@openclaw/ai 2026.9.2 → 2026.9.4

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 (115) hide show
  1. package/README.md +3 -2
  2. package/dist/anthropic-C4Qu4H0Z.mjs +377 -0
  3. package/dist/{anthropic-payload-policy-d46X2pR0.d.mts → anthropic-payload-policy-BZ8umAbk.d.mts} +16 -6
  4. package/dist/{anthropic-compaction-replay-DB2FGLQc.mjs → anthropic-stream-reducer-CILWF7JD.mjs} +931 -184
  5. package/dist/{api-registry-DWtPjzyn.d.mts → api-registry-ByUwIR0e.d.mts} +2 -2
  6. package/dist/assistant-output-tLt4H-iQ.mjs +16 -0
  7. package/dist/assistant-text-phase-C20rxWwP.mjs +55 -0
  8. package/dist/{azure-openai-responses-C5bZRfAP.mjs → azure-openai-responses-BAlqlKKc.mjs} +5 -5
  9. package/dist/{base64-CEFBpSkN.mjs → base64-D-su8YVo.mjs} +1 -27
  10. package/dist/{diagnostics-DfFyKeX_.mjs → diagnostics-QuErwCIl.mjs} +96 -21
  11. package/dist/diagnostics.d.mts +3 -1
  12. package/dist/diagnostics.mjs +3 -3
  13. package/dist/{event-stream-BhT5T1Ay.d.mts → event-stream-Bt5Y4Pav.d.mts} +11 -3
  14. package/dist/event-stream-C3WGFsum.d.mts +1 -0
  15. package/dist/{event-stream-BgDvQeum.mjs → event-stream-D8PARQfL.mjs} +45 -1
  16. package/dist/event-stream.d.mts +2 -2
  17. package/dist/event-stream.mjs +1 -1
  18. package/dist/{github-copilot-headers-NCJtz9i0.mjs → github-copilot-headers-B37TB3rB.mjs} +3 -10
  19. package/dist/github-copilot-request-facts-BTEBMeOv.mjs +16 -0
  20. package/dist/{google-4qeuE8iX.mjs → google-DPBAOaOW.mjs} +10 -10
  21. package/dist/google-messages-6JkpHrhJ.mjs +448 -0
  22. package/dist/google-shared-BvBeW9aq.mjs +185 -0
  23. package/dist/{google-vertex-O9BcnB6X.mjs → google-vertex-k-TMCAYD.mjs} +7 -6
  24. package/dist/{host-DjzGmdZ2.d.mts → host-4atIX-2V.d.mts} +5 -3
  25. package/dist/{host-CEvLw30U.mjs → host-B8YfDGd4.mjs} +181 -41
  26. package/dist/host-policy-Zcg_cNz8.mjs +37 -0
  27. package/dist/{index-FnHM2FcI.d.mts → index-DdD3qerf.d.mts} +4 -4
  28. package/dist/index.d.mts +7 -7
  29. package/dist/index.mjs +7 -6
  30. package/dist/internal/anthropic.d.mts +7 -7
  31. package/dist/internal/anthropic.mjs +4 -4
  32. package/dist/internal/google-model-family.d.mts +5 -0
  33. package/dist/internal/google-model-family.mjs +15 -0
  34. package/dist/internal/openai-responses-payload-policy.d.mts +3 -3
  35. package/dist/internal/openai-responses-payload-policy.mjs +3 -2
  36. package/dist/internal/openai.d.mts +8 -103
  37. package/dist/internal/openai.mjs +9 -10
  38. package/dist/internal/retry-after.d.mts +2 -4
  39. package/dist/internal/retry-after.mjs +57 -8
  40. package/dist/internal/runtime.d.mts +43 -7
  41. package/dist/internal/runtime.mjs +6 -7
  42. package/dist/internal/shared.d.mts +43 -3
  43. package/dist/internal/shared.mjs +5 -4
  44. package/dist/internal/tool-schema.d.mts +63 -0
  45. package/dist/internal/tool-schema.mjs +3 -0
  46. package/dist/{mistral-Tb6oalqH.mjs → mistral-CxUZ1jUb.mjs} +10 -37
  47. package/dist/model-transport-url-DKocrEsb.d.mts +28 -0
  48. package/dist/{openai-chatgpt-responses-yUXjPNfu.mjs → openai-chatgpt-responses-CgO6kZfo.mjs} +50 -41
  49. package/dist/{openai-completions-compat-CpjYUArk.d.mts → openai-completions-compat-eHgh5UPE.d.mts} +26 -4
  50. package/dist/{openai-completions-stream-BsrBe3Gg.mjs → openai-completions-stream-Da2vvl-S.mjs} +654 -453
  51. package/dist/openai-completions-yJuk7eis.mjs +187 -0
  52. package/dist/openai-prompt-cache-B4eYo2-I.d.mts +16 -0
  53. package/dist/openai-prompt-cache-Bds-n_9Q.mjs +220 -0
  54. package/dist/openai-provider-client-S2gCrM2Z.mjs +24 -0
  55. package/dist/{openai-responses-CCL41ALM.mjs → openai-responses-DaYwH05E.mjs} +14 -29
  56. package/dist/{openai-responses-compaction-window-BMVHFOtq.mjs → openai-responses-compaction-window-CIhBAkkq.mjs} +6 -246
  57. package/dist/{openai-responses-contracts-rF5DDRNS.d.mts → openai-responses-contracts-BjBAqAg_.d.mts} +5 -38
  58. package/dist/openai-responses-contracts-DDOHA62Y.mjs +245 -0
