRpmkBXR.d.mts.map} +1 -1
  9. package/dist/{Toolkit-BYPwwu-j.d.mts → Toolkit-D4mQvhjq.d.mts} +5 -5
  10. package/dist/{Toolkit-BYPwwu-j.d.mts.map → Toolkit-D4mQvhjq.d.mts.map} +1 -1
  11. package/dist/browser/BrowserTool.d.mts +1 -1
  12. package/dist/domain/Items.d.mts +1 -1
  13. package/dist/domain/Items.mjs +15 -1
  14. package/dist/domain/Items.mjs.map +1 -1
  15. package/dist/domain/Research.d.mts +1 -1
  16. package/dist/domain/Turn.d.mts +2 -1
  17. package/dist/domain/Turn.d.mts.map +1 -1
  18. package/dist/index.d.mts +5 -5
  19. package/dist/language-model/LanguageModel.d.mts +2 -2
  20. package/dist/observability/Metrics.d.mts +13 -5
  21. package/dist/observability/Metrics.d.mts.map +1 -1
  22. package/dist/observability/Metrics.mjs +30 -7
  23. package/dist/observability/Metrics.mjs.map +1 -1
  24. package/dist/observability/Metrics.test.mjs +42 -0
  25. package/dist/observability/Metrics.test.mjs.map +1 -1
  26. package/dist/retry/Retry.d.mts +3 -3
  27. package/dist/retry/Retry.d.mts.map +1 -1
  28. package/dist/retry/Retry.mjs +30 -1
  29. package/dist/retry/Retry.mjs.map +1 -1
  30. package/dist/testing/MockProvider.d.mts +1 -1
  31. package/dist/testing/MockSandbox.d.mts +1 -1
  32. package/dist/tool/Approval.d.mts +3 -3
  33. package/dist/tool/HistoryCheck.d.mts +2 -2
  34. package/dist/tool/Tool.d.mts +1 -1
  35. package/dist/tool/ToolEvent.d.mts +1 -1
  36. package/dist/tool/ToolResult.d.mts +1 -1
  37. package/dist/tool/Toolkit.d.mts +1 -1
  38. package/dist/web-read/WebReadTool.d.mts +1 -1
  39. package/dist/web-search/WebSearchTool.d.mts +1 -1
  40. package/package.json +1 -1
  41. package/src/domain/Items.ts +18 -0
  42. package/src/observability/Metrics.test.ts +65 -0
  43. package/src/observability/Metrics.ts +53 -11
  44. package/src/retry/Retry.ts +44 -3
  45. package/dist/Items-BWNu6VGp.d.mts.map +0 -1
@@ -12,7 +12,7 @@ import {
12
12
  Stream,
13
13
  } from "effect"
14
14
  import type { Usage } from "../domain/Items.js"
15
- import type { Turn, TurnEvent } from "../domain/Turn.js"
15
+ import { type Turn, TurnEvent } from "../domain/Turn.js"
16
16
 
17
17
  // ---------------------------------------------------------------------------
18
18
  // Self-describing measurements (the exporter-facing payload)
@@ -126,6 +126,7 @@ const MetricName = {
126
126
  totalTokens: "effect_uai_total_tokens",
127
127
  reasoningTokens: "effect_uai_reasoning_tokens",
128
128
  cachedInputTokens: "effect_uai_cached_input_tokens",
129
+ cacheWriteTokens: "effect_uai_cache_write_tokens",
129
130
  } as const
130
131
 
131
132
  /** Read a structural `_tag` off any value without narrowing to a closed union. */
@@ -141,6 +142,34 @@ const contentDeltaKind = (ev: unknown): Option.Option<"text" | "reasoning"> => {
141
142
  : Option.none()
142
143
  }
143
144
 
145
+ const isTextDelta = TurnEvent.$is("TextDelta")
146
+ const isReasoningDelta = TurnEvent.$is("ReasoningDelta")
147
+ const isRefusalDelta = TurnEvent.$is("RefusalDelta")
148
+ const isToolCallArgsDelta = TurnEvent.$is("ToolCallArgsDelta")
149
+
150
+ /** A delta carrying generated output, whatever the model was producing. */
151
+ export type OutputDelta = Extract<
152
+ TurnEvent,
153
+ { readonly _tag: "TextDelta" | "ReasoningDelta" | "RefusalDelta" | "ToolCallArgsDelta" }
154
+ >
155
+
156
+ /**
157
+ * Narrow to a delta that contributes to the model's output stream. Tool-call
158
+ * arguments count: an agent turn can be almost entirely `ToolCallArgsDelta`,
159
+ * and measuring only prose reports a near-zero rate for it.
