dsh-vertex-argon 1.6.3 → 2.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/adapter.js CHANGED
@@ -23,29 +23,16 @@ export const PROVIDER = 'google-vertex-argon'
23
23
  export const NATIVE_PROVIDER = 'google-vertex'
24
24
  export const MODEL = 'gemini-4-argon'
25
25
  export const GOOGLE_API_MODEL = 'gemini-4-argon-google'
26
- export const PROXY_MODEL = 'gemini-4-argon-proxy'
27
- /** Antigravity 反代暴露的 Gemini 3.8 Flash;与 Argon 共用同一份反代地址与专属密钥。 */
28
- export const PROXY_FLASH_MODEL = 'gemini-3.8-flash-proxy'
29
- export const FLASH_UPSTREAM_MODEL = 'gemini-3.8-flash'
30
- export const PROXY_MODELS = Object.freeze([PROXY_MODEL, PROXY_FLASH_MODEL])
31
- export const PROXY_MODEL_NAMES = Object.freeze({
32
- [PROXY_MODEL]: 'Gemini 4 Argon (Antigravity)',
33
- [PROXY_FLASH_MODEL]: 'Gemini 3.8 Flash (Antigravity)',
34
- })
35
- /** 反代模型判定:Argon 与 Flash 都走同一条 Antigravity 通道。 */
36
- export function isProxyModel(modelId) {
37
- return typeof modelId === 'string' && PROXY_MODELS.includes(modelId)
26
+ export const CARBON_MODEL = 'gemini-4-carbon'
27
+ export const GOOGLE_API_CARBON_MODEL = 'gemini-4-carbon-google'
28
+ /** Gemini API 官方直连模型判定(Argon 与 Carbon)。 */
29
+ export function isGoogleApiModel(modelId) {
30
+ return modelId === GOOGLE_API_MODEL || modelId === GOOGLE_API_CARBON_MODEL
38
31
  }
39
- /** 反代模型对应的上游基础 ID;档位后缀由 resolveProxyUpstreamModel 追加。 */
40
- export function proxyUpstreamBaseModel(modelId) {
41
- return modelId === PROXY_FLASH_MODEL ? FLASH_UPSTREAM_MODEL : MODEL
32
+ /** 官方直连(Vertex AI / Gemini API)对应的上游模型 ID。 */
33
+ export function officialUpstreamModel(modelId) {
34
+ return modelId === CARBON_MODEL || modelId === GOOGLE_API_CARBON_MODEL ? CARBON_MODEL : MODEL
42
35
  }
43
- /** 反代上游模型 ID:`<基础 ID>-<low|medium|high>`,与反代暴露的档位模型一一对应。 */
44
- export function resolveProxyUpstreamModel(modelId, level) {
45
- return `${proxyUpstreamBaseModel(modelId)}-${level}`
46
- }
47
- export const PROXY_API_KEY_ENV = 'ANTIGRAVITY_API_KEY'
48
- export const DEFAULT_PROXY_BASE_URL = 'http://127.0.0.1:8045/v1beta'
49
36
  export const LEVELS = ['low', 'medium', 'high']
50
37
  export const MAX_SINGLE_CALL_TOKENS = 65536
51
38
  export const MAX_LONG_DECODE_TOKENS = 1048576
@@ -53,18 +40,18 @@ const COMPACTION_HEADROOM_FLOOR = 131072
53
40
  const ENV_KEYS = ['GOOGLE_APPLICATION_CREDENTIALS', 'GOOGLE_CLOUD_PROJECT', 'GCLOUD_PROJECT', 'GOOGLE_CLOUD_LOCATION', 'GOOGLE_CLOUD_API_KEY']
54
41
  const DEFAULT_THINKING_MAP = { off: null, minimal: null, low: 'low', medium: 'medium', high: 'high', xhigh: null, max: null }
55
42
 
56
- const readVal = v => (v && typeof v === 'object' && typeof v.get === 'function') ? v.get() : v
57
-
58
- export function normalizeProxyBaseUrl(raw) {
59
- const value = typeof raw === 'string' ? raw.trim() : raw
60
- if (value != null && typeof value !== 'string') throw new LlmError('Antigravity proxy base URL must be a string.', 'INVALID_CONFIG')
61
- let url
62
- try { url = new URL(value || DEFAULT_PROXY_BASE_URL) } catch { throw new LlmError('Antigravity proxy base URL must be a valid HTTP(S) URL.', 'INVALID_CONFIG') }
63
- if (!['http:', 'https:'].includes(url.protocol)) throw new LlmError('Antigravity proxy base URL must use HTTP or HTTPS.', 'INVALID_CONFIG')
64
- url.pathname = (!url.pathname || url.pathname === '/') ? '/v1beta' : url.pathname.replace(/[/]+$/, '')
65
- return url.href
43
+ // The host's agent/request hook already narrows 'max' to 'high'. Direct adapter
44
+ // callers (advisor tools calling adapter.stream / prepareCall().stream) bypass
45
+ // that hook, so the same narrowing has to happen at this boundary too.