  59. package/dist/{openai-responses-prompt-observer-internal-D0bhBfgL.mjs → openai-responses-prompt-observer-internal-f8J7wpsk.mjs} +3 -3
  60. package/dist/{openai-responses-shared-Bmma_3Qc.mjs → openai-responses-shared-B8RdBPCv.mjs} +47 -118
  61. package/dist/openai-tool-projection-793cE3QX.d.mts +37 -0
  62. package/dist/{openai-tool-schema-_pTAJqKF.mjs → openai-tool-schema-CzjyYXun.mjs} +38 -556
  63. package/dist/openai-tool-schema-ynZBgqW-.d.mts +38 -0
  64. package/dist/openai-transport-params-9aPuV5YY.mjs +675 -0
  65. package/dist/positive-integer-41zhOdcV.mjs +8 -0
  66. package/dist/{provider-error-9TraxGvt.mjs → provider-error-BA-v_tKd.mjs} +70 -11
  67. package/dist/{provider-options-Bcc_p-TU.d.mts → provider-options-Ceqv1OKk.d.mts} +4 -39
  68. package/dist/{provider-replay-context-BuSUaAk5.mjs → provider-replay-context-CJ_YvcEW.mjs} +1 -1
  69. package/dist/{provider-transcript-transform-BaMbI1hr.mjs → provider-transcript-transform-V5YzU9zh.mjs} +1 -1
  70. package/dist/provider-transport-turn-state-D5EXOFL2.mjs +42 -0
  71. package/dist/provider-types-CVjKjsuq.d.mts +29 -0
  72. package/dist/provider-types.d.mts +6 -31
  73. package/dist/providers.d.mts +2 -2
  74. package/dist/providers.mjs +11 -11
  75. package/dist/{reasoning-tag-text-partitioner-DLCNXki6.mjs → reasoning-tag-text-partitioner-BcR5pztD.mjs} +19 -19
  76. package/dist/record-coerce-DwRYMj3t.mjs +32 -0
  77. package/dist/retry-after-CdCURCVg.d.mts +15 -0
  78. package/dist/session-affinity-Bcunsn4I.mjs +23 -0
  79. package/dist/{simple-options-BjHCCh4v.mjs → simple-options-BQbb4yQL.mjs} +24 -5
  80. package/dist/src-DDmEryvj.mjs +2 -0
  81. package/dist/{stream-first-event-timeout-DcNjoFQE.mjs → stream-first-event-timeout-2hfrquuz.mjs} +1 -1
  82. package/dist/string-coerce-fsri9iCu.mjs +34 -0
  83. package/dist/{string-normalization--fwJ4S2q.mjs → string-normalization-CmLIasuf.mjs} +1 -1
  84. package/dist/{tool-schema-json-projection-mJhXDcyz.mjs → tool-schema-json-projection-ClptDdAO.mjs} +2 -38
  85. package/dist/{transport-stream-shared-xnaqxmbP.d.mts → transport-stream-shared-DNvmoWnv.d.mts} +20 -36
  86. package/dist/{transport-stream-shared-CZqMhfIw.mjs → transport-stream-shared-zHll9BxO.mjs} +225 -13
  87. package/dist/transport-utils-zrYjICLZ.mjs +118 -0
  88. package/dist/transports.d.mts +137 -193
  89. package/dist/transports.mjs +204 -1334
  90. package/dist/types-DlfwzH3T.d.mts +1 -0
  91. package/dist/types-LFWwv0cF.mjs +8 -0
  92. package/dist/{types-CJ1-Ht7A.d.mts → types-Ntv5z2g2.d.mts} +74 -57
  93. package/dist/types.d.mts +6 -6
  94. package/dist/types.mjs +6 -5
  95. package/dist/{src-2qBGKg8O.mjs → usage-cost-BNWbbXav.mjs} +0 -1
  96. package/dist/utf16-slice-CvGodqok.mjs +29 -0
  97. package/dist/{validation-AKZBDGQd.d.mts → validation-B0t_G2H6.d.mts} +1 -1
  98. package/dist/{validation-CJZtym2g.mjs → validation-BDzVDnTs.mjs} +9 -1
  99. package/dist/validation.d.mts +1 -1
  100. package/dist/validation.mjs +1 -1
  101. package/package.json +13 -3
  102. package/dist/anthropic-BDdqdVLK.mjs +0 -886
  103. package/dist/event-stream-zctLx0yr.d.mts +0 -1
  104. package/dist/google-shared-CWeG8RIl.mjs +0 -636
  105. package/dist/google-thinking-level-C-V3tecN.mjs +0 -9
  106. package/dist/hash-CHgqbJmD.mjs +0 -16
  107. package/dist/json-parse-BuAJEbdW.mjs +0 -146
  108. package/dist/openai-completions-BIUV3RDT.mjs +0 -403
  109. package/dist/openai-prompt-cache-CGnVB74a.mjs +0 -21
  110. package/dist/openai-prompt-cache-CNoIfHYC.d.mts +0 -16
  111. package/dist/sanitize-unicode-D6xUvZaS.mjs +0 -83
  112. package/dist/transport-utils-7il795_9.mjs +0 -138
  113. package/dist/types-3Lnm-QSJ.d.mts +0 -1
  114. package/dist/types-BDdaOVi2.mjs +0 -6
  115. package/dist/utf16-slice-qz3nsy87.mjs +0 -84
@@ -1,146 +0,0 @@
1
- import { n as asNonArrayRecord } from "./utf16-slice-qz3nsy87.mjs";
2
- import { parse } from "partial-json";
3
- //#region packages/ai/src/utils/json-parse.ts
4
