@bitkyc08/opencodex 2.100.6 → 2.100.7

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 (103) hide show
  1. package/gui/dist/assets/App-BfoDjNBU.js +65 -0
  2. package/gui/dist/assets/App-hklyKJR5.css +1 -0
  3. package/gui/dist/assets/{Tray-CVqm4S4W.js → Tray-BN6pxM_V.js} +1 -1
  4. package/gui/dist/assets/brand-mark-CDZQ-wa9.js +1 -0
  5. package/gui/dist/assets/index-B85uU8rK.css +2 -0
  6. package/gui/dist/assets/index-Bt9ze3XN.js +138 -0
  7. package/gui/dist/assets/{segment-thumb-features-COdzmORE.js → segment-thumb-features-BM5n8NMX.js} +1 -1
  8. package/gui/dist/classic/assets/App-D4cjNg2I.js +64 -0
  9. package/gui/dist/classic/assets/App-DSivR528.css +1 -0
  10. package/gui/dist/classic/assets/{Tray-BDlcOLDK.js → Tray-DTgGR7wt.js} +1 -1
  11. package/gui/dist/classic/assets/index-ALDeKRg9.js +138 -0
  12. package/gui/dist/classic/assets/usage-companion-chart-BC-Nos9Q.js +1 -0
  13. package/gui/dist/classic/index.html +1 -1
  14. package/gui/dist/index.html +2 -2
  15. package/package.json +2 -2
  16. package/src/adapters/coding-agent/process-tree.ts +30 -0
  17. package/src/adapters/coding-agent/profile.ts +1 -1
  18. package/src/adapters/coding-agent/turn.ts +2 -9
  19. package/src/adapters/command-code-decision-guard.ts +25 -0
  20. package/src/adapters/cursor/catalog.ts +14 -0
  21. package/src/adapters/cursor/effort-map.ts +3 -0
  22. package/src/adapters/google-usage.ts +28 -0
  23. package/src/adapters/google.ts +9 -26
  24. package/src/adapters/qoder/adapter.ts +197 -13
  25. package/src/adapters/registry.ts +13 -0
  26. package/src/adapters/step-cli/adapter.ts +401 -0
  27. package/src/adapters/step-cli/constants.ts +11 -0
  28. package/src/adapters/step-cli/conversation.ts +159 -0
  29. package/src/adapters/step-cli/env.ts +49 -0
  30. package/src/adapters/step-cli/errors.ts +76 -0
  31. package/src/adapters/step-cli/events.ts +227 -0
  32. package/src/adapters/step-cli/profiles.ts +134 -0
  33. package/src/adapters/step-cli/projection.ts +242 -0
  34. package/src/adapters/step-cli/registry.ts +181 -0
  35. package/src/adapters/step-cli/session-pool.ts +225 -0
  36. package/src/adapters/step-cli/session.ts +842 -0
  37. package/src/adapters/step-cli/stdio-frames.ts +330 -0
  38. package/src/adapters/step-cli/tool-bridge.ts +92 -0
  39. package/src/cli/claude.ts +5 -2
  40. package/src/client/link-management-relay.ts +11 -0
  41. package/src/client/machine-listener.ts +6 -2
  42. package/src/codex/account-label.ts +1 -1
  43. package/src/codex/catalog/gather-capture.ts +2 -1
  44. package/src/codex/catalog/metadata.ts +48 -2
  45. package/src/codex/catalog/model-hints.ts +19 -5
  46. package/src/codex/catalog/model-visibility.ts +4 -7
  47. package/src/codex/catalog/native-ultrafast.ts +8 -6
  48. package/src/generated/compatibility-version.json +137 -61
  49. package/src/generated/model-metadata.ts +1 -1
  50. package/src/lib/client-error-body.ts +2 -0
  51. package/src/lib/messages-error-body.ts +3 -1
  52. package/src/lib/optional-shutdown-hooks.ts +55 -5
  53. package/src/lib/process-control.ts +3 -2
  54. package/src/lib/shutdown-budget.ts +5 -0
  55. package/src/link/remote-platform.ts +24 -9
  56. package/src/link/ssh-argv.ts +8 -2
  57. package/src/link/ssh-runner.ts +15 -2
  58. package/src/oauth/m365-copilot-web-session.ts +28 -4
  59. package/src/protocols/dto.ts +24 -12
  60. package/src/protocols/settings.ts +20 -16
  61. package/src/protocols/shadow.ts +2 -3
  62. package/src/protocols/trace.ts +24 -2
  63. package/src/providers/command-code-decision-models.ts +65 -0
  64. package/src/providers/quota.ts +4 -4
  65. package/src/providers/registry/entries-core.ts +18 -7
  66. package/src/providers/registry/entries-extended.ts +41 -0
  67. package/src/providers/registry/model-seeds.ts +42 -0
  68. package/src/providers/resolved-model-policy-merge.ts +8 -0
  69. package/src/providers/resolved-model-policy.ts +185 -165
  70. package/src/providers/slug-codec.ts +5 -0
  71. package/src/server/chat-native-eligibility.ts +5 -0
  72. package/src/server/credential-recovery-attempt.ts +2 -2
  73. package/src/server/gemini-inbound.ts +15 -11
  74. package/src/server/gemini-native-catalog.ts +27 -2
  75. package/src/server/gemini-native-wire.ts +71 -6
  76. package/src/server/gemini-native.ts +129 -34
  77. package/src/server/index/gemini-routes.ts +6 -1
  78. package/src/server/lifecycle.ts +2 -1
  79. package/src/server/management/agent-settings-routes.ts +26 -28
  80. package/src/server/management/api-access.ts +5 -3
  81. package/src/server/management/protocol-routes.ts +3 -1
  82. package/src/server/management/protocol-settings-patch.ts +25 -10
  83. package/src/server/management/shared.ts +20 -5
  84. package/src/server/messages-native-eligibility.ts +4 -0
  85. package/src/server/messages-native-tool-guard.ts +18 -5
  86. package/src/server/request-log-generation-window.ts +7 -2
  87. package/src/server/responses/command-code-decision-refusal.ts +24 -0
  88. package/src/server/responses/compact.ts +2 -2
  89. package/src/server/responses/core-combo.ts +18 -3
  90. package/src/server/responses/core-pipeline.ts +6 -0
  91. package/src/server/responses/sidecar-execution.ts +10 -7
  92. package/src/types/config.ts +2 -0
  93. package/src/usage/expected-prices.ts +17 -0
  94. package/src/usage/telemetry-contract.ts +3 -0
  95. package/gui/dist/assets/App-BI8LQ7Qk.css +0 -1
  96. package/gui/dist/assets/App-CsZ1Gr7S.js +0 -60
  97. package/gui/dist/assets/brand-mark-BebhnAtV.js +0 -1
  98. package/gui/dist/assets/index-BtwVlpeQ.css +0 -2