46
+ export function narrowMaxReasoning(options) {
47
+ if (!options || typeof options !== 'object') return options
48
+ const requested = [options.reasoning, options.reasoningEffort]
49
+ if (!requested.some(value => typeof value === 'string' && value.toLowerCase() === 'max')) return options
50
+ return { ...options, reasoning: 'high', reasoningEffort: 'high' }
66
51
  }
67
52
 
53
+ const readVal = v => (v && typeof v === 'object' && typeof v.get === 'function') ? v.get() : v
54
+
68
55
  export function isGemini4Model(model) {
69
56
  return Boolean(
70
57
  model &&
@@ -76,16 +63,26 @@ export function isGemini4Model(model) {
76
63
 
77
64
  function googleKeyEnvironment(value) {
78
65
  const name = typeof value === 'string' ? value.trim() : ''
79
- // The proxy credential namespace is reserved, including Windows case aliases.
80
- return !name || name.toUpperCase() === PROXY_API_KEY_ENV ? 'GOOGLE_API_KEY' : name
66
+ return name || 'GOOGLE_API_KEY'
81
67
  }
82
68
 
83
69
  export function detectPiAiArgonRoutes(ctx, explicitRoutes) {
84
70
  if (explicitRoutes && typeof explicitRoutes === 'object') {
71
+ const hasVertexCarbon = Boolean(explicitRoutes.hasVertexCarbon)
72
+ const hasGoogleCarbon = Boolean(explicitRoutes.hasGoogleCarbon)
73
+ const hasVertexArgon = explicitRoutes.hasVertexArgon !== undefined
74
+ ? Boolean(explicitRoutes.hasVertexArgon)
75
+ : explicitRoutes.hasVertex !== false
76
+ const hasGoogleArgon = explicitRoutes.hasGoogleArgon !== undefined
77
+ ? Boolean(explicitRoutes.hasGoogleArgon)
78
+ : Boolean(explicitRoutes.hasGoogle)
85
79
  return {
86
- hasVertex: explicitRoutes.hasVertex !== false,
87
- hasGoogle: Boolean(explicitRoutes.hasGoogle),
88
- hasProxy: explicitRoutes.hasProxy === true,
80
+ hasVertex: hasVertexArgon || hasVertexCarbon,
81
+ hasGoogle: hasGoogleArgon || hasGoogleCarbon,
82
+ hasVertexArgon,
83
+ hasGoogleArgon,
84
+ hasVertexCarbon,
85
+ hasGoogleCarbon,
89
86
  googleApiKeyEnv: googleKeyEnvironment(explicitRoutes.googleApiKeyEnv),
90
87
  googleBaseUrl: explicitRoutes.googleBaseUrl,
91
88
  vertexBaseUrl: explicitRoutes.vertexBaseUrl,
@@ -98,22 +95,68 @@ export function detectPiAiArgonRoutes(ctx, explicitRoutes) {
98
95
  const providers = readVal(entry?.fiber?.config?.providers) ?? readVal(entry?.options?.config?.providers) ?? {}
99
96
  const googleCfg = providers?.google
100
97
  const vertexCfg = providers?.['google-vertex']
101
- const hasModel = list => Array.isArray(list) && list.some(m => /^(?:models\/)?gemini-4-argon\b/i.test(String(readVal(m?.id) ?? '')))
102
- const hasGoogle = hasModel(readVal(googleCfg?.models))
103
- const hasVertex = hasModel(readVal(vertexCfg?.models))
98
+ const hasModel = (list, re) => Array.isArray(list) && list.some(m => re.test(String(readVal(m?.id) ?? '')))
99
+ const googleModels = readVal(googleCfg?.models)
100
+ const vertexModels = readVal(vertexCfg?.models)
101
+ const hasGoogleArgon = hasModel(googleModels, /^(?:models\/)?gemini-4-argon\b/i)
102
+ const hasVertexArgon = hasModel(vertexModels, /^(?:models\/)?gemini-4-argon\b/i)
103
+ const hasGoogleCarbon = hasModel(googleModels, /^(?:models\/)?gemini-4-carbon\b/i)
104
+ const hasVertexCarbon = hasModel(vertexModels, /^(?:models\/)?gemini-4-carbon\b/i)
105