- const VALID_JSON_ESCAPES = /* @__PURE__ */ new Set([
5
- "\"",
6
- "\\",
7
- "/",
8
- "b",
9
- "f",
10
- "n",
11
- "r",
12
- "t",
13
- "u"
14
- ]);
15
- const JSON_CONTROL_ESCAPES = /* @__PURE__ */ new Set([
16
- "b",
17
- "f",
18
- "n",
19
- "r",
20
- "t"
21
- ]);
22
- function isControlCharacter(char) {
23
- const codePoint = char.codePointAt(0);
24
- return codePoint !== void 0 && codePoint >= 0 && codePoint <= 31;
25
- }
26
- function escapeControlCharacter(char) {
27
- switch (char) {
28
- case "\b": return "\\b";
29
- case "\f": return "\\f";
30
- case "\n": return "\\n";
31
- case "\r": return "\\r";
32
- case " ": return "\\t";
33
- default: return `\\u${char.codePointAt(0)?.toString(16).padStart(4, "0") ?? "0000"}`;
34
- }
35
- }
36
- /**
37
- * Repairs malformed JSON string literals by:
38
- * - escaping raw control characters inside strings
39
- * - doubling backslashes before invalid escape characters
40
- */
41
- function repairJson(json) {
42
- let repaired = "";
43
- let inString = false;
44
- let stringValuePrefix = "";
45
- for (let index = 0; index < json.length; index++) {
46
- const char = json.charAt(index);
47
- if (!inString) {
48
- repaired += char;
49
- if (char === "\"") {
50
- inString = true;
51
- stringValuePrefix = "";
52
- }
53
- continue;
54
- }
55
- if (char === "\"") {
56
- repaired += char;
57
- inString = false;
58
- stringValuePrefix = "";
59
- continue;
60
- }
61
- if (char === "\\") {
62
- const nextChar = json.charAt(index + 1);
63
- if (!nextChar) {
64
- repaired += "\\\\";
65
- continue;
66
- }
67
- if (nextChar === "u") {
68
- const unicodeDigits = json.slice(index + 2, index + 6);
69
- if (/^[0-9a-fA-F]{4}$/.test(unicodeDigits)) {
70
- repaired += `\\u${unicodeDigits}`;
71
- stringValuePrefix += `\\u${unicodeDigits}`;
72
- index += 5;
73
- continue;
74
- }
75
- repaired += "\\\\";
76
- stringValuePrefix += "\\";
77
- continue;
78
- }
79
- if (JSON_CONTROL_ESCAPES.has(nextChar) && looksLikeWindowsPathPrefix(stringValuePrefix)) {
80
- repaired += "\\\\";
81
- stringValuePrefix += "\\";
82
- continue;
83
- }
84
- if (VALID_JSON_ESCAPES.has(nextChar)) {
85
- repaired += `\\${nextChar}`;
86
- stringValuePrefix += nextChar === "\\" ? "\\" : `\\${nextChar}`;
87
- index += 1;
88
- continue;
89
- }
90
- repaired += "\\\\";
91
- stringValuePrefix += "\\";
92
- continue;
93
- }
94
- repaired += isControlCharacter(char) ? escapeControlCharacter(char) : char;
95
- stringValuePrefix += char;
96
- }
97
- return repaired;
98
- }
99
- function parseJsonWithRepair(json) {
100
- return JSON.parse(repairJson(json));
101
- }
102
- function looksLikeWindowsPathPrefix(prefix) {
103
- const tail = prefix.slice(-160);
104
- return /(?:^|[^A-Za-z0-9])[A-Za-z]:(?:[\\/][^"\\/:*?<>|\r\n]*)*$/.test(tail);
105
- }
106
- /**
107
- * Attempts to parse potentially incomplete JSON during streaming.
108
- * Always returns a valid object, even if the JSON is incomplete.
109
- *
110
- * @param partialJson The partial JSON string from streaming
111
- * @returns Parsed object or empty object if parsing fails
112
- */
113
- function parseStreamingJson(partialJson) {
114
- if (!partialJson || partialJson.trim() === "") return {};
115
- try {
116
- return asNonArrayRecord(parseJsonWithRepair(partialJson));
117
- } catch {
118
- try {
119
- return asNonArrayRecord(parse(partialJson));
120
- } catch {
121
- try {
122
- return asNonArrayRecord(parse(repairJson(partialJson)));
123
- } catch {
124
- return {};
125
- }
126
- }
127
- }
128
- }
129
- const TOOL_ARGUMENT_PREVIEW_FIRST_CHECKPOINT_CHARS = 512;
130
- /**
131
- * Streamed tool-call arguments are preview-only; the terminal parse re-reads
132
- * the full buffer authoritatively at content_block_stop. Reparsing every delta
133
- * scans an ever-growing buffer and makes assembly quadratic in the argument
134
- * size, so refresh previews on a geometric length schedule instead — bounded
135
- * staleness, linear total parse work.