  99. package/gui/dist/assets/index-CEeMegqE.js +0 -138
  100. package/gui/dist/classic/assets/App-BJgUWNJ8.js +0 -59
  101. package/gui/dist/classic/assets/App-CoiVLRW3.css +0 -1
  102. package/gui/dist/classic/assets/index-B9Jgli_W.js +0 -138
  103. package/gui/dist/classic/assets/usage-companion-chart-vbzDcND0.js +0 -1
@@ -123,9 +123,21 @@ export function clampObservedModelLimits(
123
123
  ...(maxOutputTokens !== undefined ? { maxOutputTokens } : {}),
124
124
  });
125
125
  }
126
+ type PerModelPolicyInput = Pick<ResolveModelPolicyInput, "modelId" | "modelCapabilities" | "inboundWire" | "effectiveAlias" | "effectiveAliasSource">;
127
+ type ProviderPolicyInput = Omit<ResolveModelPolicyInput, keyof PerModelPolicyInput>;
128
+
126
129
  export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedModelPolicy {
127
- const entry = input.transportMatchedRegistry ? input.registryEntry : undefined;
128
- const provider = input.provider;
130
+ return prepareProviderPolicy(input)(input);
131
+ }
132
+
133
+ /** Batch projection for captured catalog providers; the capture already owns registry enrichment. */
134
+ export function prepareModelPolicy(input: Pick<ProviderPolicyInput, "providerName" | "provider" | "effectiveAuth" | "transportMatchedRegistry"> & { transportMatchedRegistry: false }) {
135
+ return prepareProviderPolicy({ ...input, provider: recursivelyFreeze(detachedClone(input.provider)) });
136
+ }
137
+
138
+ function prepareProviderPolicy(shared: ProviderPolicyInput): (modelInput: PerModelPolicyInput) => ResolvedModelPolicy {
139
+ const entry = shared.transportMatchedRegistry ? shared.registryEntry : undefined;
140
+ const provider = shared.provider;
129
141
  const providerPolicy: Partial<StaticProviderPolicyShape> = {};
130
142
  const providerProvenance: Partial<Record<StaticProviderPolicyField, StaticPolicySource>> = {};
131
143
  const [baseUrl, baseUrlSource] = resolvedBaseUrl(entry, provider);
@@ -154,11 +166,11 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
154
166
  putScalar("apiKeyTransport", entry?.apiKeyTransport);
155
167
  const [headers, headersSource] = staticHeaders(entry?.staticHeaders, provider.headers);
156
168
  put("headers", headers, headersSource);
157
- const resolvedAuthMode = input.effectiveAuth?.authMode
169
+ const resolvedAuthMode = shared.effectiveAuth?.authMode
158
170
  ?? (entry?.authKind === "forward" || entry?.authKind === "oauth"
159
171
  ? entry.authKind
160
172
  : provider.authMode === "forward" ? undefined : provider.authMode ?? entry?.authKind);
161
- put("authMode", resolvedAuthMode, input.effectiveAuth
173
+ put("authMode", resolvedAuthMode, shared.effectiveAuth
162
174
  ? "captured-auth"
163
175
  : entry?.authKind === "forward" || entry?.authKind === "oauth" ? "registry"
164
176
  : provider.authMode !== undefined ? "operator" : entry ? "registry" : "unknown");
@@ -186,7 +198,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
186
198
  ] as const) putScalar(key, entry?.[key] as StaticProviderPolicyShape[typeof key] | undefined);
187
199
  // Operator-only display policy; no registry entry declares it (#8363).
188
200
  put("thinkingDisplay", provider.thinkingDisplay, provider.thinkingDisplay !== undefined ? "operator" : "unknown");
189
- const legacyClinePassLadder = entry !== undefined && input.providerName === "cline-pass"
201
+ const legacyClinePassLadder = entry !== undefined && shared.providerName === "cline-pass"
190
202
  && provider.reasoningWireFormat === "gateway-object"
191
203
  && sameStringArray(provider.reasoningEfforts, ["low"]);
192
204
  put("reasoningEfforts", legacyClinePassLadder ? entry?.reasoningEfforts : provider.reasoningEfforts ?? entry?.reasoningEfforts,
@@ -194,7 +206,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
194
206
  put("chatServiceTier", provider.chatServiceTier ?? keyAuthDefaults?.chatServiceTier ?? entry?.chatServiceTier,
195
207
  provider.chatServiceTier !== undefined ? "operator" : keyAuthDefaults?.chatServiceTier !== undefined ? "registry" : entry?.chatServiceTier !== undefined ? "registry" : "unknown");
196
208
  // The Fast switch overrides every capability source; see providerFastSwitchOff.
197
- const fastOff = fastSwitchOff(provider, input.registryEntry);
209
+ const fastOff = fastSwitchOff(provider, shared.registryEntry);
198
210
  if (fastOff) put("supportsServiceTier", false, provider.fastEnabled === false ? "operator" : "registry");
199
211
  else put("supportsServiceTier", provider.supportsServiceTier ?? keyAuthDefaults?.supportsServiceTier ?? entry?.supportsServiceTier,
200
212
  provider.supportsServiceTier !== undefined ? "operator" : keyAuthDefaults?.supportsServiceTier !== undefined ? "registry" : entry?.supportsServiceTier !== undefined ? "registry" : "unknown");
@@ -227,7 +239,7 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
227
239
  put("modelCapabilities", mergeReasoningModelCapabilities(entry?.modelCapabilities, provider.modelCapabilities),
228
240
  provider.modelCapabilities !== undefined ? "operator-capability" : entry?.modelCapabilities !== undefined ? "registry" : "unknown");
229
241
  for (const key of ["modelContextWindows", "modelMaxInputTokens"] as const) {
230
- const [value, source] = input.providerName === "openai-apikey"
242
+ const [value, source] = shared.providerName === "openai-apikey"
231
243
  ? positiveCapMap(entry?.[key], provider[key])
232
244
  : mapFill(entry?.[key], provider[key]);
233
245
  put(key, value, source);
@@ -284,163 +296,171 @@ export function resolveModelPolicy(input: ResolveModelPolicyInput): ResolvedMode