+ const hasGoogle = hasGoogleArgon || hasGoogleCarbon
106
+ const hasVertex = hasVertexArgon || hasVertexCarbon || !hasGoogle
104
107
  return {
105
- hasVertex: hasVertex || !hasGoogle,
108
+ hasVertex,
106
109
  hasGoogle,
110
+ hasVertexArgon: hasVertexArgon || (!hasGoogle && !hasVertexCarbon),
111
+ hasGoogleArgon,
112
+ hasVertexCarbon,
113
+ hasGoogleCarbon,
107
114
  googleApiKeyEnv: googleKeyEnvironment(readVal(googleCfg?.apiKeyEnv)),
108
115
  googleBaseUrl: readVal(googleCfg?.baseURL),
109
116
  vertexBaseUrl: readVal(vertexCfg?.baseURL),
110
117
  }
111
118
  }
112
119
  } catch {}
113
- return { hasVertex: true, hasGoogle: false, googleApiKeyEnv: 'GOOGLE_API_KEY' }
120
+ return {
121
+ hasVertex: true,
122
+ hasGoogle: false,
123
+ hasVertexArgon: true,
124
+ hasGoogleArgon: false,
125
+ hasVertexCarbon: false,
126
+ hasGoogleCarbon: false,
127
+ googleApiKeyEnv: 'GOOGLE_API_KEY',
128
+ }
114
129
  }
115
130
 
116
131
  const TRANSIENT_GEMINI_STOP_RE = /(?:Unhandled stop reason:\s*(?:MALFORMED_RESPONSE|MALFORMED_FUNCTION_CALL|UNEXPECTED_TOOL_CALL|TOO_MANY_TOOL_CALLS)|Provider stopped with:\s*(?:MALFORMED_RESPONSE|MALFORMED_FUNCTION_CALL|UNEXPECTED_TOOL_CALL|OTHER|FINISH_REASON_UNSPECIFIED))/i
132
+ const AUTO_TOOL_MODE_TTL_MS = 180_000
133
+ const autoToolModeFallbacks = new Map()
134
+
135
+ function retryKeyOf(requestModel, context, options) {
136
+ const tail = context?.messages?.at?.(-1)
137
+ const tailHint = typeof tail?.content === 'string' ? tail.content.slice(0, 160) : JSON.stringify(tail?.content ?? '').slice(0, 160)
138
+ return `${options?.sessionId ?? 'anon'}:${requestModel?.id ?? ''}:${context?.messages?.length ?? 0}:${tailHint}`
139
+ }
140
+
141
+ function markAutoToolModeFallback(key) {
142
+ if (!key) return
143
+ if (autoToolModeFallbacks.size >= 256) autoToolModeFallbacks.delete(autoToolModeFallbacks.keys().next().value)
144
+ autoToolModeFallbacks.set(key, Date.now())
145
+ }
146
+
147
+ function shouldFallbackToAutoToolMode(key) {
148
+ const ts = key ? autoToolModeFallbacks.get(key) : undefined
149
+ if (ts === undefined) return false
150
+ if (Date.now() - ts > AUTO_TOOL_MODE_TTL_MS) {
151
+ autoToolModeFallbacks.delete(key)
152
+ return false
153
+ }
154
+ return true
155
+ }
156
+
157
+ function clearAutoToolModeFallback(key) {
158
+ if (key) autoToolModeFallbacks.delete(key)
159
+ }
117
160
 
118
161
  let apiClientProtoRef = null
119
162
  let apiClientPatch = null
@@ -154,13 +197,14 @@ export function ensureGemini4ApiClientPatch() {
154
197
  return patch.dispose
155
198
  }
156
199
 
157
- function wrapGemini4Stream(rawStream, tracker, signal) {
200
+ function wrapGemini4Stream(rawStream, tracker, signal, retryKey) {
158
201
  if (!rawStream || typeof rawStream[Symbol.asyncIterator] !== 'function') return rawStream
159
202
  return {
160
203
  [Symbol.asyncIterator]: async function* () {
161
204
  try {
162
205
  for await (const event of rawStream) {
163
206
  if (event?.type === 'done') {
207
+ clearAutoToolModeFallback(retryKey)
164
208
  const stop = event.message?.stopReason
165
209
  const outcome = stop === 'toolUse' ? 'tool-calls' : stop === 'length' ? 'max-tokens' : 'done'
166
210
  tracker?.finishStream({ outcome })
@@ -168,6 +212,9 @@ function wrapGemini4Stream(rawStream, tracker, signal) {
168