136
- */
137
- function createToolArgumentPreviewSchedule() {
138
- let nextCheckpointChars = TOOL_ARGUMENT_PREVIEW_FIRST_CHECKPOINT_CHARS;
139
- return (accumulatedChars) => {
140
- if (accumulatedChars < nextCheckpointChars) return false;
141
- nextCheckpointChars = accumulatedChars * 2;
142
- return true;
143
- };
144
- }
145
- //#endregion
146
- export { repairJson as i, parseJsonWithRepair as n, parseStreamingJson as r, createToolArgumentPreviewSchedule as t };
@@ -1,403 +0,0 @@
1
- import { t as __exportAll } from "./rolldown-runtime-BhDjJH2R.mjs";
2
- import { t as AssistantMessageEventStream } from "./event-stream-BgDvQeum.mjs";
3
- import { i as clampThinkingLevel } from "./sanitize-unicode-D6xUvZaS.mjs";
4
- import { n as getAiTransportHost } from "./host-CEvLw30U.mjs";
5
- import { n as clampOpenAIPromptCacheKey } from "./openai-prompt-cache-CGnVB74a.mjs";
6
- import { t as resolveCacheRetention } from "./cache-retention-0x979a5V.mjs";
7
- import { m as sortPromptCacheToolsByName, n as buildBaseOptions, o as reasoningTagTextPolicy, u as splitSystemPromptCacheBoundary } from "./simple-options-BjHCCh4v.mjs";
8
- import { n as getEnvApiKey } from "./env-api-keys-bktO00EJ.mjs";
9
- import { _ as transportAbortError, t as assignTransportErrorDetails, y as withProviderResponseHook } from "./transport-stream-shared-CZqMhfIw.mjs";
10
- import { E as tagUnresolvedTextAsCommentary, S as shouldOmitOllamaCompatResponseFormat, _ as hasToolCallHistory, b as resolveOpenAICompletionsCompat, g as convertMessages, h as resolveOpenAIReasoningEffortMap, t as processCompletionsStream, v as finalizeOpenAICompletionsToolCalls, w as clearPendingCommentaryText, x as resolveOpenAICompletionsResponseFormat } from "./openai-completions-stream-BsrBe3Gg.mjs";
11
- import { C as createOpenAIProviderAcceptanceHook, T as isOpenAICompletionsThinkingEnabled, j as resolveOpenAIClientBaseUrl, v as projectOpenAITools, y as reconcileOpenAICompletionsToolChoice } from "./openai-tool-schema-_pTAJqKF.mjs";
12
- import { i as getFirstStreamEventTimeoutMs, r as getFirstStreamEventTimeoutHandler, t as createFirstStreamEventAbortController } from "./stream-first-event-timeout-DcNjoFQE.mjs";
13
- import { i as resolveCloudflareBaseUrl, n as hasCopilotVisionInput, r as isCloudflareProvider, t as buildCopilotDynamicHeaders } from "./github-copilot-headers-NCJtz9i0.mjs";
14
- import OpenAI from "openai";
15
- //#region packages/ai/src/providers/openai-completions.ts
16
- var openai_completions_exports = /* @__PURE__ */ __exportAll({
17
- convertMessages: () => convertMessages,
18
- streamOpenAICompletions: () => streamOpenAICompletions,
19
- streamSimpleOpenAICompletions: () => streamSimpleOpenAICompletions
20
- });
21
- const streamOpenAICompletions = (model, context, options) => {
22
- const stream = new AssistantMessageEventStream();
23
- (async () => {
24
- const output = {
25
- role: "assistant",
26
- content: [],
27
- api: model.api,
28
- provider: model.provider,
29
- model: model.id,
30
- usage: {
31
- input: 0,
32
- output: 0,
33
- cacheRead: 0,
34
- cacheWrite: 0,
35
- totalTokens: 0,
36
- cost: {
37
- input: 0,
38
- output: 0,
39
- cacheRead: 0,
40
- cacheWrite: 0,
41
- total: 0
42
- }
43
- },
44
- stopReason: "stop",
45
- timestamp: Date.now()
46
- };
47
- const provisionalCommentaryTags = /* @__PURE__ */ new Map();
48
- let firstEventAbort;
49
- try {
50
- const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
51
- const compat = resolveOpenAICompletionsCompat(model);
52
- const shouldEmitReasoning = Boolean(model.reasoning && options?.reasoningEffort && isOpenAICompletionsThinkingEnabled(options.reasoningEffort));
53
- const cacheRetention = resolveCacheRetention(options?.cacheRetention);
54
- const cacheSessionId = cacheRetention === "none" ? void 0 : options?.sessionId;
55
- const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat);
56
- let params = buildParams(model, context, options, compat, cacheRetention);
57
- const nextParams = await options?.onPayload?.(params, model);
58
- if (nextParams !== void 0) params = nextParams;
59
- firstEventAbort = createFirstStreamEventAbortController(options?.signal);
60
- const requestOptions = {
61
- signal: firstEventAbort.signal,
62
- ...options?.timeoutMs !== void 0 ? { timeout: options.timeoutMs } : {},
63
- maxRetries: 0
64
- };
65
- const { data: openaiStream, response } = await client.chat.completions.create(params, requestOptions).withResponse();
66
- const hookedOpenAIStream = withProviderResponseHook({
67
- stream: openaiStream,
68
- signal: firstEventAbort.signal,
69
- abort: firstEventAbort.abort,
70
- hook: createOpenAIProviderAcceptanceHook(options, response, model),
71
- onReady: () => stream.push({
72
- type: "start",
73
- partial: output
74
- })
75
- });
76
- const finishedBlocks = /* @__PURE__ */ new Set();
77
- const contentIndices = /* @__PURE__ */ new WeakMap();
78
- let openTextBlock;
79
- let openThinkingBlock;
80
- const finishBlock = (block) => {
81
- const contentIndex = contentIndices.get(block);
82
- if (contentIndex === void 0 || finishedBlocks.has(block)) return;
83
- finishedBlocks.add(block);
84
- if (block.type === "text") {
85
- openTextBlock = void 0;
86
- stream.push({
87
- type: "text_end",
88
- contentIndex,
89
- content: block.text,
90
- partial: output
91
- });
92
- } else if (block.type === "thinking") {