160
+ *
161
+ * The single place the output-delta tag list lives — the throughput
162
+ * accumulator and `unitCount` both read it. Distinct from `contentDeltaKind`,
163
+ * which answers a narrower question for first-token latency.
164
+ */
165
+ const outputDelta = (ev: unknown): Option.Option<OutputDelta> =>
166
+ isTextDelta(ev) || isReasoningDelta(ev) || isRefusalDelta(ev) || isToolCallArgsDelta(ev)
167
+ ? Option.some(ev)
168
+ : Option.none()
169
+
170
+ const outputTextOf = (delta: OutputDelta): string =>
171
+ isToolCallArgsDelta(delta) ? delta.delta : delta.text
172
+
144
173
  const isTurnCompleteEvent = (ev: unknown): boolean => tagOf(ev) === "TurnComplete"
145
174
 
146
175
  const turnOf = (ev: unknown): Turn =>
@@ -154,6 +183,10 @@ const addUsage = (a: Usage, b: Usage): Usage => {
154
183
  a.input_tokens_details?.cached_tokens,
155
184
  b.input_tokens_details?.cached_tokens,
156
185
  )
186
+ const cacheWrite = sumOptional(
187
+ a.input_tokens_details?.cache_write_tokens,
188
+ b.input_tokens_details?.cache_write_tokens,
189
+ )
157
190
  const reasoning = sumOptional(
158
191
  a.output_tokens_details?.reasoning_tokens,
159
192
  b.output_tokens_details?.reasoning_tokens,
@@ -162,7 +195,13 @@ const addUsage = (a: Usage, b: Usage): Usage => {
162
195
  input_tokens: sumOptional(a.input_tokens, b.input_tokens),
163
196
  output_tokens: sumOptional(a.output_tokens, b.output_tokens),
164
197
  total_tokens: sumOptional(a.total_tokens, b.total_tokens),
165
- input_tokens_details: cached === undefined ? undefined : { cached_tokens: cached },
198
+ input_tokens_details:
199
+ cached === undefined && cacheWrite === undefined
200
+ ? undefined
201
+ : {
202
+ ...(cached !== undefined && { cached_tokens: cached }),
203
+ ...(cacheWrite !== undefined && { cache_write_tokens: cacheWrite }),
204
+ },
166
205
  output_tokens_details: reasoning === undefined ? undefined : { reasoning_tokens: reasoning },
167
206
  }
168
207
  }
@@ -175,6 +214,7 @@ const usageMeasurements = (usage: Usage): ReadonlyArray<Measurement> => {
175
214
  [MetricName.totalTokens, usage.total_tokens],
176
215
  [MetricName.reasoningTokens, usage.output_tokens_details?.reasoning_tokens],
177
216
  [MetricName.cachedInputTokens, usage.input_tokens_details?.cached_tokens],
217
+ [MetricName.cacheWriteTokens, usage.input_tokens_details?.cache_write_tokens],
178
218
  ]
179
219
  return Arr.filterMap(pairs, ([name, value]) =>
180
220
  value === undefined
@@ -349,8 +389,11 @@ export type ThroughputOptions = {
349
389
  readonly every?: Duration.Input
350
390
  /** What a "unit" is. Default `"char"` (exact and chunking-independent). */
351
391
  readonly unit?: "char" | "token" | "event"
352
- /** Token counter for `unit: "token"`. Absent => approximate 1 per content delta. */
353
- readonly tokenizer?: (event: TurnEvent) => Effect.Effect<number>
392
+ /**
393
+ * Token counter for `unit: "token"`. Called only for deltas that carry
394
+ * generated output. Absent => approximate 1 per such delta.