212
  const isTransient =
169
213
  typeof event?.error?.errorMessage === 'string' &&
170
214
  TRANSIENT_GEMINI_STOP_RE.test(event.error.errorMessage)
215
+ if (isTransient || event?.error?.code === 'STREAM_PHASE_TIMEOUT' || /stream phase timed out/i.test(String(event?.error?.errorMessage ?? ''))) {
216
+ markAutoToolModeFallback(retryKey)
217
+ }
171
218
  tracker?.finishStream({
172
219
  outcome: signal?.aborted ? 'aborted' : 'error',
173
220
  error: event.error,
@@ -187,6 +234,9 @@ function wrapGemini4Stream(rawStream, tracker, signal) {
187
234
  yield event
188
235
  }
189
236
  } catch (error) {
237
+ if (error?.code === 'STREAM_PHASE_TIMEOUT' || /stream phase timed out/i.test(String(error?.message ?? ''))) {
238
+ markAutoToolModeFallback(retryKey)
239
+ }
190
240
  tracker?.finishStream({
191
241
  outcome: signal?.aborted ? 'aborted' : 'error',
192
242
  error,
@@ -216,11 +266,11 @@ export function gemini4StreamSimple(
216
266
  assertPreparedStructuredOutput(structuredOutput)
217
267
  ensureGemini4ApiClientPatch()
218
268
  const fallback = defaultLevel ?? (/cyber-permissive/i.test(requestModel?.id ?? '') ? 'high' : 'medium')
219
- const rawLevel = options.reasoning ?? fallback
269
+ const requestedLevel = options.reasoning ?? fallback
270
+ const rawLevel = typeof requestedLevel === 'string' && requestedLevel.toLowerCase() === 'max' ? 'high' : requestedLevel
220
271
  const mapped = requestModel?.thinkingLevelMap?.[rawLevel] ?? rawLevel
221
272
  const level = typeof mapped === 'string' ? mapped.toLowerCase() : fallback
222
- if (!LEVELS.includes(level)) throw new LlmError('Gemini 4 Argon requires low, medium or high.', 'UNSUPPORTED_REASONING_EFFORT')
223
- const resolvedUpstreamModelId = isProxyModel(requestModel?.id) ? resolveProxyUpstreamModel(requestModel.id, level) : upstreamModelId
273
+ if (!LEVELS.includes(level)) throw new LlmError('Gemini 4 (Argon / Carbon) requires low, medium or high.', 'UNSUPPORTED_REASONING_EFFORT')
224
274
  const { reasoning, thinking, thinkingBudgets, temperature, onPayload: prior, onProviderStreamEvent: priorEvent, ...rest } = options
225
275
  const requestedMaxTokens = options.maxTokens ?? requestModel.maxTokens
226
276
  const effectiveMaxTokens =
@@ -242,6 +292,7 @@ export function gemini4StreamSimple(
242
292
  captureOptions,
243
293
  )
244
294
  const signatures = createGoogleSignatureBridge(context, requestModel)
295
+ const retryKey = retryKeyOf(requestModel, context, options)
245
296
  return wrapGemini4Stream(
246
297
  signatures.stream(nativeStream(requestModel, signatures.context, {
247
298
  ...rest,
@@ -251,10 +302,10 @@ export function gemini4StreamSimple(
251
302
  const replacement = await prior?.(payload, payloadModel)
252
303
  let body = replacement === undefined ? payload : replacement
253
304
  if (!body || typeof body !== 'object' || Array.isArray(body) || !Array.isArray(body.contents) || !body.config || typeof body.config !== 'object' || Array.isArray(body.config)) {
254
- throw new LlmError('Gemini 4 Argon received an incompatible native payload.', 'INVALID_REQUEST')
305
+ throw new LlmError('Gemini 4 received an incompatible native payload.', 'INVALID_REQUEST')
255
306
  }
256
- if (body.model !== requestModel.id && body.model !== upstreamModelId && body.model !== resolvedUpstreamModelId) {