284
296
  (providerPolicy.modelReasoningEffortMap ??= {})[directModel] = {};
285
297
  }
286
298
  }
287
- const modelValue = <T>(record: Readonly<Record<string, T>> | undefined): T | undefined =>
288
- legacyModelValue(record, input.modelId);
289
- const modelSource = <T>(operator: Readonly<Record<string, T>> | undefined, registry: Readonly<Record<string, T>> | undefined): StaticPolicySource =>
290
- legacyModelSource(operator, registry, input.modelId);
291
- const modelOrProviderSource = <T>(
292
- operatorMap: Readonly<Record<string, T>> | undefined, registryMap: Readonly<Record<string, T>> | undefined,
293
- operatorDefault: T | undefined, registryDefault: T | undefined,
294
- ): StaticPolicySource => {
295
- const exact = modelSource(operatorMap, registryMap);
296
- return exact !== "unknown" ? exact : operatorDefault !== undefined ? "operator" : registryDefault !== undefined ? "registry" : "unknown";
297
- };
298
- const configuredAdapter = provider.modelAdapters?.[input.modelId];
299
- const pin = pinnedWireAdapter(input.providerName, input.modelId, provider);
300
- const normalizedModelId = input.modelId.trim().toLowerCase();
301
- const registryWire = wireDefault(entry?.modelWireDefaults?.[normalizedModelId], provider, entry,
302
- input.inboundWire ?? "responses", resolvedAuthMode);
303
- const explicitWire = configuredAdapter && MODEL_ADAPTER_OVERRIDE_ALLOWED.has(configuredAdapter)
304
- ? configuredAdapter
305
- : undefined;
306
- const providerAdapter = providerPolicy.adapter ?? provider.adapter;
307
- const effectiveProviderForWire = resolvedAuthMode === provider.authMode
308
- ? provider
309
- : { ...provider, authMode: resolvedAuthMode };
310
- const canonicalForward = isCanonicalOpenAiForwardProvider(effectiveProviderForWire);
311
- const adapter = pin
312
- ?? (!canonicalForward ? explicitWire ?? registryWire : undefined)
313
- ?? providerAdapter;
314
- const adapterSource: StaticPolicySource = pin
315
- ? "hard-pin"
316
- : explicitWire && !canonicalForward
317
- ? "operator"
318
- : registryWire && !canonicalForward
319
- ? "registry"
320
- : "provider-default";
321
- const declaredModalities = input.modelCapabilities?.inputModalities;
322
- const capabilityModalities = Array.isArray(declaredModalities) && declaredModalities.length > 0
323
- ? [...declaredModalities]
324
- : undefined;
325
- const exactContextWindow = modelValue(providerPolicy.modelContextWindows);
326
- const familyContextWindow = providerPolicy.adapter === "anthropic"
327
- ? anthropicFamilyContextWindow(providerPolicy.modelContextWindows, input.modelId)
328
- : undefined;
329
- const modelContextWindow = exactContextWindow ?? familyContextWindow ?? providerPolicy.contextWindow;
330
- const exactMaxInputTokens = modelValue(providerPolicy.modelMaxInputTokens);
331
- const modelMaxInputTokens = exactMaxInputTokens !== undefined && modelContextWindow !== undefined
332
- ? Math.min(exactMaxInputTokens, modelContextWindow)
333
- : exactMaxInputTokens;
334
- const exactMaxOutputTokens = modelValue(providerPolicy.modelMaxOutputTokens);
335
- const modelMaxOutputTokens = exactMaxOutputTokens ?? providerPolicy.defaultMaxOutputTokens;
336
- const reasoningCapabilities = reasoningCapabilitiesFromConfig({
337
- ...providerPolicy,
338
- adapter: providerPolicy.adapter ?? provider.adapter,
339
- baseUrl: providerPolicy.baseUrl ?? provider.baseUrl,
340
- ...(input.modelCapabilities ? { modelCapabilities: mergeReasoningModelCapabilities(
341
- providerPolicy.modelCapabilities, { [input.modelId]: input.modelCapabilities },
342
- ) } : {}),
343
- }, input.modelId);
344
- const modelReasoningEfforts = reasoningCapabilities.supportsReasoningEffort === false ? []
345
- : modelValue(providerPolicy.modelReasoningEfforts) ?? providerPolicy.reasoningEfforts;
346
- const modelSupportsReasoningSummaries = modelValue(providerPolicy.modelSupportsReasoningSummaries)
347
- ?? (modelValue(providerPolicy.modelReasoningSummaryDelivery) !== undefined ? true : undefined);
348
- const modelSupportsVerbosity = modelValue(providerPolicy.modelSupportsVerbosity) ?? providerPolicy.supportsVerbosity;
349
- const modelSupportsServiceTier = fastOff
350
- ? false
351
- : modelValue(providerPolicy.modelSupportsServiceTier) ?? providerPolicy.supportsServiceTier;
352
- const modelSupportsResponsesCustomTools = input.modelCapabilities?.responsesCustomTools
353
- ?? providerPolicy.supportsResponsesCustomTools;
354
- const model: ResolvedPerModelStaticPolicy = {
355
- adapter,
356
- ...reasoningCapabilities,
357
- ...(modelContextWindow !== undefined ? { contextWindow: modelContextWindow } : {}),
358
- ...(capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) !== undefined
359
- ? { inputModalities: capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) }
360
- : {}),
361
- ...(modelMaxInputTokens !== undefined ? { maxInputTokens: modelMaxInputTokens } : {}),
362
- ...(modelMaxOutputTokens !== undefined ? { maxOutputTokens: modelMaxOutputTokens } : {}),
363
- ...(modelReasoningEfforts !== undefined ? { reasoningEfforts: modelReasoningEfforts } : {}),
364
- ...(reasoningCapabilities.supportsReasoningEffort !== false && modelValue(providerPolicy.modelDefaultReasoningEfforts) !== undefined ? { defaultReasoningEffort: modelValue(providerPolicy.modelDefaultReasoningEfforts) } : {}),
365
- ...(modelSupportsReasoningSummaries !== undefined ? { supportsReasoningSummaries: modelSupportsReasoningSummaries } : {}),
366
- ...(modelSupportsVerbosity !== undefined ? { supportsVerbosity: modelSupportsVerbosity } : {}),
367