395
+ */
396
+ readonly tokenizer?: (event: OutputDelta) => Effect.Effect<number>
354
397
  /** Rate definition. Default `"windowed"` (current speed, no cold-start bias). */
355
398
  readonly mode?: "windowed" | "cumulative"
356
399
  /** EWMA smoothing. Omitted => off; `"default"` => a sensible factor. */
@@ -434,14 +477,13 @@ const unitCount = (
434
477
  unit: "char" | "token" | "event",
435
478
  tokenizer: ThroughputOptions["tokenizer"],
436
479
  ): Effect.Effect<number> => {
437
- if (tagOf(ev) !== "TextDelta") return Effect.succeed(0)
438
- const text = (ev as { readonly text: string }).text
480
+ const output = outputDelta(ev)
481
+ if (Option.isNone(output)) return Effect.succeed(0)
482
+ const delta = output.value
439
483
  return Match.value(unit).pipe(
440
- Match.when("char", () => Effect.succeed([...text].length)),
484
+ Match.when("char", () => Effect.succeed([...outputTextOf(delta)].length)),
441
485
  Match.when("event", () => Effect.succeed(1)),
442
- Match.when("token", () =>
443
- tokenizer === undefined ? Effect.succeed(1) : tokenizer(ev as TurnEvent),
444
- ),
486
+ Match.when("token", () => (tokenizer === undefined ? Effect.succeed(1) : tokenizer(delta))),
445
487
  Match.exhaustive,
446
488
  )
447
489
  }
@@ -465,7 +507,7 @@ export const throughput =
465
507
  Stream.tap((ev) =>
466
508
  isTurnCompleteEvent(ev)
467
509
  ? Ref.update(ref, resetTurn)
468
- : tagOf(ev) === "TextDelta"
510
+ : Option.isSome(outputDelta(ev))
469
511
  ? Effect.flatMap(Clock.currentTimeMillis, (now) =>
470
512
  Effect.flatMap(unitCount(ev, unit, tokenizer), (n) =>
471
513
  Ref.update(ref, (state) => addUnits(state, n, now)),
@@ -29,14 +29,55 @@ import type * as AiError from "../domain/AiError.js"
29
29
  * variants if needed (e.g. a custom schedule that observes `cause`).
30
30
  */
31
31
  export class Retryable extends Data.TaggedError("RetryableAi")<{
32
- readonly cause: AiError.RateLimited | AiError.Unavailable | AiError.Timeout
32
+ readonly cause:
33
+ | AiError.RateLimited
34
+ | AiError.Unavailable
35
+ | AiError.Timeout
36
+ | AiError.InvalidRequest
33
37
  }> {}
34
38
 
39
+ /**
40
+ * `InvalidRequest` payloads that are a transient fault on the provider's side
41
+ * rather than a defect in the request.
42
+ *
43
+ * The class exists because a provider can only answer within its own error
44
+ * vocabulary: a constrained-decoding grammar that times out while compiling is
45
+ * a cold-start timeout, but it arrives as `invalid_request_error` because that
46
+ * is where the vendor files schema problems. Replaying it succeeds — the same
47
+ * schema compiles on the next call.
48
+ *
49
+ * An allow-list rather than a general softening of `InvalidRequest`, and it
50
+ * must stay one. The whole point of that tag being fatal is that a genuinely
51
+ * malformed request retried on a schedule burns quota to arrive at the same
52
+ * rejection, slowly. An entry earns its place by having been observed to
53
+ * succeed on replay.
54
+ */
55
+ const TRANSIENT_INVALID_REQUEST: ReadonlyArray<RegExp> = [/grammar compilation timed out/i]
56
+
57
+ /** The payload as text, for matching. `raw` is whatever the provider handed
58
+ * back — usually its response body as a string, sometimes a parsed object,
59
+ * and occasionally something that will not serialise. */
60
+ const rawText = (raw: unknown): string => {
61
+ if (typeof raw === "string") return raw
62
+ try {
63
+ return JSON.stringify(raw) ?? ""
64
+ } catch {
65
+ return String(raw)
66
+ }
67
+ }
68
+
69
+ const isTransientInvalidRequest = (e: AiError.AiError): boolean =>
70
+ e._tag === "InvalidRequest" &&
71