257
- throw new LlmError('Gemini 4 Argon received an incompatible model payload.', 'INVALID_REQUEST')
307
+ if (body.model !== requestModel.id && body.model !== upstreamModelId) {
308
+ throw new LlmError('Gemini 4 received an incompatible model payload.', 'INVALID_REQUEST')
258
309
  }
259
310
  body = signatures.restorePayload(body)
260
311
  if (structuredOutput) body = structuredOutput.generateContent(body)
@@ -263,15 +314,16 @@ export function gemini4StreamSimple(
263
314
  cleanConfig.httpOptions && typeof cleanConfig.httpOptions === 'object' && !Array.isArray(cleanConfig.httpOptions)
264
315
  ? cleanConfig.httpOptions
265
316
  : {}
317
+ const toolMode = shouldFallbackToAutoToolMode(retryKey) ? 'AUTO' : 'VALIDATED'
266
318
  return {
267
319
  ...body,
268
- model: resolvedUpstreamModelId,
320
+ model: upstreamModelId,
269
321
  config: {
270
322
  ...cleanConfig,
271
323
  ...(cleanConfig.tools?.some(tool => tool.functionDeclarations?.length) ? {
272
324
  toolConfig: {
273
325
  ...cleanConfig.toolConfig,
274
- functionCallingConfig: { ...cleanConfig.toolConfig?.functionCallingConfig, mode: 'VALIDATED' },
326
+ functionCallingConfig: { ...cleanConfig.toolConfig?.functionCallingConfig, mode: toolMode },
275
327
  },
276
328
  } : {}),
277
329
  automaticContinuation: false,
@@ -291,6 +343,7 @@ export function gemini4StreamSimple(
291
343
  })),
292
344
  tracker,
293
345
  options.signal,
346
+ retryKey,
294
347
  )
295
348
  }
296
349
 
@@ -366,10 +419,6 @@ export function resolveConfigValues(raw = {}) {
366
419
  const googleServiceTier = readVal(source?.googleServiceTier) ?? 'standard'
367
420
  const vertexServiceTier = readVal(source?.vertexServiceTier) ?? 'standard'
368
421
  const vertexPaygoOnly = readVal(source?.vertexPaygoOnly) ?? false
369
- const proxyEnabled = readVal(source?.proxyEnabled) ?? false
370
- const proxyBaseUrl = normalizeProxyBaseUrl(readVal(source?.proxyBaseUrl))
371
- const proxyApiKey = readVal(source?.proxyApiKey) ?? ''
372
- if (typeof proxyEnabled !== 'boolean' || typeof proxyApiKey !== 'string') throw new LlmError('Invalid Antigravity proxy configuration.', 'INVALID_CONFIG')
373
422
  const retainRawResponses = readVal(source?.retainRawResponses) ?? false
374
423
  if (typeof retainRawResponses !== 'boolean') throw new LlmError('Raw response retention consent must be an explicit boolean.', 'INVALID_CONFIG')
375
424
  if (![googleServiceTier, vertexServiceTier].every(tier => SERVICE_TIERS.includes(tier))) throw new LlmError('Invalid Argon service tier.', 'INVALID_CONFIG')
@@ -399,9 +448,6 @@ export function resolveConfigValues(raw = {}) {
399
448
  googleServiceTier,
400
449
  vertexServiceTier,
401
450
  vertexPaygoOnly,
402
- proxyEnabled,
403
- proxyBaseUrl,
404
- proxyApiKey: proxyApiKey.trim(),
405
451
  retainRawResponses,
406
452
  interactionsBackground,
407
453
  allowInteractionStorage,
@@ -409,44 +455,43 @@ export function resolveConfigValues(raw = {}) {
409
455
  }
410
456
  }
411
457
 
412
- /** Serve Vertex AI, Gemini API and isolated Antigravity proxy models in one plugin group. */
458
+ /** Serve the official Vertex AI and Gemini API models in one plugin group. */
413
459