- ...(modelSupportsServiceTier !== undefined ? { supportsServiceTier: modelSupportsServiceTier } : {}),
368
- ...(modelSupportsResponsesCustomTools !== undefined ? { supportsResponsesCustomTools: modelSupportsResponsesCustomTools } : {}),
369
- ...(modelSupportsServiceTier === true && entry?.fastTierDescription !== undefined
370
- ? { fastTierDescription: entry.fastTierDescription }
371
- : {}),
372
- ...(entry?.directReasoningEffortModels?.includes(input.modelId) ? { directReasoningEffort: true } : {}),
373
- ...(entry?.modelResponsesUpstreamStreaming?.[normalizedModelId] !== undefined
374
- ? { responsesUpstreamStreaming: entry.modelResponsesUpstreamStreaming[normalizedModelId] }
375
- : {}),
376
- ...(entry?.modelResponsesTerminalRepair?.[normalizedModelId] !== undefined
377
- ? { responsesTerminalRepair: entry.modelResponsesTerminalRepair[normalizedModelId] }
378
- : {}),
379
- };
380
- const modelProvenance: Partial<Record<keyof ResolvedPerModelStaticPolicy, StaticPolicySource>> = {
381
- adapter: adapterSource,
382
- };
383
- for (const field of REASONING_CAPABILITY_FIELDS) {
384
- modelProvenance[field] = reasoningCapabilities[field] === undefined ? "unknown"
385
- : modelInList(provider.noReasoningModels, input.modelId) ? "operator"
386
- : modelInList(entry?.noReasoningModels, input.modelId) ? "registry"
387
- : input.modelCapabilities?.[field] !== undefined || declaredModelCapabilities(provider.modelCapabilities, input.modelId)?.[field] !== undefined
388
- ? "operator-capability" : declaredModelCapabilities(entry?.modelCapabilities, input.modelId)?.[field] !== undefined ? "registry" : "operator";
389
- }
390
- modelProvenance.contextWindow = modelOrProviderSource(provider.modelContextWindows, entry?.modelContextWindows, provider.contextWindow, entry?.contextWindow);
391
- if (modelProvenance.contextWindow === "unknown" && familyContextWindow !== undefined) {
392
- modelProvenance.contextWindow = anthropicFamilyContextWindow(provider.modelContextWindows, input.modelId) !== undefined
393
- ? "operator"
394
- : "registry";
395
- }
396
- modelProvenance.inputModalities = capabilityModalities
397
- ? "operator-capability"
398
- : modelSource(provider.modelInputModalities, entry?.modelInputModalities);
399
- modelProvenance.maxInputTokens = modelSource(provider.modelMaxInputTokens, entry?.modelMaxInputTokens);
400
- modelProvenance.maxOutputTokens = modelOrProviderSource(
401
- provider.modelMaxOutputTokens,
402
- entry?.modelMaxOutputTokens,
403
- provider.defaultMaxOutputTokens,
404
- entry?.defaultMaxOutputTokens,
405
- );
406
- modelProvenance.reasoningEfforts = modelOrProviderSource(provider.modelReasoningEfforts, entry?.modelReasoningEfforts, provider.reasoningEfforts, entry?.reasoningEfforts);
407
- modelProvenance.defaultReasoningEffort = modelSource(
408
- provider.modelDefaultReasoningEfforts,
409
- entry?.modelDefaultReasoningEfforts,
410
- );
411
- modelProvenance.supportsReasoningSummaries = modelSource(provider.modelSupportsReasoningSummaries, entry?.modelSupportsReasoningSummaries);
412
- if (modelProvenance.supportsReasoningSummaries === "unknown" && modelValue(provider.modelReasoningSummaryDelivery) !== undefined) {
413
- modelProvenance.supportsReasoningSummaries = "operator";
299
+ const frozenProvider = recursivelyFreeze(providerPolicy) as ResolvedProviderStaticPolicy;
300
+ const frozenProviderProvenance = Object.freeze(providerProvenance);
301
+ return modelInput => resolvePreparedModel({ ...shared, ...modelInput });
302
+
303
+ function resolvePreparedModel(input: ResolveModelPolicyInput): ResolvedModelPolicy {
304
+ const modelValue = <T>(record: Readonly<Record<string, T>> | undefined): T | undefined =>
305
+ legacyModelValue(record, input.modelId);
306
+ const modelSource = <T>(operator: Readonly<Record<string, T>> | undefined, registry: Readonly<Record<string, T>> | undefined): StaticPolicySource =>
307
+ legacyModelSource(operator, registry, input.modelId);
308
+ const modelOrProviderSource = <T>(
309
+ operatorMap: Readonly<Record<string, T>> | undefined, registryMap: Readonly<Record<string, T>> | undefined,
310
+ operatorDefault: T | undefined, registryDefault: T | undefined,
311
+ ): StaticPolicySource => {
312
+ const exact = modelSource(operatorMap, registryMap);
313
+ return exact !== "unknown" ? exact : operatorDefault !== undefined ? "operator" : registryDefault !== undefined ? "registry" : "unknown";
314
+ };
315
+ const configuredAdapter = provider.modelAdapters?.[input.modelId];
316
+ const pin = pinnedWireAdapter(input.providerName, input.modelId, provider);
317
+ const normalizedModelId = input.modelId.trim().toLowerCase();
318
+ const registryWire = wireDefault(entry?.modelWireDefaults?.[normalizedModelId], provider, entry,
319
+ input.inboundWire ?? "responses", resolvedAuthMode);
320
+ const explicitWire = configuredAdapter && MODEL_ADAPTER_OVERRIDE_ALLOWED.has(configuredAdapter)
321
+ ? configuredAdapter
322
+ : undefined;
323
+ const providerAdapter = providerPolicy.adapter ?? provider.adapter;
324
+ const effectiveProviderForWire = resolvedAuthMode === provider.authMode
325
+ ? provider
326
+ : { ...provider, authMode: resolvedAuthMode };
327
+ const canonicalForward = isCanonicalOpenAiForwardProvider(effectiveProviderForWire);
328
+ const adapter = pin
329
+ ?? (!canonicalForward ? explicitWire ?? registryWire : undefined)
330
+ ?? providerAdapter;
331
+ const adapterSource: StaticPolicySource = pin
332
+ ? "hard-pin"
333
+ : explicitWire && !canonicalForward
334
+ ? "operator"
335
+ : registryWire && !canonicalForward
336
+ ? "registry"
