Repository navigation
Expand file tree
/
Copy pathopen-responses.ts
More file actions
1507 lines (1406 loc) · 59.2 KB
/
Copy pathopen-responses.ts
File metadata and controls
1507 lines (1406 loc) · 59.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import { Effect, Option, Schema } from "effect"
import type { Content } from "@opencode/schema/tool"
import { HttpTransport } from "../route/transport/index.js"
import { Protocol } from "../route/protocol.js"
import {
AIError,
LLMEvent,
ProviderInternalError,
Usage,
type FinishReason,
type LLMRequest,
type MediaPart,
type ProviderMetadata,
type ReasoningPart,
type TextPart,
type ToolCallPart,
type ToolDefinition,
type ToolResultPart,
} from "../schema/index.js"
import type { Media } from "../media.js"
import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared.js"
import { classifyProviderFailure } from "../provider-error.js"
import { effortUpdate } from "../effort-updates.js"
import { OpenResponsesOptions } from "./utils/open-responses-options.js"
import { Lifecycle } from "./utils/lifecycle.js"
import { ToolStream } from "./utils/tool-stream.js"
const ADAPTER = "open-responses"
const NAME = "Open Responses"
export const PATH = "/responses"
// =============================================================================
// Request Body Schema
// =============================================================================
export const OpenResponsesInputText = Schema.Struct({
type: Schema.tag("input_text"),
text: Schema.String,
})
export const OpenResponsesInputImage = Schema.Struct({
type: Schema.tag("input_image"),
image_url: Schema.String,
detail: Schema.optional(Schema.String),
})
export const OpenResponsesInputFile = Schema.Struct({
type: Schema.tag("input_file"),
filename: Schema.String,
detail: Schema.optional(Schema.String),
file_data: Schema.optional(Schema.String),
file_url: Schema.optional(Schema.String),
})
const OpenResponsesInputVideo = Schema.Struct({
type: Schema.tag("input_video"),
video_url: Schema.String,
})
const MediaInput = Schema.Union([OpenResponsesInputImage, OpenResponsesInputFile])
export type MediaInput = Schema.Schema.Type<typeof MediaInput>
const OpenResponsesInputContent = Schema.Union([OpenResponsesInputText, MediaInput])
export const OpenResponsesOutputText = Schema.Struct({
type: Schema.tag("output_text"),
text: Schema.String,
})
export const MessagePhase = Schema.NullOr(Schema.Literals(["commentary", "final_answer"]))
type MessagePhase = Schema.Schema.Type<typeof MessagePhase>
export const MessageMetadata = Schema.Struct({
itemId: Schema.optional(Schema.String),
status: Schema.optional(Schema.String),
phase: Schema.optional(MessagePhase),
})
const messagePhase = (value: unknown): MessagePhase | undefined => {
if (value === null || value === "commentary" || value === "final_answer") return value
return undefined
}
const OpenResponsesReasoningSummaryText = Schema.Struct({
type: Schema.tag("summary_text"),
text: Schema.String,
})
export const OpenResponsesReasoningItem = Schema.Struct({
type: Schema.tag("reasoning"),
id: Schema.optionalKey(Schema.String),
summary: Schema.Array(OpenResponsesReasoningSummaryText),
encrypted_content: optionalNull(Schema.String),
})
const OpenResponsesWebSearchCall = Schema.StructWithRest(
Schema.Struct({
type: Schema.tag("web_search_call"),
id: Schema.String,
status: Schema.optional(Schema.String),
action: optionalNull(JsonObject),
}),
[JsonObject],
)
const OpenResponsesFileSearchCall = Schema.StructWithRest(
Schema.Struct({
type: Schema.tag("file_search_call"),
id: Schema.String,
status: Schema.optional(Schema.String),
queries: Schema.optional(Schema.Array(Schema.String)),
results: optionalNull(Schema.Array(JsonObject)),
}),
[JsonObject],
)
const OpenResponsesCodeInterpreterCall = Schema.StructWithRest(
Schema.Struct({
type: Schema.tag("code_interpreter_call"),
id: Schema.String,
status: Schema.optional(Schema.String),
code: optionalNull(Schema.String),
container_id: optionalNull(Schema.String),
outputs: optionalNull(Schema.Array(JsonObject)),
}),
[JsonObject],
)
const OpenResponsesMCPCall = Schema.StructWithRest(
Schema.Struct({
type: Schema.tag("mcp_call"),
id: Schema.String,
status: Schema.optional(Schema.String),
server_label: Schema.optional(Schema.String),
name: Schema.optional(Schema.String),
arguments: Schema.optional(Schema.String),
output: optionalNull(Schema.String),
error: Schema.optional(Schema.Unknown),
}),
[JsonObject],
)
export const HostedToolItem = Schema.Union([
OpenResponsesWebSearchCall,
OpenResponsesFileSearchCall,
OpenResponsesCodeInterpreterCall,
OpenResponsesMCPCall,
])
export type HostedToolItem = Schema.Schema.Type<typeof HostedToolItem>
// `function_call_output.output` accepts either a plain string or an ordered
// array of content items so tools can return images and files in addition to text.