  export function argonProvider(nativeVertex, models, defaultLevel = 'medium', nativeGoogle = googleProvider(), longDecode = true, recordDiagnostic, serviceConfig = {}, responseHub, structuredOutput) {
414
460
  const modelMap = new Map((Array.isArray(models) ? models : [models]).map(m => [m.id, m]))
415
461
  return {
416
462
  id: PROVIDER,
417
- name: 'Vertex Argon',
463
+ name: 'Gemini 4 Adapter',
418
464
  auth: nativeVertex.auth,
419
465
  getModels: () => [...modelMap.values()],
420
466
  streamSimple(requestModel, context, options = {}) {
421
467
  const configured = modelMap.get(requestModel?.id)
422
468
  if (!configured || requestModel.provider !== PROVIDER) {
423
- throw new LlmError('Vertex Argon only serves configured Gemini 4 Argon routes.', 'UNSUPPORTED_OPTION')
469
+ throw new LlmError('Gemini 4 Adapter only serves configured Gemini 4 routes.', 'UNSUPPORTED_OPTION')
424
470
  }
425
471
  const { env: rawEnv, apiKey: vertexApiKey, ...restOptions } = options
426
472
  const cleanEnv = rawEnv ? { ...rawEnv } : undefined
427
473
  if (cleanEnv) {
428
474
  delete cleanEnv.__GOOGLE_API_KEY__
429
- delete cleanEnv.__ANTIGRAVITY_API_KEY__
430
475
  }
431
476
  if (configured.api === 'google-generative-ai') {
432
- const proxy = isProxyModel(configured.id)
433
- const apiKey = proxy ? rawEnv?.__ANTIGRAVITY_API_KEY__ : rawEnv?.__GOOGLE_API_KEY__
434
- if (!apiKey) throw new LlmError(proxy ? 'Antigravity proxy requires its dedicated API key.' : 'Gemini API credentials are not configured.', 'MISSING_CREDENTIAL')
477
+ const apiKey = rawEnv?.__GOOGLE_API_KEY__
478
+ if (!apiKey) throw new LlmError('Gemini API credentials are not configured.', 'MISSING_CREDENTIAL')
479
+ const policy = resolveServicePolicy(serviceConfig, 'google-generative-ai', 'generate-content')
435
480
  return gemini4StreamSimple(
436
481
  (m, c, o) => nativeGoogle.stream(m, c, o),
437
482
  requestModel,
438
483
  context,
439
484
  { ...restOptions, ...(cleanEnv ? { env: cleanEnv } : {}), ...(apiKey ? { apiKey } : {}), maxTokens: options.maxTokens ?? configured.maxTokens },
440
485
  defaultLevel,
441
- proxy ? proxyUpstreamBaseModel(configured.id) : MODEL,
486
+ officialUpstreamModel(configured.id),
442
487
  longDecode,
443
488
  recordDiagnostic,
444
- resolveServicePolicy(serviceConfig, 'google-generative-ai', 'generate-content'),
489
+ policy,
445
490
  { responseHub, retainRawResponses: serviceConfig.retainRawResponses }, structuredOutput,
446
491
  )
447
492
  }
448
493
  if (requestModel.api !== 'google-vertex') {
449
- throw new LlmError('Vertex Argon only serves google-vertex or google-generative-ai.', 'UNSUPPORTED_OPTION')
494
+ throw new LlmError('Gemini 4 Adapter only serves google-vertex or google-generative-ai.', 'UNSUPPORTED_OPTION')
450
495
  }
451
496
  return gemini4StreamSimple(
452
497
  (m, c, o) => nativeVertex.stream(m, c, o),
@@ -454,7 +499,7 @@ export function argonProvider(nativeVertex, models, defaultLevel = 'medium', nat
454
499
  context,
455
500
  { ...restOptions, ...(cleanEnv ? { env: cleanEnv } : {}), ...(vertexApiKey ? { apiKey: vertexApiKey } : {}), maxTokens: options.maxTokens ?? configured.maxTokens },
456
501
  defaultLevel,
457
- MODEL,
502
+ officialUpstreamModel(configured.id),
458
503
  longDecode,
459
504
  recordDiagnostic,
460
505