337
+ : "provider-default";
338
+ const declaredModalities = input.modelCapabilities?.inputModalities;
339
+ const capabilityModalities = Array.isArray(declaredModalities) && declaredModalities.length > 0
340
+ ? [...declaredModalities]
341
+ : undefined;
342
+ const exactContextWindow = modelValue(providerPolicy.modelContextWindows);
343
+ const familyContextWindow = providerPolicy.adapter === "anthropic"
344
+ ? anthropicFamilyContextWindow(providerPolicy.modelContextWindows, input.modelId)
345
+ : undefined;
346
+ const modelContextWindow = exactContextWindow ?? familyContextWindow ?? providerPolicy.contextWindow;
347
+ const exactMaxInputTokens = modelValue(providerPolicy.modelMaxInputTokens);
348
+ const modelMaxInputTokens = exactMaxInputTokens !== undefined && modelContextWindow !== undefined
349
+ ? Math.min(exactMaxInputTokens, modelContextWindow)
350
+ : exactMaxInputTokens;
351
+ const exactMaxOutputTokens = modelValue(providerPolicy.modelMaxOutputTokens);
352
+ const modelMaxOutputTokens = exactMaxOutputTokens ?? providerPolicy.defaultMaxOutputTokens;
353
+ const reasoningCapabilities = reasoningCapabilitiesFromConfig({
354
+ ...providerPolicy,
355
+ adapter: providerPolicy.adapter ?? provider.adapter,
356
+ baseUrl: providerPolicy.baseUrl ?? provider.baseUrl,
357
+ // Reasoning reads only this model's declaration; keep the other axes without cloning the roster.
358
+ ...(input.modelCapabilities ? { modelCapabilities: { [input.modelId]: {
359
+ ...declaredModelCapabilities(providerPolicy.modelCapabilities, input.modelId),
360
+ ...input.modelCapabilities,
361
+ } } } : {}),
362
+ }, input.modelId);
363
+ const modelReasoningEfforts = reasoningCapabilities.supportsReasoningEffort === false ? []
364
+ : modelValue(providerPolicy.modelReasoningEfforts) ?? providerPolicy.reasoningEfforts;
365
+ const modelSupportsReasoningSummaries = modelValue(providerPolicy.modelSupportsReasoningSummaries)
366
+ ?? (modelValue(providerPolicy.modelReasoningSummaryDelivery) !== undefined ? true : undefined);
367
+ const modelSupportsVerbosity = modelValue(providerPolicy.modelSupportsVerbosity) ?? providerPolicy.supportsVerbosity;
368
+ const modelSupportsServiceTier = fastOff
369
+ ? false
370
+ : modelValue(providerPolicy.modelSupportsServiceTier) ?? providerPolicy.supportsServiceTier;
371
+ const modelSupportsResponsesCustomTools = input.modelCapabilities?.responsesCustomTools
372
+ ?? providerPolicy.supportsResponsesCustomTools;
373
+ const model: ResolvedPerModelStaticPolicy = {
374
+ adapter,
375
+ ...reasoningCapabilities,
376
+ ...(modelContextWindow !== undefined ? { contextWindow: modelContextWindow } : {}),
377
+ ...(capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) !== undefined
378
+ ? { inputModalities: capabilityModalities ?? modelValue(providerPolicy.modelInputModalities) }
379
+ : {}),
380
+ ...(modelMaxInputTokens !== undefined ? { maxInputTokens: modelMaxInputTokens } : {}),
381
+ ...(modelMaxOutputTokens !== undefined ? { maxOutputTokens: modelMaxOutputTokens } : {}),
382
+ ...(modelReasoningEfforts !== undefined ? { reasoningEfforts: modelReasoningEfforts } : {}),
383
+ ...(reasoningCapabilities.supportsReasoningEffort !== false && modelValue(providerPolicy.modelDefaultReasoningEfforts) !== undefined ? { defaultReasoningEffort: modelValue(providerPolicy.modelDefaultReasoningEfforts) } : {}),
384
+ ...(modelSupportsReasoningSummaries !== undefined ? { supportsReasoningSummaries: modelSupportsReasoningSummaries } : {}),
385
+ ...(modelSupportsVerbosity !== undefined ? { supportsVerbosity: modelSupportsVerbosity } : {}),
386
+ ...(modelSupportsServiceTier !== undefined ? { supportsServiceTier: modelSupportsServiceTier } : {}),
387
+ ...(modelSupportsResponsesCustomTools !== undefined ? { supportsResponsesCustomTools: modelSupportsResponsesCustomTools } : {}),
388
+ ...(modelSupportsServiceTier === true && entry?.fastTierDescription !== undefined
389
+ ? { fastTierDescription: entry.fastTierDescription }
390
+ : {}),
391
+ ...(entry?.directReasoningEffortModels?.includes(input.modelId) ? { directReasoningEffort: true } : {}),
392
+ ...(entry?.modelResponsesUpstreamStreaming?.[normalizedModelId] !== undefined
393
+ ? { responsesUpstreamStreaming: entry.modelResponsesUpstreamStreaming[normalizedModelId] }
394
+ : {}),
395
+ ...(entry?.modelResponsesTerminalRepair?.[normalizedModelId] !== undefined
396
+ ? { responsesTerminalRepair: entry.modelResponsesTerminalRepair[normalizedModelId] }
397
+ : {}),
398
+ };
399
+ const modelProvenance: Partial<Record<keyof ResolvedPerModelStaticPolicy, StaticPolicySource>> = {
400
+ adapter: adapterSource,
401
+ };
402
+ for (const field of REASONING_CAPABILITY_FIELDS) {
403
+ modelProvenance[field] = reasoningCapabilities[field] === undefined ? "unknown"
404
+ : modelInList(provider.noReasoningModels, input.modelId) ? "operator"
405
+ : modelInList(entry?.noReasoningModels, input.modelId) ? "registry"
406
+ : input.modelCapabilities?.[field] !== undefined || declaredModelCapabilities(provider.modelCapabilities, input.modelId)?.[field] !== undefined
407
+ ? "operator-capability" : declaredModelCapabilities(entry?.modelCapabilities, input.modelId)?.[field] !== undefined ? "registry" : "operator";
408
+ }
409
+ modelProvenance.contextWindow = modelOrProviderSource(provider.modelContextWindows, entry?.modelContextWindows, provider.contextWindow, entry?.contextWindow);
410
+ if (modelProvenance.contextWindow === "unknown" && familyContextWindow !== undefined) {
411
+ modelProvenance.contextWindow = anthropicFamilyContextWindow(provider.modelContextWindows, input.modelId) !== undefined