// https://www.openresponses.org/reference
const OpenResponsesFunctionCallOutputContent = Schema.Union([
OpenResponsesInputText,
OpenResponsesInputImage,
OpenResponsesInputFile,
OpenResponsesInputVideo,
])
const OpenResponsesFunctionCallOutput = Schema.Union([
Schema.String,
Schema.Array(OpenResponsesFunctionCallOutputContent),
])
export const CompactionItem = Schema.Struct({
type: Schema.Literal("compaction"),
id: optionalNull(Schema.String),
encrypted_content: Schema.String,
})
// Kept out of the baseline `InputItem` union: only the OpenAI extension accepts it.
export const ConfigurationUpdate = Schema.Struct({
type: Schema.Literal("configuration_update"),
reasoning: Schema.Struct({ effort: OpenResponsesOptions.ReasoningEffort }),
})
export type ConfigurationUpdate = Schema.Schema.Type<typeof ConfigurationUpdate>
export const HostedToolReplay = Schema.StructWithRest(
Schema.Struct({
type: Schema.String,
id: Schema.String,
}),
[JsonObject],
)
export type HostedToolReplayItem = Schema.Schema.Type<typeof HostedToolReplay>
export const InputItem = Schema.Union([
CompactionItem,
ConfigurationUpdate,
Schema.Struct({ type: Schema.tag("message"), role: Schema.tag("system"), content: Schema.String }),
Schema.Struct({ type: Schema.tag("message"), role: Schema.tag("developer"), content: Schema.String }),
Schema.Struct({
type: Schema.tag("message"),
role: Schema.tag("user"),
content: Schema.Array(OpenResponsesInputContent),
id: Schema.optional(Schema.String),
status: Schema.optional(Schema.String),
}),
Schema.Struct({
type: Schema.tag("message"),
id: Schema.optionalKey(Schema.String),
role: Schema.tag("assistant"),
content: Schema.Array(OpenResponsesOutputText),
phase: Schema.optionalKey(MessagePhase),
status: Schema.optional(Schema.String),
}),
OpenResponsesReasoningItem,
Schema.Struct({
type: Schema.tag("function_call"),
id: Schema.optionalKey(Schema.String),
call_id: Schema.String,
name: Schema.String,
namespace: Schema.optional(Schema.String),
arguments: Schema.String,
}),
Schema.Struct({
type: Schema.tag("function_call_output"),
call_id: Schema.String,
output: OpenResponsesFunctionCallOutput,
}),
HostedToolItem,
HostedToolReplay,
])
type OpenResponsesInputItem = Schema.Schema.Type<typeof InputItem>
// Mutable counterpart of the schema reasoning item so `lowerMessages` can fold
// multiple streamed summary parts into the same item before flushing.
type OpenResponsesReasoningInput = {
type: "reasoning"
id?: string
summary: Array<{ type: "summary_text"; text: string }>
encrypted_content?: string | null
}
export const Tool = Schema.Struct({
type: Schema.tag("function"),
name: Schema.String,
description: Schema.String,
parameters: JsonObject,
strict: Schema.optional(Schema.Boolean),
})
export const HostedTool = Schema.StructWithRest(
Schema.Struct({
type: Schema.String,
}),
[JsonObject],
)
export type HostedTool = Schema.Schema.Type<typeof HostedTool>
export const ToolChoice = Schema.Union([
Schema.Literals(["auto", "none", "required"]),
Schema.Struct({ type: Schema.tag("function"), name: Schema.String }),
Schema.Struct({
type: Schema.tag("allowed_tools"),
mode: Schema.Literals(["auto", "none", "required"]),
tools: Schema.Array(Schema.Struct({ type: Schema.tag("function"), name: Schema.String })),
}),
])
// Fields shared between the HTTP body and the WebSocket `response.create`
// message. The HTTP body adds `stream: true`; the WebSocket message adds
// `type: "response.create"`. Defining the shared shape once keeps the two
// transports in sync without a destructure-and-strip dance.