  resolveServicePolicy(serviceConfig, 'google-vertex', 'generate-content'),
@@ -477,7 +522,7 @@ export class ArgonAdapter extends LlmAdapter {
477
522
  imageRequestPricing(provider, model) { return this.inner.imageRequestPricing?.(provider, model) }
478
523
  imageRequestTarget(provider, model, image) {
479
524
  const cfg = this.configSnapshot?.()
480
- const protocol = (model === GOOGLE_API_MODEL || isProxyModel(model)) ? cfg?.googleProtocol : cfg?.vertexProtocol
525
+ const protocol = isGoogleApiModel(model) ? cfg?.googleProtocol : cfg?.vertexProtocol
481
526
  if (provider === PROVIDER && protocol === 'interactions' && this.interactionHandler) return interactionImageRequestTarget(image)
482
527
  return this.inner.imageRequestTarget(provider, model, image)
483
528
  }
@@ -488,13 +533,14 @@ export class ArgonAdapter extends LlmAdapter {
488
533
  resolveModel(provider, model, signal) { return this.inner.resolveModel(provider, model, signal) }
489
534
  async prepareCall(provider, model, signal) {
490
535
  const cfg = this.configSnapshot?.()
491
- if (provider === PROVIDER && cfg) resolveServicePolicy(cfg, (model === GOOGLE_API_MODEL || isProxyModel(model)) ? 'google-generative-ai' : 'google-vertex', (model === GOOGLE_API_MODEL || isProxyModel(model)) ? cfg.googleProtocol : cfg.vertexProtocol)
536
+ const googleApi = isGoogleApiModel(model)
537
+ const protocol = googleApi ? cfg?.googleProtocol : cfg?.vertexProtocol
538
+ if (provider === PROVIDER && cfg) resolveServicePolicy(cfg, googleApi ? 'google-generative-ai' : 'google-vertex', protocol)
492
539
  const prepared = await this.inner.prepareCall(provider, model, signal)
493
- const protocol = (model === GOOGLE_API_MODEL || isProxyModel(model)) ? cfg?.googleProtocol : cfg?.vertexProtocol
494
540
  if (provider === PROVIDER && protocol === 'interactions' && this.interactionHandler) {
495
- return { model: prepared.model, stream: options => this.interactionHandler(options, cfg) }
541
+ return { model: prepared.model, stream: options => this.interactionHandler(narrowMaxReasoning(options), cfg) }
496
542
  }
497
- return { ...prepared, stream: options => prepared.stream(generateContentHistory(options, model)) }
543
+ return { ...prepared, stream: options => prepared.stream(generateContentHistory(narrowMaxReasoning(options), model)) }
498
544
  }
499
545
  async *stream(options) {
500
546
  const prepared = await this.prepareCall(options.provider, options.model, options.signal)
@@ -554,12 +600,12 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
554
600
  vt: cfg.vertexServiceTier,
555
601
  paygo: cfg.vertexPaygoOnly,
556
602
  raw: cfg.retainRawResponses,
557
- pe: cfg.proxyEnabled,
558
- pb: cfg.proxyBaseUrl,
559
- pk: cfg.proxyApiKey,
560
- p: detected.hasProxy,
561
603
  v: detected.hasVertex,
562
604
  g: detected.hasGoogle,
605
+ va: detected.hasVertexArgon,
606
+ ga: detected.hasGoogleArgon,
607
+ vc: detected.hasVertexCarbon,
608
+ gc: detected.hasGoogleCarbon,
563
609
  ge: detected.googleApiKeyEnv,
564
610
  gb: detected.googleBaseUrl,
565
611
  vb: detected.vertexBaseUrl,
@@ -579,21 +625,19 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
579
625
  (await input.ctx.env(detected.googleApiKeyEnv || 'GOOGLE_API_KEY')) ??
580
626
  (await input.ctx.env('GOOGLE_API_KEY')) ??