93
- openThinkingBlock = void 0;
94
- stream.push({
95
- type: "thinking_end",
96
- contentIndex,
97
- content: block.thinking,
98
- partial: output
99
- });
100
- } else if (block.type === "toolCall") stream.push({
101
- type: "toolcall_end",
102
- contentIndex,
103
- toolCall: block,
104
- partial: output
105
- });
106
- };
107
- const directEventStream = { push(event) {
108
- if (event.type === "text_start" || event.type === "thinking_start" || event.type === "toolcall_start") {
109
- const block = output.content[event.contentIndex];
110
- if (block) {
111
- contentIndices.set(block, event.contentIndex);
112
- if (block.type === "text") openTextBlock = block;
113
- else if (block.type === "thinking") openThinkingBlock = block;
114
- }
115
- }
116
- stream.push(event);
117
- } };
118
- try {
119
- await processCompletionsStream(hookedOpenAIStream, output, model, directEventStream, {
120
- mode: "direct",
121
- beforeContentBlock(nextType) {
122
- if (openThinkingBlock) finishBlock(openThinkingBlock);
123
- if (openTextBlock && nextType !== "toolCall") finishBlock(openTextBlock);
124
- },
125
- provisionalCommentaryTags,
126
- signal: options?.signal,
127
- emitReasoning: shouldEmitReasoning,
128
- strictReasoningTags: reasoningTagTextPolicy.isStrict(options),
129
- firstEventTimeoutMs: getFirstStreamEventTimeoutMs(options),
130
- abortFirstEventStream: firstEventAbort.abort,
131
- onFirstEventTimeout: getFirstStreamEventTimeoutHandler(options)
132
- });
133
- if (options?.signal?.aborted) throw transportAbortError(options.signal);
134
- if (output.stopReason === "aborted" || output.stopReason === "error") throw new Error(output.errorMessage || (output.stopReason === "aborted" ? "Request was aborted" : "Provider returned an invalid tool call"));
135
- } catch (error) {
136
- for (const block of output.content) if (block.type !== "toolCall") finishBlock(block);
137
- throw error;
138
- }
139
- for (const block of output.content) if (block.type !== "toolCall" || output.stopReason === "toolUse") finishBlock(block);
140
- stream.push({
141
- type: "done",
142
- reason: output.stopReason,
143
- message: output
144
- });
145
- stream.end();
146
- } catch (error) {
147
- const terminal = assignTransportErrorDetails(output, error, options?.signal);
148
- finalizeOpenAICompletionsToolCalls(output, { allowSilentToolCallPromotion: false });
149
- clearPendingCommentaryText(provisionalCommentaryTags);
150
- tagUnresolvedTextAsCommentary(output);
151
- for (const block of output.content) {
152
- delete block.index;
153
- delete block.partialArgs;
154
- delete block.streamIndex;
155
- }
156
- stream.push({
157
- type: "error",
158
- reason: terminal.stopReason,
159
- error: output
160
- });
161
- stream.end();
162
- } finally {
163
- firstEventAbort?.dispose();
164
- }
165
- })();
166
- return stream;
167
- };
168
- const streamSimpleOpenAICompletions = (model, context, options) => {
169
- const apiKey = options?.apiKey || getEnvApiKey(model.provider);
170
- if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
171
- const base = buildBaseOptions(model, options, apiKey);
172
- const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : void 0;
173
- const reasoningEffort = clampedReasoning === "off" ? void 0 : clampedReasoning === "max" ? "xhigh" : clampedReasoning;
174
- const toolChoice = options?.toolChoice;
175
- return streamOpenAICompletions(model, context, {
176
- ...base,
177
- reasoningEffort,
178
- toolChoice
179
- });
180
- };
181
- function createClient(model, context, apiKey, optionsHeaders, sessionId, compat = resolveOpenAICompletionsCompat(model)) {
182
- if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
183
- const headers = { ...model.headers };
184
- if (model.provider === "github-copilot") {
185
- const hasImages = hasCopilotVisionInput(context.messages);
186
- const copilotHeaders = buildCopilotDynamicHeaders({
187
- messages: context.messages,
188
- hasImages
189
- });
190
- Object.assign(headers, copilotHeaders);
191
- }
192
- if (sessionId && compat.sessionAffinity !== "none") {
193
- if (compat.sessionAffinity === "openrouter") headers["x-session-id"] = sessionId;
194
- else {
195
- headers.session_id = sessionId;
196
- headers["x-client-request-id"] = sessionId;
197
- headers["x-session-affinity"] = sessionId;
198
- }
199
- }
200
- if (optionsHeaders) Object.assign(headers, optionsHeaders);
201
- const defaultHeaders = model.provider === "cloudflare-ai-gateway" ? {
202
- ...headers,
203
- Authorization: headers.Authorization ?? null,
204
- "cf-aig-authorization": `Bearer ${apiKey}`
205
- } : headers;
206
- const baseUrl = isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model) : model.baseUrl;
207
- return new OpenAI({
208
- apiKey,
209
- baseURL: resolveOpenAIClientBaseUrl(model, baseUrl),
210
- dangerouslyAllowBrowser: true,
211
- defaultHeaders,
212
- maxRetries: 0,
213
- fetch: getAiTransportHost().buildModelFetch(model)
214
- });
215
- }
216
- function buildParams(model, context, options, compat = resolveOpenAICompletionsCompat(model), cacheRetention = resolveCacheRetention(options?.cacheRetention)) {
217
- const cacheControl = getCompatCacheControl(compat, cacheRetention);
218
- const cacheOptOutIndexes = /* @__PURE__ */ new Set();
219
- const messages = convertMessages(model, context, compat, {
220