export const coreFields = {
model: Schema.String,
input: Schema.Array(InputItem),
instructions: Schema.optional(Schema.String),
tools: optionalArray(Schema.Union([Tool, HostedTool])),
tool_choice: Schema.optional(ToolChoice),
store: Schema.optional(Schema.Boolean),
metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
safety_identifier: Schema.optional(Schema.String),
stream_options: Schema.optional(
Schema.Struct({
include_obfuscation: Schema.optional(Schema.Boolean),
}),
),
top_logprobs: Schema.optional(Schema.Int.check(Schema.isBetween({ minimum: 0, maximum: 20 }))),
truncation: Schema.optional(OpenResponsesOptions.TruncationSchema),
service_tier: Schema.optional(OpenResponsesOptions.ServiceTierSchema),
prompt_cache_key: Schema.optional(Schema.String),
include: optionalArray(OpenResponsesOptions.ResponseIncludableSchema),
reasoning: Schema.optional(
Schema.Struct({
effort: Schema.optional(OpenResponsesOptions.ReasoningEffort),
summary: Schema.optional(Schema.Literals(["auto", "concise", "detailed"])),
}),
),
text: Schema.optional(
Schema.Struct({
verbosity: Schema.optional(OpenResponsesOptions.TextVerbositySchema),
}),
),
max_output_tokens: Schema.optional(Schema.Number),
max_tool_calls: Schema.optional(Schema.Int),
parallel_tool_calls: Schema.optional(Schema.Boolean),
temperature: Schema.optional(Schema.Number),
top_p: Schema.optional(Schema.Number),
presence_penalty: Schema.optional(Schema.Number),
frequency_penalty: Schema.optional(Schema.Number),
}
export const OpenResponsesBody = Schema.Struct({
...coreFields,
stream: Schema.Literal(true),
})
export type OpenResponsesBody = Schema.Schema.Type<typeof OpenResponsesBody>
export const OpenResponsesUsage = Schema.StructWithRest(
Schema.Struct({
input_tokens: Schema.optional(Schema.Number),
input_tokens_details: optionalNull(
Schema.StructWithRest(
Schema.Struct({
cached_tokens: Schema.optional(Schema.Number),
cache_write_tokens: Schema.optional(Schema.Number),
}),
[JsonObject],
),
),
output_tokens: Schema.optional(Schema.Number),
output_tokens_details: optionalNull(
Schema.StructWithRest(Schema.Struct({ reasoning_tokens: Schema.optional(Schema.Number) }), [JsonObject]),
),
total_tokens: Schema.optional(Schema.Number),
}),
[JsonObject],
)
type OpenResponsesUsage = Schema.Schema.Type<typeof OpenResponsesUsage>
// The spec requires `id` on every output item, but some gateways drop it from
// later item events (Bedrock Mantle renames it to `item_id` on
// `output_item.done` and `response.completed.output`). Decode it as optional
// and let `normalize` recover or mint it once before the parser runs.
// https://www.openresponses.org/specification#extending-items
export const StreamItem = Schema.StructWithRest(
Schema.Struct({
type: Schema.String,
id: Schema.optional(Schema.String),
call_id: Schema.optional(Schema.String),
name: Schema.optional(Schema.String),
namespace: Schema.optional(Schema.String),
arguments: Schema.optional(Schema.String),
encrypted_content: optionalNull(Schema.String),
}),
[Schema.Record(Schema.String, Schema.Unknown)],
)
export type StreamItem = Schema.Schema.Type<typeof StreamItem>
export type OutputItem = StreamItem & { readonly id: string }
// Responses-compatible providers put error details at the top level, under `error`, or under
// `response.error`, and gateways reshape them freely: strings, numeric codes, extra fields. Those
// fields decode as opaque values and `errorDetail` reads them defensively, so an error frame can
// only fail on invalid JSON and otherwise always classifies with the raw body as the fallback.
// https://www.openresponses.org/specification
const asText = (value: unknown) =>
typeof value === "string" && value.length > 0 ? value : typeof value === "number" ? String(value) : undefined
export const Event = Schema.StructWithRest(
Schema.Struct({
type: Schema.String,
delta: Schema.optional(Schema.String),
arguments: Schema.optional(Schema.String),
text: Schema.optional(Schema.String),
item_id: Schema.optional(Schema.String),
output_index: Schema.optional(Schema.Number),
summary_index: Schema.optional(Schema.Number),
// OutputItemAdded/Done permit a null item in the Open Responses OpenAPI schema.