+ TRANSIENT_INVALID_REQUEST.some((pattern) => pattern.test(rawText(e.raw)))
72
+
35
73
  /** Type-narrowing predicate for the retryable subset. */
36
74
  export const isRetryable = (
37
75
  e: AiError.AiError,
38
- ): e is AiError.RateLimited | AiError.Unavailable | AiError.Timeout =>
39
- e._tag === "RateLimited" || e._tag === "Unavailable" || e._tag === "Timeout"
76
+ ): e is AiError.RateLimited | AiError.Unavailable | AiError.Timeout | AiError.InvalidRequest =>
77
+ e._tag === "RateLimited" ||
78
+ e._tag === "Unavailable" ||
79
+ e._tag === "Timeout" ||
80
+ isTransientInvalidRequest(e)
40
81
 
41
82
  // Lift events to Items, non-retryable failures to Terminal values
42
83
  // (escape retry), retryable failures to wrapped errors (only thing
@@ -1 +0,0 @@
1
- {"version":3,"file":"Items-BWNu6VGp.d.mts","names":[],"sources":["../src/domain/Items.ts"],"mappings":";;;;;;;cAOa,SAAA,EAAS,MAAA,CAAA,MAAA;EAAA;;;KAIV,SAAA,UAAmB,SAAA,CAAU,IAAA;;;;;;;;;cAU5B,UAAA,EAAU,MAAA,CAAA,MAAA;EAAA;;;KAIX,UAAA,UAAoB,UAAA,CAAW,IAAA;;;;;;;;;cAgB9B,WAAA,EAAW,MAAA,CAAA,MAAA;EAAA;;;;;;;;KASZ,WAAA,UAAqB,WAAA,CAAY,IAAA;AAAA,cAEhC,YAAA,EAAY,MAAA,CAAA,MAAA;EAAA;;;;KAKb,YAAA,UAAsB,YAAA,CAAa,IAAA;AAAA,cAElC,qBAAA,EAAqB,MAAA,CAAA,MAAA;EAAA;;;;;;KAOtB,qBAAA,UAA+B,qBAAA,CAAsB,IAAA;AAAA,cAEpD,QAAA,EAAQ,MAAA,CAAA,MAAA;EAAA;;;;KAKT,QAAA,UAAkB,QAAA,CAAS,IAAA;AAAA,cAE1B,UAAA,EAAU,MAAA,CAAA,KAAA,WAAA,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;KACX,UAAA,UAAoB,UAAA,CAAW,IAAA;AAAA,cAE9B,aAAA,MAAa,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;;;;cACb,cAAA,MAAc,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;;cACd,uBAAA,MAAuB,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;;;;cACvB,UAAA,MAAU,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;;cAEV,UAAA,EAAU,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;KAKX,UAAA,UAAoB,UAAA,CAAW,IAAA;;;;;;AAtC3C;cA8Ca,OAAA,EAAO,MAAA,CAAA,MAAA;EAAA;;;KAIR,OAAA,UAAiB,OAAA,CAAQ,IAAA;AAAA,cAExB,YAAA,EAAY,MAAA,CAAA,KAAA,WAAA,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KACb,YAAA,UAAsB,YAAA,CAAa,IAAA;AAAA,cAElC,IAAA,EAAI,MAAA,CAAA,QAAA;AAAA,KACL,IAAA,UAAc,IAAA,CAAK,IAAA;AAAA,cAelB,OAAA,EAAO,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAMR,OAAA,UAAiB,OAAA,CAAQ,IAAA;AAAA,cAExB,QAAA,EAAQ,MAAA,CAAA,MAAA;EAAA;;;;;;KAQT,QAAA,UAAkB,QAAA,CAAS,IAAA;AAAA,cAE1B,cAAA,EAAc,MAAA,CAAA,MAAA;EAAA;;;;;KAMf,cAAA,UAAwB,cAAA,CAAe,IAAA;;;;;;;cAQtC,SAAA,EAAS,MAAA,CAAA,MAAA;EAAA;;;;;;KAOV,SAAA,UAAmB,SAAA,CAAU,IAAA;AAAA,cAE5B,WAAA,EAAW,MAAA,CAAA,KAAA,WAAA,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KACZ,WAAA,UAAqB,WAAA,CAAY,IAAA;AAAA,cAMhC,WAAA,MAAW,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;cACX,YAAA,MAAY,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;cACZ,YAAA,MAAY,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;cACZ,SAAA,MAAS,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA,GAAA,MAAA,CAAA,MAAA,CAAA,YAAA;EAAA;;;cAET,SAAA,MAAS,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;cACT,UAAA,MAAU,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;;cACV,gBAAA,MAAgB,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;cAChB,WAAA,MAAW,KAAA,EAAA,CAAA,KAAA,KAAA,IAAA,CAAA;EAAA;;;;;;cAMX,kBAAA,EAAkB,MAAA,CAAA,MAAA;EAAA;;KAGnB,kBAAA,UAA4B,kBAAA,CAAmB,IAAA;AAAA,cAE9C,mBAAA,EAAmB,MAAA,CAAA,MAAA;EAAA;;KAGpB,mBAAA,UAA6B,mBAAA,CAAoB,IAAA;AAAA,cAEhD,KAAA,EAAK,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;KAON,KAAA,UAAe,KAAA,CAAM,IAAA;AAAA,cAEpB,UAAA,EAAU,MAAA,CAAA,QAAA;AAAA,KAUX,UAAA,UAAoB,UAAA,CAAW,IAAA;AAAA,cAM9B,QAAA,GAAY,IAAA,aAAe,OAAA;AAAA,cAM3B,UAAA,GAAc,IAAA,aAAe,OAAA;AAAA,cAM7B,aAAA,GAAiB,IAAA,aAAe,OAAA;AAAA,cAMhC,cAAA,GAAkB,OAAA,UAAiB,MAAA,aAAiB,cAAA"}