- cacheOptOutIndexes,
221
- preserveSystemPromptCacheBoundary: cacheControl !== void 0
222
- });
223
- const supportsPromptCacheKey = model.baseUrl.includes("api.openai.com") || compat.supportsPromptCacheKey;
224
- const promptCacheKey = supportsPromptCacheKey && cacheRetention !== "none" ? clampOpenAIPromptCacheKey(options?.promptCacheKey ?? options?.sessionId) : void 0;
225
- const params = {
226
- model: model.id,
227
- messages,
228
- stream: true,
229
- prompt_cache_key: promptCacheKey,
230
- prompt_cache_retention: supportsPromptCacheKey && cacheRetention === "long" && compat.supportsLongCacheRetention ? "24h" : void 0
231
- };
232
- if (compat.supportsUsageInStreaming) params.stream_options = { include_usage: true };
233
- if (compat.supportsStore) params.store = false;
234
- if (options?.maxTokens) {
235
- const maxTokens = clampOpenAICompletionsMaxTokens(model, options.maxTokens);
236
- if (compat.maxTokensField === "max_tokens") params.max_tokens = maxTokens;
237
- else params.max_completion_tokens = maxTokens;
238
- }
239
- if (options?.temperature !== void 0) params.temperature = options.temperature;
240
- if (options?.stop !== void 0 && options.stop.length > 0) params.stop = options.stop;
241
- const requestedResponseFormat = options?.responseFormat;
242
- const responseFormat = requestedResponseFormat === void 0 ? void 0 : resolveOpenAICompletionsResponseFormat(shouldOmitOllamaCompatResponseFormat({
243
- provider: model.provider,
244
- baseUrl: model.baseUrl,
245
- hasTools: () => Boolean(context.tools?.length)
246
- }) ? void 0 : requestedResponseFormat, compat.supportsJsonSchemaResponseFormat);
247
- if (responseFormat !== void 0) params.response_format = responseFormat;
248
- let toolProjection;
249
- if (context.tools) {
250
- const converted = convertTools(context.tools, compat);
251
- toolProjection = converted.projection;
252
- if (converted.tools.length > 0) params.tools = converted.tools;
253
- else if (hasToolCallHistory(context.messages)) params.tools = [];
254
- if (compat.zaiToolStream && converted.tools.length > 0) params.tool_stream = true;
255
- } else if (hasToolCallHistory(context.messages)) params.tools = [];
256
- if (cacheControl) applyAnthropicCacheControl(messages, params.tools, cacheControl, cacheOptOutIndexes);
257
- if (options?.toolChoice) {
258
- const toolChoice = reconcileOpenAICompletionsToolChoice(options.toolChoice, toolProjection ?? projectOpenAITools([]));
259
- if (toolChoice !== void 0) params.tool_choice = toolChoice;
260
- }
261
- const reasoningEffortMap = resolveOpenAIReasoningEffortMap(model);
262
- const thinkingLevelMap = model.thinkingLevelMap;
263
- const offReasoningEffort = reasoningEffortMap.off ?? model.thinkingLevelMap?.off;
264
- const reasoningEffort = options?.reasoningEffort === void 0 ? offReasoningEffort ?? void 0 : reasoningEffortMap[options.reasoningEffort] ?? thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort;
265
- const reasoningEnabled = reasoningEffort !== void 0 && reasoningEffort !== "none";
266
- if (compat.thinkingFormat === "zai" && model.reasoning) params.thinking = reasoningEnabled ? {
267
- type: "enabled",
268
- clear_thinking: false
269
- } : { type: "disabled" };
270
- else if (compat.thinkingFormat === "qwen" && model.reasoning) params.enable_thinking = reasoningEnabled;
271
- else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) params.chat_template_kwargs = {
272
- enable_thinking: reasoningEnabled,
273
- preserve_thinking: true
274
- };
275
- else if (compat.thinkingFormat === "deepseek" && model.reasoning) {
276
- params.thinking = { type: reasoningEnabled ? "enabled" : "disabled" };
277
- if (reasoningEnabled && compat.supportsReasoningEffort) params.reasoning_effort = reasoningEffort;
278
- } else if (compat.thinkingFormat === "openrouter" && model.reasoning) {
279
- const openRouterParams = params;
280
- if (reasoningEnabled) openRouterParams.reasoning = { effort: reasoningEffort };
281
- else if (offReasoningEffort !== null) openRouterParams.reasoning = { effort: offReasoningEffort ?? "none" };
282
- } else if (compat.thinkingFormat === "together" && model.reasoning) {
283
- const togetherParams = params;
284
- togetherParams.reasoning = { enabled: reasoningEnabled };
285
- if (reasoningEnabled && compat.supportsReasoningEffort) togetherParams.reasoning_effort = reasoningEffort;
286
- } else if (reasoningEnabled && model.reasoning && compat.supportsReasoningEffort) params.reasoning_effort = reasoningEffort;
287
- else if (model.reasoning && compat.supportsReasoningEffort) {
288
- if (typeof offReasoningEffort === "string") params.reasoning_effort = offReasoningEffort;
289
- }
290
- if (compat.openRouterRouting) params.provider = compat.openRouterRouting;
291
- if (model.baseUrl.includes("ai-gateway.vercel.sh") && model.compat?.vercelGatewayRouting) {
292
- const routing = model.compat.vercelGatewayRouting;
293
- if (routing.only || routing.order) {
294
- const gatewayOptions = {};
295
- if (routing.only) gatewayOptions.only = routing.only;
296
- if (routing.order) gatewayOptions.order = routing.order;
297
- params.providerOptions = { gateway: gatewayOptions };
298
- }
299
- }
300
- return params;
301
- }
302
- function clampOpenAICompletionsMaxTokens(model, requestedMaxTokens) {
303