item: optionalNull(StreamItem),
response: Schema.optional(
Schema.StructWithRest(
Schema.Struct({
id: Schema.optional(Schema.String),
service_tier: optionalNull(Schema.String),
incomplete_details: optionalNull(Schema.Struct({ reason: Schema.optional(Schema.String) })),
output: Schema.optional(Schema.Array(StreamItem)),
usage: optionalNull(OpenResponsesUsage),
error: Schema.optional(Schema.Unknown),
}),
[Schema.Record(Schema.String, Schema.Unknown)],
),
),
code: Schema.optional(Schema.Unknown),
message: Schema.optional(Schema.Unknown),
error: Schema.optional(Schema.Unknown),
status: Schema.optional(Schema.Unknown),
status_code: Schema.optional(Schema.Unknown),
}),
[Schema.Record(Schema.String, Schema.Unknown)],
)
export type Event = Schema.Schema.Type<typeof Event>
export type NormalizedEvent = Event & { readonly item?: OutputItem | null }
const decodeEventValue = Schema.decodeUnknownEffect(Event)
const decodeFrame = Schema.decodeUnknownEffect(ProviderShared.Json)
/**
* Decodes one WebSocket frame. Some providers and gateways answer a rejected `response.create` with a bare
* `{ "error": ... }` envelope and no event type; that reads as an error event so it classifies instead of
* failing decoding.
*/
export const decodeChannelEvent = (frame: string) =>
decodeFrame(frame).pipe(
Effect.flatMap((value) =>
decodeEventValue(
ProviderShared.isRecord(value) && value.type === undefined && value.error != null
? { ...value, type: "error" }
: value,
),
),
)
export interface ProviderAdapter {
readonly id: string
readonly name: string
readonly nativeTool?: (native: NonNullable<ToolDefinition["native"]>) => Effect.Effect<HostedTool, AIError>
readonly lowerMedia?: (input: {
readonly part: MediaPart
readonly media: Media.Inline | undefined
readonly request: LLMRequest
}) => MediaInput | undefined
readonly restoreHostedToolItem?: (item: unknown) => HostedToolReplayItem | undefined
}
const BASE_ADAPTER: ProviderAdapter = { id: ADAPTER, name: NAME }
export interface ParserState {
readonly provider: LLMRequest["model"]["provider"]
readonly completedCompactions: ReadonlySet<string>
readonly id: string
readonly name: string
readonly providerMetadataKey: string
readonly tools: ToolStream.State<string>
readonly hasFunctionCall: boolean
readonly lifecycle: Lifecycle.State
readonly outputItems: Readonly<Record<number, string>>
readonly message: { readonly id: string; readonly phase: MessagePhase | null | undefined } | undefined
readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
}
type ReasoningSummaryStatus = "active" | "can-conclude" | "concluded"
interface ReasoningStreamItem {
readonly encryptedContent: string | null | undefined
// Keyed by the wire protocol's numeric `summary_index`. JS object keys coerce to
// strings, but typing the map as `Record<number, ...>` documents intent
// and matches the wire field.
readonly summaryParts: Readonly<Record<number, ReasoningSummaryStatus>>
// Summary indexes that received at least one streamed delta. The `:0` block
// is started eagerly when the item opens, so block existence cannot tell
// whether a `.done` final would duplicate streamed text.
readonly deltaIndexes: ReadonlySet<number>
}
// =============================================================================
// Request Lowering
// =============================================================================
export const lowerTool = Effect.fnUntraced(function* (protocolName: string, tool: ToolDefinition) {
if (tool.native !== undefined)
return yield* ProviderShared.invalidRequest(`${protocolName} does not support provider-native tool ${tool.name}`)
return {
type: "function" as const,
name: tool.name,
description: tool.description,
parameters: tool.inputSchema,
// The common tool definition does not currently express Responses strict-schema policy.
strict: false,
}
})
export const lowerTools = (tools: ReadonlyArray<ToolDefinition>, adapter: ProviderAdapter) =>
Effect.forEach(
tools,
(tool): Effect.Effect<Schema.Schema.Type<typeof Tool> | HostedTool, AIError> =>
tool.native !== undefined && adapter.nativeTool ? adapter.nativeTool(tool.native) : lowerTool(adapter.name, tool),
)
export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
ProviderShared.matchToolChoice(protocolName, toolChoice, {
auto: () => "auto" as const,
none: () => "none" as const,
required: () => "required" as const,
tool: (toolName) => ({ type: "function" as const, name: toolName }),
})
// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
// provider-defined and does not necessarily identify the item's semantic type.