412
+ ? "operator"
413
+ : "registry";
414
+ }
415
+ modelProvenance.inputModalities = capabilityModalities
416
+ ? "operator-capability"
417
+ : modelSource(provider.modelInputModalities, entry?.modelInputModalities);
418
+ modelProvenance.maxInputTokens = modelSource(provider.modelMaxInputTokens, entry?.modelMaxInputTokens);
419
+ modelProvenance.maxOutputTokens = modelOrProviderSource(
420
+ provider.modelMaxOutputTokens,
421
+ entry?.modelMaxOutputTokens,
422
+ provider.defaultMaxOutputTokens,
423
+ entry?.defaultMaxOutputTokens,
424
+ );
425
+ modelProvenance.reasoningEfforts = modelOrProviderSource(provider.modelReasoningEfforts, entry?.modelReasoningEfforts, provider.reasoningEfforts, entry?.reasoningEfforts);
426
+ modelProvenance.defaultReasoningEffort = modelSource(
427
+ provider.modelDefaultReasoningEfforts,
428
+ entry?.modelDefaultReasoningEfforts,
429
+ );
430
+ modelProvenance.supportsReasoningSummaries = modelSource(provider.modelSupportsReasoningSummaries, entry?.modelSupportsReasoningSummaries);
431
+ if (modelProvenance.supportsReasoningSummaries === "unknown" && modelValue(provider.modelReasoningSummaryDelivery) !== undefined) {
432
+ modelProvenance.supportsReasoningSummaries = "operator";
433
+ }
434
+ modelProvenance.supportsVerbosity = modelOrProviderSource(provider.modelSupportsVerbosity, entry?.modelSupportsVerbosity, provider.supportsVerbosity, entry?.supportsVerbosity);
435
+ const exactServiceTierSource = modelSource(provider.modelSupportsServiceTier, registryServiceTierDefaults);
436
+ modelProvenance.supportsServiceTier = !fastOff && exactServiceTierSource !== "unknown" ? exactServiceTierSource
437
+ : providerProvenance.supportsServiceTier ?? "unknown";
438
+ modelProvenance.supportsResponsesCustomTools = input.modelCapabilities?.responsesCustomTools !== undefined
439
+ ? "operator-capability"
440
+ : providerProvenance.supportsResponsesCustomTools ?? "unknown";
441
+ modelProvenance.responsesUpstreamStreaming = model.responsesUpstreamStreaming === undefined ? "unknown" : "registry";
442
+ modelProvenance.responsesTerminalRepair = model.responsesTerminalRepair === undefined ? "unknown" : "registry";
443
+ modelProvenance.fastTierDescription = model.fastTierDescription === undefined ? "unknown" : "registry";
444
+ modelProvenance.directReasoningEffort = model.directReasoningEffort === undefined ? "unknown" : "registry";
445
+ const alias = input.effectiveAlias !== undefined
446
+ ? input.effectiveAlias
447
+ : provider.alias !== undefined
448
+ ? provider.alias
449
+ : entry?.alias;
450
+ const aliasSource: StaticPolicySource = input.effectiveAlias !== undefined
451
+ ? input.effectiveAliasSource ?? "operator"
452
+ : provider.alias !== undefined
453
+ ? "operator"
454
+ : entry?.alias !== undefined ? "registry" : "unknown";
455
+ return Object.freeze({
456
+ version: 1 as const,
457
+ providerName: input.providerName,
458
+ modelId: input.modelId,
459
+ transportMatchedRegistry: entry !== undefined,
460
+ ...(alias !== undefined ? { effectiveAlias: alias } : {}),
461
+ provider: frozenProvider,
462
+ model: recursivelyFreeze(detachedClone(model)),
463
+ provenance: Object.freeze({ alias: aliasSource, provider: frozenProviderProvenance, model: Object.freeze(modelProvenance) }),
464
+ });
414
465
  }
415
- modelProvenance.supportsVerbosity = modelOrProviderSource(provider.modelSupportsVerbosity, entry?.modelSupportsVerbosity, provider.supportsVerbosity, entry?.supportsVerbosity);
416
- const exactServiceTierSource = modelSource(provider.modelSupportsServiceTier, registryServiceTierDefaults);
417
- modelProvenance.supportsServiceTier = !fastOff && exactServiceTierSource !== "unknown" ? exactServiceTierSource
418
- : providerProvenance.supportsServiceTier ?? "unknown";
419
- modelProvenance.supportsResponsesCustomTools = input.modelCapabilities?.responsesCustomTools !== undefined
420
- ? "operator-capability"
421
- : providerProvenance.supportsResponsesCustomTools ?? "unknown";
422
- modelProvenance.responsesUpstreamStreaming = model.responsesUpstreamStreaming === undefined ? "unknown" : "registry";
423
- modelProvenance.responsesTerminalRepair = model.responsesTerminalRepair === undefined ? "unknown" : "registry";
424
- modelProvenance.fastTierDescription = model.fastTierDescription === undefined ? "unknown" : "registry";
425
- modelProvenance.directReasoningEffort = model.directReasoningEffort === undefined ? "unknown" : "registry";
426
- const alias = input.effectiveAlias !== undefined
427
- ? input.effectiveAlias
428
- : provider.alias !== undefined
429
- ? provider.alias
430
- : entry?.alias;
431
- const aliasSource: StaticPolicySource = input.effectiveAlias !== undefined
432
- ? input.effectiveAliasSource ?? "operator"
433
- : provider.alias !== undefined
434
- ? "operator"
435
- : entry?.alias !== undefined ? "registry" : "unknown";
436
- return recursivelyFreeze(detachedClone({
437
- version: 1 as const,
438
- providerName: input.providerName,
439
- modelId: input.modelId,
440
- transportMatchedRegistry: entry !== undefined,
441
- ...(alias !== undefined ? { effectiveAlias: alias } : {}),
442
- provider: providerPolicy as ResolvedProviderStaticPolicy,
443
- model,
444
- provenance: { alias: aliasSource, provider: providerProvenance, model: modelProvenance },
445
- }));
446
466
  }
@@ -85,6 +85,11 @@ export function slugEquals(stored: string, provider: string, id: string): boolea
85
85
  return stored === `${provider}/${id}` || stored === routedSlug(provider, id);
86
86
  }
87
87
 