- const modelMaxTokens = typeof model.maxTokens === "number" && Number.isFinite(model.maxTokens) && model.maxTokens > 0 ? Math.floor(model.maxTokens) : void 0;
304
- return modelMaxTokens === void 0 || requestedMaxTokens <= modelMaxTokens ? requestedMaxTokens : modelMaxTokens;
305
- }
306
- function getCompatCacheControl(compat, cacheRetention) {
307
- if (compat.cacheControlFormat !== "anthropic" || cacheRetention === "none") return;
308
- const ttl = cacheRetention === "long" && compat.supportsLongCacheRetention ? "1h" : void 0;
309
- return {
310
- type: "ephemeral",
311
- ...ttl ? { ttl } : {}
312
- };
313
- }
314
- function applyAnthropicCacheControl(messages, tools, cacheControl, cacheOptOutIndexes) {
315
- addCacheControlToSystemPrompt(messages, cacheControl);
316
- addCacheControlToLastTool(tools, cacheControl);
317
- addCacheControlToLastConversationMessage(messages, cacheControl, cacheOptOutIndexes);
318
- }
319
- function addCacheControlToSystemPrompt(messages, cacheControl) {
320
- for (const message of messages) if (message.role === "system" || message.role === "developer") {
321
- addCacheControlToInstructionMessage(message, cacheControl);
322
- return;
323
- }
324
- }
325
- function addCacheControlToLastConversationMessage(messages, cacheControl, cacheOptOutIndexes) {
326
- for (let i = messages.length - 1; i >= 0; i--) {
327
- const message = messages[i];
328
- if (!message || cacheOptOutIndexes.has(i)) continue;
329
- if (message.role === "user" || message.role === "assistant") {
330
- if (addCacheControlToMessage(message, cacheControl)) return;
331
- }
332
- }
333
- }
334
- function addCacheControlToLastTool(tools, cacheControl) {
335
- if (!tools || tools.length === 0) return;
336
- const lastTool = tools.at(-1);
337
- if (!lastTool) return;
338
- lastTool.cache_control = cacheControl;
339
- }
340
- function addCacheControlToInstructionMessage(message, cacheControl) {
341
- return addCacheControlToTextContent(message, cacheControl);
342
- }
343
- function addCacheControlToMessage(message, cacheControl) {
344
- if (message.role === "user" || message.role === "assistant") return addCacheControlToTextContent(message, cacheControl);
345
- return false;
346
- }
347
- function addCacheControlToTextContent(message, cacheControl) {
348
- const content = message.content;
349
- if (typeof content === "string") {
350
- if (content.length === 0) return false;
351
- message.content = buildCacheControlledTextParts(content, cacheControl);
352
- return true;
353
- }
354
- if (!Array.isArray(content)) return false;
355
- for (let i = content.length - 1; i >= 0; i--) {
356
- const part = content[i];
357
- if (part?.type === "text") {
358
- const text = part.text;
359
- content.splice(i, 1, ...buildCacheControlledTextParts(text, cacheControl));
360
- return true;
361
- }
362
- }
363
- return false;
364
- }
365
- function buildCacheControlledTextParts(text, cacheControl) {
366
- const split = splitSystemPromptCacheBoundary(text);
367
- if (!split) return [{
368
- type: "text",
369
- text,
370
- cache_control: cacheControl
371
- }];
372
- const parts = [];
373
- if (split.stablePrefix) parts.push({
374
- type: "text",
375
- text: split.stablePrefix,
376
- cache_control: cacheControl
377
- });
378
- if (split.dynamicSuffix) parts.push({
379
- type: "text",
380
- text: split.dynamicSuffix
381
- });
382
- return parts.length > 0 ? parts : [{
383
- type: "text",
384
- text: ""
385
- }];
386
- }
387
- function convertTools(tools, compat) {
388
- const projection = projectOpenAITools(tools);
389
- return {
390
- projection,
391
- tools: sortPromptCacheToolsByName(projection.tools).map((tool) => ({
392
- type: "function",
393
- function: {
394
- name: tool.name,
395
- description: tool.description,
396
- parameters: tool.parameters,
397
- ...compat.supportsStrictMode && { strict: false }
398
- }
399
- }))
400
- };
401
- }
402
- //#endregion
403
- export { streamOpenAICompletions as n, streamSimpleOpenAICompletions as r, openai_completions_exports as t };
@@ -1,21 +0,0 @@
1
- //#region packages/ai/src/providers/openai-prompt-cache.ts
2
- /** Selects the documented cache lifetime fields for a Responses request. */
3
- function resolveOpenAIResponsesCacheParams(model, cacheRetention, supportsLongCacheRetention) {
4
- if (cacheRetention === "none") return {};
5
- if (model.id === "gpt-6-astra" && model.api === "openai-responses") {
6
- const endpoint = URL.parse(model.baseUrl ?? "https://api.openai.com/v1");
7
- if (endpoint?.protocol === "https:" && (endpoint.hostname === "api.openai.com" || endpoint.hostname.endsWith(".api.openai.com"))) return { prompt_cache_options: { ttl: "30m" } };
8
- }
9
- return cacheRetention === "long" && supportsLongCacheRetention ? { prompt_cache_retention: "24h" } : {};
10
- }
11
- /** Maximum prompt cache key length accepted by OpenAI-compatible request metadata. */
12
- const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;
13
- /** Truncates a prompt cache key by Unicode code point count. */
14
- function clampOpenAIPromptCacheKey(key) {
15
- if (key === void 0) return;
16
- const chars = Array.from(key);
17
- if (chars.length <= 64) return key;
18
- return chars.slice(0, 64).join("");
19
- }
20
- //#endregion
21
- export { clampOpenAIPromptCacheKey as n, resolveOpenAIResponsesCacheParams as r, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH as t };
@@ -1,16 +0,0 @@
1