const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
const metadata = providerMetadata?.[providerMetadataKey]
if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
const separator = metadata.itemId.indexOf("_")
return separator > 0 && separator < metadata.itemId.length - 1 ? metadata.itemId : undefined
}
const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
const id = itemID(part.providerMetadata, providerMetadataKey)
return {
type: "function_call",
...(id === undefined ? {} : { id }),
call_id: part.id,
name: part.name,
namespace: part.namespace,
arguments: ProviderShared.encodeJson(part.input === undefined ? {} : part.input),
}
}
const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
const metadata = part.providerMetadata?.[providerMetadataKey]
if (!ProviderShared.isRecord(metadata)) return undefined
const id = itemID(part.providerMetadata, providerMetadataKey)
const encryptedContent =
typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
? metadata.reasoningEncryptedContent
: undefined
if (part.text.length === 0 && !encryptedContent) return undefined
return {
type: "reasoning",
...(id === undefined ? {} : { id }),
summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
encrypted_content: encryptedContent,
}
}
const decodeImageDetail = ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenResponsesInputImage.fields.detail))
const decodeMessageMetadata = ProviderShared.validateWith(Schema.decodeUnknownEffect(MessageMetadata))
const lowerMedia = Effect.fnUntraced(function* (
part: MediaPart,
request: LLMRequest,
adapter: ProviderAdapter,
target: "message" | "tool-result",
) {
const media = part.media.inline()
const providerMedia = adapter.lowerMedia?.({ part, media, request })
if (providerMedia) return providerMedia
const rawDetail = part.providerMetadata?.[metadataKey(request.model)]?.detail
const detail = rawDetail === undefined ? undefined : yield* decodeImageDetail(rawDetail)
const mime = part.media.mediaType.toLowerCase()
const url = ProviderShared.mediaUrl(part.media)
const location = url ?? (yield* ProviderShared.requireInlineMedia(adapter.name, part.media)).dataUrl
if (part.media.kind !== "image") {
if (target === "tool-result" && part.media.kind === "video")
return { type: "input_video" as const, video_url: location }
return {
type: "input_file" as const,
filename: part.filename ?? (mime === "application/pdf" ? "document.pdf" : "file"),
detail,
...(url ? { file_url: url } : { file_data: location }),
}
}
return {
type: "input_image" as const,
image_url: location,
detail,
}
})
const lowerUserContent = Effect.fnUntraced(function* (
part: LLMRequest["messages"][number]["content"][number],
request: LLMRequest,
adapter: ProviderAdapter,
) {
if (part.type === "text") return { type: "input_text" as const, text: part.text }
if (part.type === "media") return yield* lowerMessageMedia(part, request, adapter)
return yield* ProviderShared.unsupportedContent(adapter.name, "user", ["text", "media"])
})
const lowerMessageMedia = Effect.fnUntraced(function* (part: MediaPart, request: LLMRequest, adapter: ProviderAdapter) {
const lowered = yield* lowerMedia(part, request, adapter, "message")
if (lowered.type === "input_video")
return yield* ProviderShared.invalidRequest(`${adapter.name} user messages do not support input_video`)
return lowered
})
// Tool results may carry structured text, images, and files. Keep media as provider-native
// content instead of JSON-stringifying base64 into a prompt string.
const lowerToolResultContentItem = Effect.fnUntraced(function* (
item: Content,
request: LLMRequest,
adapter: ProviderAdapter,
) {
if (item.type === "text") return { type: "input_text" as const, text: item.text }
return yield* lowerMedia(ProviderShared.toolFileMedia(item), request, adapter, "tool-result")
})
const lowerHostedToolResultContentItem = Effect.fnUntraced(function* (
item: Content,
request: LLMRequest,
adapter: ProviderAdapter,
) {
if (item.type === "text") return { type: "input_text" as const, text: item.text }
return yield* lowerMessageMedia(ProviderShared.toolFileMedia(item), request, adapter)
})
const lowerToolResultOutput = Effect.fnUntraced(function* (
part: ToolResultPart,
request: LLMRequest,
adapter: ProviderAdapter,
) {
// Text/json/error results are encoded as a plain string for backward
// compatibility with existing cassettes and provider expectations.
if (part.result.type !== "content") return ProviderShared.toolResultText(part)
// Preserve the narrowed array element type when compiled through a consumer package.
const content: ReadonlyArray<Content> = part.result.value
return yield* Effect.forEach(content, (item) => lowerToolResultContentItem(item, request, adapter))
})
const DEFAULT_EFFORT = "medium"
const lowerMessages = Effect.fnUntraced(function* (
request: LLMRequest,
adapter: ProviderAdapter,
) {
const input: OpenResponsesInputItem[] = []
const providerMetadataKey = metadataKey(request.model)
for (const message of request.messages) {
const rawMetadata = message.providerMetadata?.[providerMetadataKey]
const metadata = rawMetadata === undefined ? undefined : yield* decodeMessageMetadata(rawMetadata)
if (message.role === "system") {
const update = effortUpdate(message)
if (update) {
// Consecutive updates are rejected, so a newer one replaces its predecessor.