88
+ /** The same raw-or-encoded match as slugEquals, using a caller's existing index. */
89
+ export function hasRoutedSlug(slugs: ReadonlySet<string>, provider: string, id: string): boolean {
90
+ return slugs.has(`${provider}/${id}`) || (id.includes("/") && slugs.has(routedSlug(provider, id)));
91
+ }
92
+
88
93
  /** Stable key for the exact equivalence relation used by catalog/config slug matching. */
89
94
  export function slugEquivalenceKey(slug: string): string {
90
95
  const slash = slug.indexOf("/");
@@ -4,6 +4,7 @@ import type { ProtocolReasonCode } from "../protocols/contract";
4
4
  import type { RouteResult } from "../router";
5
5
  import type { OcxConfig } from "../types";
6
6
  import { isModelTextOnly, requiresVisionPreprocessing } from "../vision";
7
+ import { isCommandCodeDecisionRoute } from "../providers/command-code-decision-models";
7
8
 
8
9
  type Rec = Record<string, unknown>;
9
10
 
@@ -22,6 +23,7 @@ export type NativeChatDeclineReason = Extract<
22
23
  | "vision-preprocessing"
23
24
  | "hosted-tool"
24
25
  | "opaque-state-stripped"
26
+ | "compatibility-reject"
25
27
  >;
26
28
 
27
29
  /**
@@ -36,6 +38,9 @@ export function nativeChatDeclineReason(
36
38
  ): NativeChatDeclineReason | undefined {
37
39
  const provider = route.provider;
38
40
  if (provider.adapter !== "openai-chat") return "cross-wire-ir";
41
+ // Command Code's System One decision models are not chat models (#8997). The native lane would
42
+ // forward the body verbatim; the translated path refuses it at the adapter guard before a send.
43
+ if (isCommandCodeDecisionRoute(provider, route.modelId)) return "compatibility-reject";
39
44
  if (provider.authMode !== undefined && provider.authMode !== "key" && provider.authMode !== "local") return "auth-mode-not-native";
40
45
  // Combo and policy execution own multi-candidate retries in the Responses pipeline.
41
46
  if (route.combo || route.routeKind === "combo" || route.routeKind === "policy") return "combo-or-policy-route";
@@ -18,8 +18,8 @@ export function sealCredentialRefusal(logCtx: RequestLogContext, status: number,
18
18
  type AttemptSendProvider = Parameters<typeof noteProviderAttemptSend>[2];
19
19
  type AttemptSendRecovery = Parameters<typeof noteProviderAttemptSend>[4];
20
20
 
21
- /** Open a native compact attempt at its first physical send and count each send against it. */
22
- export function noteCompactAttemptSend(
21
+ /** Open an attempt at its first physical send and count each send against it. */
22
+ export function noteCredentialAttemptSend(
23
23
  logCtx: RequestLogContext, route: { providerName: string; modelId: string }, provider: AttemptSendProvider, recovery?: AttemptSendRecovery,
24
24
  ): void {
25
25
  const { providerName, modelId } = route;
@@ -3,8 +3,8 @@
3
3
  * /v1beta|v1/models/{model}:generateContent|streamGenerateContent (countTokens and the embed
4
4
  * methods answer 501).
5
5
  *
6
- * Same shape as the Ollama inbound (./ollama-inbound): the model list describes configured
7
- * provider routes, generation is translated into an internal Chat Completions request handled by
6
+ * Like the Ollama inbound (./ollama-inbound), the catalog starts with configured
7
+ * provider routes. Unless the native leg applies, generation is translated into an internal Chat Completions request handled by
8
8
  * the ordinary Chat path, and this handler owns the final request-log row. Errors use the Gemini
9
9
  * `{ error: { code, message, status } }` body with a fixed message chosen by HTTP status.
10
10
  */
@@ -21,7 +21,7 @@ import { estimateInputTokens } from "./responses/input-admission";
21
21
  import type { DataPlaneAdmission } from "./auth-cors";
22
22
  import { resolveAdmissionModelScope, routeAllowedByScope } from "./admission-model-scope";
23
23
  import type { GeminiMethod } from "./gemini-native-paths";
24
- import { geminiNativeCatalogAliases, geminiNativeCatalogLookup } from "./gemini-native-catalog";
24
+ import { geminiNativeCatalogAliases, geminiNativeCatalogLookup, geminiNativeDiscoveredRoutes } from "./gemini-native-catalog";
25
25
  import {
26
26
  applyLocalRefusalCode,
27
27
  commitCancelled,
@@ -66,10 +66,11 @@ export function toGeminiRefusal(response: Response): Response {
66
66
  return applyLocalRefusalCode(geminiError(parts.status, parts.message, parts.retryAfter), parts.localRefusalCode);
67
67
  }
68
68
 