581
627
  (await input.ctx.env('GEMINI_API_KEY'))
582
- const proxyKey = cfg.proxyApiKey || await input.ctx.env(PROXY_API_KEY_ENV)
583
628
  if (googleKey) env.__GOOGLE_API_KEY__ = googleKey
584
- if (proxyKey) env.__ANTIGRAVITY_API_KEY__ = proxyKey
585
629
  if (vertexRes) {
586
630
  return { ...vertexRes, env: { ...env, ...vertexRes.env } }
587
631
  }
588
- if (googleKey || proxyKey) {
589
- return { auth: {}, env, source: googleKey ? (detected.googleApiKeyEnv || 'GOOGLE_API_KEY') : PROXY_API_KEY_ENV }
632
+ if (googleKey) {
633
+ return { auth: {}, env, source: detected.googleApiKeyEnv || 'GOOGLE_API_KEY' }
590
634
  }
591
635
  return undefined
592
636
  },
593
637
  },
594
638
  }
595
639
  const models = []
596
- if (detected.hasVertex) {
640
+ if (detected.hasVertexArgon ?? detected.hasVertex) {
597
641
  models.push({
598
642
  id: MODEL,
599
643
  name: detected.hasGoogle ? 'Gemini 4 Argon (Vertex AI)' : 'Gemini 4 Argon',
@@ -608,13 +652,13 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
608
652
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
609
653
  })
610
654
  }
611
- if (detected.hasGoogle) {
655
+ if (detected.hasVertexCarbon) {
612
656
  models.push({
613
- id: GOOGLE_API_MODEL,
614
- name: 'Gemini 4 Argon (Gemini API)',
657
+ id: CARBON_MODEL,
658
+ name: detected.hasGoogle ? 'Gemini 4 Carbon (Vertex AI)' : 'Gemini 4 Carbon',
615
659
  provider: PROVIDER,
616
- api: 'google-generative-ai',
617
- baseUrl: detected.googleBaseUrl || 'https://generativelanguage.googleapis.com/v1beta',
660
+ api: 'google-vertex',
661
+ baseUrl: detected.vertexBaseUrl || 'https://{location}-aiplatform.googleapis.com',
618
662
  reasoning: true,
619
663
  input: ['text', 'image'],
620
664
  contextWindow: cfg.contextWindow,
@@ -623,13 +667,13 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
623
667
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
624
668
  })
625
669
  }
626
- if (cfg.proxyEnabled || detected.hasProxy) {
670
+ if (detected.hasGoogleArgon ?? detected.hasGoogle) {
627
671
  models.push({
628
- id: PROXY_MODEL,
629
- name: PROXY_MODEL_NAMES[PROXY_MODEL],
672
+ id: GOOGLE_API_MODEL,
673
+ name: 'Gemini 4 Argon (Gemini API)',
630
674
  provider: PROVIDER,
631
675
  api: 'google-generative-ai',
632
- baseUrl: cfg.proxyBaseUrl,
676
+ baseUrl: detected.googleBaseUrl || 'https://generativelanguage.googleapis.com/v1beta',
633
677
  reasoning: true,
634
678
  input: ['text', 'image'],
635
679
  contextWindow: cfg.contextWindow,
@@ -637,12 +681,14 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
637
681
  thinkingLevelMap: { ...DEFAULT_THINKING_MAP },
638
682
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
639
683
  })
684
+ }
685
+ if (detected.hasGoogleCarbon) {
640
686
  models.push({
641
- id: PROXY_FLASH_MODEL,
642
- name: PROXY_MODEL_NAMES[PROXY_FLASH_MODEL],
687
+ id: GOOGLE_API_CARBON_MODEL,
688
+ name: 'Gemini 4 Carbon (Gemini API)',
643
689
  provider: PROVIDER,
644
690
  api: 'google-generative-ai',
645
- baseUrl: cfg.proxyBaseUrl,
691
+ baseUrl: detected.googleBaseUrl || 'https://generativelanguage.googleapis.com/v1beta',
646
692
  reasoning: true,
647
693
  input: ['text', 'image'],
648
694
  contextWindow: cfg.contextWindow,
@@ -656,10 +702,10 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
656
702
  cachedKey = key
657
703
  cachedProfiles = new Map([[PROVIDER, {
658
704
  provider: PROVIDER,
659
- displayName: 'Vertex Argon',
705
+ displayName: 'Gemini 4 Adapter',
660
706
  piProvider: provider,
661
707
  reasoning: cfg.reasoning,
662
- streamIdleTimeoutMs: cfg.longDecode ? STREAM_READ_TIMEOUT_MS + 60000 : 300000,
708
+ streamIdleTimeoutMs: cfg.longDecode ? STREAM_READ_TIMEOUT_MS + 60000 : STREAM_READ_TIMEOUT_MS,
663
709
  maxRequestImageBytes: 20 * 1024 * 1024,
664
710
  requestImagePixelBudget: 2048 * 2048,
665
711
  requestImageMaxBytes: 1024 * 1024,