const last = input.at(-1)
if (last !== undefined && "type" in last && last.type === "configuration_update") input.pop()
input.push({ type: "configuration_update", reasoning: { effort: update.effort ?? DEFAULT_EFFORT } })
continue
}
input.push({
type: "message",
role: "developer",
content: ProviderShared.joinText(yield* ProviderShared.systemUpdateText(adapter.name, message)),
})
continue
}
if (message.role === "user") {
const content = yield* Effect.forEach(message.content, (part) => lowerUserContent(part, request, adapter))
if (content.length > 0)
input.push({ type: "message", role: "user", content, id: metadata?.itemId, status: metadata?.status })
continue
}
if (message.role === "assistant") {
const content: TextPart[] = []
const reasoningItems: Record<string, OpenResponsesReasoningInput> = {}
const hostedToolItems = new Set<string>()
const flushText = () => {
if (content.length === 0) return
const groups = content.reduce<
Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
>((groups, part) => {
const partMetadata = part.providerMetadata?.[providerMetadataKey]
const id = itemID(part.providerMetadata, providerMetadataKey) ?? metadata?.itemId
const partPhase = messagePhase(partMetadata?.phase)
const phase = partPhase === undefined ? metadata?.phase : partPhase
const group = groups.at(-1)
if (group && group.id === id && group.phase === phase) group.parts.push(part)
else groups.push({ id, phase, parts: [part] })
return groups
}, [])
input.push(
...groups.map((group) => ({
type: "message" as const,
...(group.id === undefined ? {} : { id: group.id }),
role: "assistant" as const,
// Replayed text is a finished input item, even if generation was cut short.
status: "completed",
content: group.parts.map((part) => ({ type: "output_text" as const, text: part.text })),
...(group.phase === undefined ? {} : { phase: group.phase }),
})),
)
content.splice(0, content.length)
}
for (const part of message.content) {
if (part.type === "compaction") {
flushText()
if (part.provider !== request.model.provider || part.encrypted === undefined)
return yield* ProviderShared.invalidRequest(
"Compaction state must be replayed to its originating provider and API",
)
input.push({ type: "compaction", id: part.id, encrypted_content: part.encrypted })
continue
}
if (part.type === "text") {
content.push(part)
continue
}
if (part.type === "reasoning") {
flushText()
const reasoning = lowerReasoning(part, providerMetadataKey)
if (!reasoning) continue
const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
if (existing) {
existing.summary.push(...reasoning.summary)
if (typeof reasoning.encrypted_content === "string")
existing.encrypted_content = reasoning.encrypted_content
continue
}
if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
input.push(reasoning)
continue
}
if (part.type === "tool-call") {
flushText()
if (part.providerExecuted === true) continue
input.push(lowerToolCall(part, providerMetadataKey))
continue
}
if (part.type === "tool-result" && part.providerExecuted === true) {
flushText()
const id = itemID(part.providerMetadata, providerMetadataKey)
const hosted =
part.result.type !== "json"
? undefined
: Schema.is(HostedToolItem)(part.result.value)
? part.result.value
: adapter.restoreHostedToolItem?.(part.result.value)
if (id !== undefined && hosted?.id === id) {
if (!hostedToolItems.has(id)) {
input.push(hosted)
hostedToolItems.add(id)
}
continue
}
const content: ReadonlyArray<Content> =
part.result.type === "content"
? part.result.value
: [{ type: "text", text: ProviderShared.toolResultText(part) }]
input.push({
type: "message",
role: "user",
content: yield* Effect.forEach(content, (item) => lowerHostedToolResultContentItem(item, request, adapter)),
})
continue
}
if (part.type === "media") {
flushText()
// Responses has no assistant-authored image item; replay generated media (e.g. from Gemini) as user input.