- import { D as Model, l as CacheRetention } from "./types-CJ1-Ht7A.mjs";
2
- import "./types-3Lnm-QSJ.mjs";
3
- //#region packages/ai/src/providers/openai-prompt-cache.d.ts
4
- /** Selects the documented cache lifetime fields for a Responses request. */
5
- declare function resolveOpenAIResponsesCacheParams(model: Pick<Model, "id" | "api" | "baseUrl">, cacheRetention: CacheRetention, supportsLongCacheRetention: boolean): {
6
- prompt_cache_retention?: "24h";
7
- prompt_cache_options?: {
8
- ttl: "30m";
9
- };
10
- };
11
- /** Maximum prompt cache key length accepted by OpenAI-compatible request metadata. */
12
- declare const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;
13
- /** Truncates a prompt cache key by Unicode code point count. */
14
- declare function clampOpenAIPromptCacheKey(key: string | undefined): string | undefined;
15
- //#endregion
16
- export { clampOpenAIPromptCacheKey as n, resolveOpenAIResponsesCacheParams as r, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH as t };
@@ -1,83 +0,0 @@
1
- import { l as requiresClaudeMandatoryAdaptiveThinking, p as resolveClaudeNativeThinkingLevelMap, t as calculateUsageCost } from "./src-2qBGKg8O.mjs";
2
- //#region packages/ai/src/model-utils.ts
3
- /** Calculates and stores model cost fields from token usage and per-million pricing. */
4
- function calculateCost(model, usage) {
5
- Object.assign(usage.cost, calculateUsageCost(usage, model.cost));
6
- return usage.cost;
7
- }
8
- /** Replaces the catalog estimate when the provider reports an authoritative billed total. */
9
- function applyProviderReportedUsageCost(usage, reportedCost) {
10
- if (typeof reportedCost !== "number" || !Number.isFinite(reportedCost) || reportedCost < 0) return;
11
- usage.cost.total = reportedCost;
12
- usage.cost.totalOrigin = "provider-billed";
13
- }
14
- const EXTENDED_THINKING_LEVELS = [
15
- "off",
16
- "minimal",
17
- "low",
18
- "medium",
19
- "high",
20
- "xhigh",
21
- "max"
22
- ];
23
- function resolveThinkingLevelMap(model) {
24
- return model.api === "anthropic-messages" ? resolveClaudeNativeThinkingLevelMap(model) ?? model.thinkingLevelMap : model.thinkingLevelMap;
25
- }
26
- /** Returns thinking levels exposed by a reasoning-capable model. */
27
- function getSupportedThinkingLevels(model) {
28
- const mandatoryAdaptiveContract = model.api === "anthropic-messages" && requiresClaudeMandatoryAdaptiveThinking(model);
29
- if (!model.reasoning && !mandatoryAdaptiveContract) return ["off"];
30
- const thinkingLevelMap = resolveThinkingLevelMap(model);
31
- return EXTENDED_THINKING_LEVELS.filter((level) => {
32
- const mapped = thinkingLevelMap?.[level];
33
- if (mapped === null) return false;
34
- if (level === "xhigh" || level === "max") return mapped !== void 0;
35
- return true;
36
- });
37
- }
38
- /** Clamps a requested thinking level to the closest supported level for a model. */
39
- function clampThinkingLevel(model, level) {
40
- const availableLevels = getSupportedThinkingLevels(model);
41
- if (availableLevels.includes(level)) return level;
42
- const requestedIndex = EXTENDED_THINKING_LEVELS.indexOf(level);
43
- if (requestedIndex === -1) return availableLevels[0] ?? "off";
44
- const thinkingLevelMap = resolveThinkingLevelMap(model);
45
- if ((level === "xhigh" || level === "max") && thinkingLevelMap?.[level] === null) {
46
- for (const candidate of EXTENDED_THINKING_LEVELS.slice(0, requestedIndex).toReversed()) if (availableLevels.includes(candidate)) return candidate;
47
- }
48
- for (const candidate of EXTENDED_THINKING_LEVELS.slice(requestedIndex)) if (availableLevels.includes(candidate)) return candidate;
49
- for (const candidate of EXTENDED_THINKING_LEVELS.slice(0, requestedIndex).toReversed()) if (availableLevels.includes(candidate)) return candidate;
50
- return availableLevels[0] ?? "off";
51
- }
52
- /** Compares model identity by provider and id. */
53
- function modelsAreEqual(a, b) {
54
- if (!a || !b) return false;
55
- return a.id === b.id && a.provider === b.provider;
56
- }
57
- //#endregion
58
- //#region packages/ai/src/utils/sanitize-unicode.ts
59
- /**
60
- * Removes unpaired Unicode surrogate characters from a string.
61
- *
62
- * Unpaired surrogates (high surrogates 0xD800-0xDBFF without matching low surrogates 0xDC00-0xDFFF,
63
- * or vice versa) cause JSON serialization errors in many API providers.
64
- *
65
- * Valid emoji and other characters outside the Basic Multilingual Plane use properly paired
66
- * surrogates and will NOT be affected by this function.
67
- *
68
- * @param text - The text to sanitize
69
- * @returns The sanitized text with unpaired surrogates removed
70
- *
71
- * @example
72
- * // Valid emoji (properly paired surrogates) are preserved
73
- * sanitizeSurrogates("Hello 🙈 World") // => "Hello 🙈 World"
74
- *
75
- * // Unpaired high surrogate is removed
76
- * const unpaired = String.fromCharCode(0xD83D); // high surrogate without low
77
- * sanitizeSurrogates(`Text ${unpaired} here`) // => "Text here"
78
- */
79
- function sanitizeSurrogates(text) {
80
- return text.replace(/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/g, "");
81
- }
82
- //#endregion
83
- export { getSupportedThinkingLevels as a, clampThinkingLevel as i, applyProviderReportedUsageCost as n, modelsAreEqual as o, calculateCost as r, sanitizeSurrogates as t };