69
- function eligibleRoutes(config: OcxConfig, req: Request, admission: DataPlaneAdmission) {
70
- const scope = resolveAdmissionModelScope(config, admission);
71
- return listOllamaRoutes(config, req)
72
- .filter(row => routeAllowedByScope(scope, { providerName: row.route.providerName, modelId: row.route.modelId }));
69
+ function geminiCatalogRoutes(config: OcxConfig, req: Request) {
70
+ const configured = listOllamaRoutes(config, req);
71
+ const selectors = new Set(configured.map(row => row.selector));
72
+ const discovered = geminiNativeDiscoveredRoutes(config, req).filter(row => !selectors.has(row.selector));
73
+ return [...configured, ...discovered];
73
74
  }
74
75
 
75
76
  /** Token limits from configuration first, then generated model metadata; omitted when unknown. */
@@ -96,16 +97,19 @@ const modelResource = (config: OcxConfig, row: { selector: string; route: RouteR
96
97
  });
97
98
 
98
99
  export function handleGeminiModels(config: OcxConfig, req: Request, admission: DataPlaneAdmission): Response {
99
- const rows = eligibleRoutes(config, req, admission);
100
- const aliases = geminiNativeCatalogAliases(config, req, rows, resolveAdmissionModelScope(config, admission));
100
+ const scope = resolveAdmissionModelScope(config, admission);
101
+ const candidates = geminiCatalogRoutes(config, req);
102
+ const rows = candidates.filter(row => routeAllowedByScope(scope, row.route));
103
+ // A key can allow a concrete effort tier while excluding its source family.
104
+ const aliases = geminiNativeCatalogAliases(config, req, candidates, scope);
101
105
  return Response.json({ models: [...rows, ...aliases].map(row => modelResource(config, row)) });
102
106
  }
103
107
 
104
108
  function findGeminiCatalogRow(config: OcxConfig, req: Request, admission: DataPlaneAdmission, model: string) {
105
- const row = findOllamaRoute(config, model, req);
109
+ const row = findOllamaRoute(config, model, req) ?? geminiNativeDiscoveredRoutes(config, req, model)[0];
106
110
  const scope = resolveAdmissionModelScope(config, admission);
107
111
  if (row && routeAllowedByScope(scope, row.route)) return row;
108
- return geminiNativeCatalogLookup(config, req, model, () => eligibleRoutes(config, req, admission), scope);
112
+ return geminiNativeCatalogLookup(config, req, model, () => geminiCatalogRoutes(config, req), scope);
109
113
  }
110
114
 
111
115
  export function handleGeminiModel(config: OcxConfig, req: Request, admission: DataPlaneAdmission, model: string): Response {
@@ -1,5 +1,8 @@
1
+ import { getStaleCached } from "../codex/model-cache";
2
+ import { filterCatalogVisibleModels } from "../codex/catalog/model-visibility";
1
3
  import { resolveProtocolSettings } from "../protocols/settings";
2
4
  import { resolveAntigravityEffortWireModel } from "../providers/antigravity-models";
5
+ import { slugEquals } from "../providers/slug-codec";
3
6
  import { effectiveGoogleMode } from "../providers/registry";
4
7
  import type { RouteResult } from "../router";
5
8
  import type { OcxConfig } from "../types";
@@ -35,7 +38,7 @@ export function nativeGeminiDestination(config: OcxConfig, req: Request, selecto
35
38
 
36
39
  /** A native alias must route to the same wire ID and pass scope under its own resolved destination. */
37
40
  function allowedWireAlias(config: OcxConfig, req: Request, alias: string, scope?: AdmissionModelScope): CatalogRow | undefined {
38
- if (!GEMINI_MODEL.test(alias)) return undefined;
41
+ if (!GEMINI_MODEL.test(alias) || nativeModelDisabled(config, alias)) return undefined;
39
42
  const destination = resolveNativeDestination(config, req, alias);
40
43
  if (destination?.wireModel !== alias || !routeAllowedByScope(scope, destination.route)) return undefined;
41
44
  return { selector: alias, route: destination.route };
@@ -43,6 +46,27 @@ function allowedWireAlias(config: OcxConfig, req: Request, alias: string, scope?
43
46
 
44
47
  type CatalogRow = { selector: string; route: RouteResult };
45
48
 
49
+ /** Read the same accepted provider roster as the other catalogs; do not start discovery here. */
50
+ export function geminiNativeDiscoveredRoutes(config: OcxConfig, req: Request, selector?: string): CatalogRow[] {
51
+ const provider = config.providers[PROVIDER];
52
+ if (!resolveProtocolSettings(config).rollout.geminiNativeAntigravity
53
+ || !provider || provider.disabled === true || provider.liveModels === false
54
+ || (selector !== undefined && !selector.startsWith(`${PROVIDER}/`))) return [];
55
+ const cached = getStaleCached(PROVIDER) ?? [];
56
+ const models = filterCatalogVisibleModels(cached.filter(model =>
57
+ model.provider === PROVIDER && GEMINI_MODEL.test(model.id)), config);
58
+ return models.flatMap(({ id }) => {
59
+ const qualified = `${PROVIDER}/${id}`;
60
+ if (selector !== undefined && selector !== qualified) return [];
61
+ const destination = resolveNativeDestination(config, req, qualified);
62
+ return destination?.route.modelId === id ? [{ selector: qualified, route: destination.route }] : [];
63
+ });
64
+ }
65
+
66
+ function nativeModelDisabled(config: OcxConfig, modelId: string): boolean {
67
+ return config.disabledModels?.some(slug => slugEquals(slug, PROVIDER, modelId)) ?? false;
68
+ }
69
+
46
70
  export function geminiNativeCatalogLookup(
47
71
  config: OcxConfig, req: Request, alias: string, rows: () => readonly CatalogRow[], scope?: AdmissionModelScope,
48
72
  ): CatalogRow | undefined {
@@ -52,7 +76,8 @@ export function geminiNativeCatalogLookup(
52
76
  }
53
77
 
54
78
  function advertisedWireModels(config: OcxConfig, route: RouteResult): string[] {
55
- return [undefined, "low", "medium", "high"].flatMap(effort => nativeWireModel(config, route, effort) ?? []);
79
+ if (route.providerName !== PROVIDER || nativeModelDisabled(config, route.modelId)) return [];
80
+ return [undefined, "low", "medium", "high", "xhigh"].flatMap(effort => nativeWireModel(config, route, effort) ?? []);
56
81
  }
57
82
 
58
83
  export function geminiNativeCatalogAliases(