input.push({
type: "message",
role: "user",
content: [yield* lowerMessageMedia(part, request, adapter)],
})
continue
}
return yield* ProviderShared.unsupportedContent(adapter.name, "assistant", [
"text",
"reasoning",
"tool-call",
"tool-result",
"media",
])
}
flushText()
continue
}
for (const part of message.content) {
if (!ProviderShared.supportsContent(part, ["tool-result"]))
return yield* ProviderShared.unsupportedContent(adapter.name, "tool", ["tool-result"])
input.push({
type: "function_call_output",
call_id: part.id,
output: yield* lowerToolResultOutput(part, request, adapter),
})
}
}
return input
})
export const lowerConversation = Effect.fnUntraced(function* (
request: LLMRequest,
adapter: ProviderAdapter,
) {
const instructions = ProviderShared.joinText(request.system)
return {
model: request.model.id,
input: yield* lowerMessages(request, adapter),
...(instructions ? { instructions } : {}),
}
})
export const lowerGeneration = (request: LLMRequest, options = OpenResponsesOptions.resolve(request)) => {
const generation = request.generation
const cacheKey = ProviderShared.promptCacheKey(request)
const parallelToolCalls = resolveParallelToolCalls(request)
return {
stream: true as const,
max_output_tokens: generation?.maxTokens,
temperature: generation?.temperature,
top_p: generation?.topP,
presence_penalty: generation?.presencePenalty,
frequency_penalty: generation?.frequencyPenalty,
...(options.store !== undefined ? { store: options.store } : {}),
...(options.metadata ? { metadata: options.metadata } : {}),
...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
...(options.streamOptions?.includeObfuscation !== undefined
? { stream_options: { include_obfuscation: options.streamOptions.includeObfuscation } }
: {}),
...(options.topLogprobs !== undefined ? { top_logprobs: options.topLogprobs } : {}),
...(cacheKey ? { prompt_cache_key: cacheKey } : {}),
...(options.include ? { include: options.include } : {}),
...(options.reasoningEffort || options.reasoningSummary
? { reasoning: { effort: options.reasoningEffort, summary: options.reasoningSummary } }
: {}),
...(options.textVerbosity ? { text: { verbosity: options.textVerbosity } } : {}),
...(options.serviceTier ? { service_tier: options.serviceTier } : {}),
...(options.maxToolCalls !== undefined ? { max_tool_calls: options.maxToolCalls } : {}),
...(parallelToolCalls !== undefined ? { parallel_tool_calls: parallelToolCalls } : {}),
...(options.truncation ? { truncation: options.truncation } : {}),
}
}
export const resolveParallelToolCalls = (request: LLMRequest) => {
const configured = OpenResponsesOptions.resolve(request).parallelToolCalls
if (configured !== undefined) return configured
const disabled = request.toolChoice?.disableParallelToolUse
return disabled === undefined ? undefined : !disabled
}
export const allowedToolChoice = (request: LLMRequest) => {
const allowed = OpenResponsesOptions.resolve(request).allowedTools
if (!allowed) return undefined
return {
type: "allowed_tools" as const,
mode: allowed.mode,
tools: allowed.toolNames.map((name) => ({ type: "function" as const, name })),
}
}
export const fromRequestWithAdapter = Effect.fn("OpenResponses.fromRequestWithAdapter")(function* (
request: LLMRequest,
adapter: ProviderAdapter,
) {
const projected = ProviderShared.flattenToolRequest(request)
return {
...(yield* lowerConversation(projected.request, adapter)),
...lowerGeneration(request),
tools: projected.tools.length === 0 ? undefined : yield* lowerTools(projected.tools, adapter),
tool_choice:
allowedToolChoice(request) ??
(request.toolChoice ? yield* lowerToolChoice(adapter.name, request.toolChoice) : undefined),
}
})
export const fromRequest = (request: LLMRequest) => fromRequestWithAdapter(request, BASE_ADAPTER)
// =============================================================================
// Stream Parsing
// =============================================================================
// Responses APIs report `input_tokens` (inclusive total) with a
// cached-read and cache-write subsets, and `output_tokens` (inclusive total)
// with a `reasoning_tokens` subset. Pass the totals through and derive the
// non-cached breakdown.
export const mapUsage = (usage: OpenResponsesUsage | null | undefined, providerMetadataKey: string) => {
if (!usage) return undefined
const cached = usage.input_tokens_details?.cached_tokens
const cacheWrite = usage.input_tokens_details?.cache_write_tokens
const reasoning = usage.output_tokens_details?.reasoning_tokens
const nonCached = ProviderShared.subtractTokens(usage.input_tokens, ProviderShared.sumTokens(cached, cacheWrite))
return new Usage({
inputTokens: usage.input_tokens,
outputTokens: usage.output_tokens,
nonCachedInputTokens: nonCached,
cacheReadInputTokens: cached,
cacheWriteInputTokens: cacheWrite,
reasoningTokens: reasoning,
totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, usage.total_tokens),
providerMetadata: { [providerMetadataKey]: usage },
})
}
const mapFinishReason = (event: Event, hasFunctionCall: boolean): FinishReason => {
const reason = event.response?.incomplete_details?.reason
if (reason === undefined || reason === null) {
if (hasFunctionCall) return "tool-calls"
if (event.type === "response.incomplete") return "unknown"
return "stop"
}
if (reason === "max_output_tokens") return "length"
if (reason === "content_filter") return "content-filter"
return hasFunctionCall ? "tool-calls" : "unknown"
}
export const metadataKey = (model: LLMRequest["model"]) => model.route.providerMetadataKey ?? "openresponses"
export const providerMetadata = (state: ParserState, metadata: Record<string, unknown>): ProviderMetadata => ({
[state.providerMetadataKey]: metadata,
})
export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
const NO_EVENTS: StepResult["1"] = []
// `response.completed` / `response.incomplete` are clean finishes that emit a
// `finish` event; `response.failed` and `error` are hard failures. All four end
// the stream, so keep this set aligned with `step` and the protocol's terminal predicate.
const TERMINAL_TYPES = new Set(["error", "response.completed", "response.incomplete", "response.failed"])
export const terminal = (event: Event) => TERMINAL_TYPES.has(event.type)
const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepResult => {
if (!event.delta || state.message?.id !== id) return [state, NO_EVENTS]
const events: LLMEvent[] = []
const phase = state.message.phase
const metadata = providerMetadata(state, { itemId: id, ...(phase === undefined ? {} : { phase }) })
const lifecycle = Lifecycle.textStart(state.lifecycle, events, id, metadata)
return [{ ...state, lifecycle: Lifecycle.textDelta(lifecycle, events, id, event.delta) }, events]
}
const onOutputTextDone = (state: ParserState, event: Event, id: string): StepResult => {
if (state.message?.id === id) {
if (state.lifecycle.text.has(id) || event.text === undefined) return [state, NO_EVENTS]
return onOutputTextDelta(state, { ...event, delta: event.text }, id)
}
const events: LLMEvent[] = []
return [{ ...state, lifecycle: Lifecycle.textEnd(state.lifecycle, events, id) }, events]
}
const decodeMessagePart = Schema.decodeUnknownOption(
Schema.Union([OpenResponsesOutputText, Schema.Struct({ type: Schema.tag("refusal"), refusal: Schema.String })]),
)
const decodeSummaryPart = Schema.decodeUnknownOption(OpenResponsesReasoningSummaryText)
const decodeReasoningPart = Schema.decodeUnknownOption(
Schema.Struct({ type: Schema.tag("reasoning_text"), text: Schema.String }),
)
const joinReasoningText = (parts: ReadonlyArray<string | undefined>) => {
if (!parts.some((part) => part !== undefined && part.length > 0)) return undefined
return parts.filter((part) => part !== undefined).join("\n\n")
}
const outputItemID = (state: Pick<ParserState, "outputItems">, event: Event) =>
event.output_index === undefined ? event.item_id : (state.outputItems[event.output_index] ?? event.item_id)
const ITEM_ID_PREFIX: Readonly<Record<string, string>> = {
message: "msg",
reasoning: "rs",
function_call: "fc",
compaction: "cmp",
}
// An item without an id adopts the id already open in its output slot,
// otherwise it gets a locally minted one.
const resolveItem = (
state: Pick<ParserState, "outputItems">,
item: StreamItem,
index: number | undefined,
): OutputItem => ({
...item,
id:
item.id ??
(index === undefined ? undefined : state.outputItems[index]) ??
`${ITEM_ID_PREFIX[item.type] ?? "item"}_${crypto.randomUUID().replaceAll("-", "")}`,
})
// Registered output slots are authoritative for `item_id` routing, and items
// are resolved here so everything downstream can rely on `item.id`.
export const normalize = (state: Pick<ParserState, "outputItems">, input: Event): NormalizedEvent => ({
...input,
item_id: input.item_id === undefined ? undefined : outputItemID(state, input),
item: input.item ? resolveItem(state, input.item, input.output_index) : input.item,
})
const startReasoningSummaryPart = (state: ParserState, itemID: string, index: number): StepResult => {
const item = state.reasoningItems[itemID]
if (!item || index === 0 || item.summaryParts[index] !== undefined) return [state, NO_EVENTS]
const events: LLMEvent[] = []
const lifecycle = Object.entries(item.summaryParts)
.filter((entry) => entry[1] !== "concluded")
.reduce(
(lifecycle, entry) =>
Lifecycle.reasoningEnd(lifecycle, events, `${itemID}:${entry[0]}`, providerMetadata(state, { itemId: itemID })),
state.lifecycle,
)
return [
{
...state,
lifecycle: Lifecycle.reasoningStart(
lifecycle,
events,
`${itemID}:${index}`,
providerMetadata(state, { itemId: itemID, reasoningEncryptedContent: item.encryptedContent ?? null }),
),
reasoningItems: {
...state.reasoningItems,
[itemID]: {
...item,
summaryParts: {
...Object.fromEntries(
Object.entries(item.summaryParts).map((entry) =>
entry[1] === "concluded" ? entry : [entry[0], "concluded" as const],
),
),
[index]: "active",
},
},
},
},
events,
]