@opengeni/worker-bundle 0.15.1 → 0.16.0

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@opengeni/worker-bundle",
3
- "version": "0.15.1",
3
+ "version": "0.16.0",
4
4
  "description": "OpenGeni worker entry and reusable embedded lifecycle, shipped with a release-coherent pre-bundled Temporal workflow artifact.",
5
5
  "license": "Apache-2.0",
6
6
  "repository": {
@@ -40,17 +40,17 @@
40
40
  },
41
41
  "dependencies": {
42
42
  "@opengeni/agent-proto": "^0.3.0",
43
- "@opengeni/codex": "^0.2.9",
44
- "@opengeni/config": "^0.10.1",
45
- "@opengeni/contracts": "^0.31.1",
46
- "@opengeni/core": "^0.18.1",
47
- "@opengeni/db": "^0.22.2",
48
- "@opengeni/documents": "^0.2.70",
49
- "@opengeni/events": "^0.3.60",
50
- "@opengeni/github": "^0.4.17",
51
- "@opengeni/observability": "^0.4.4",
52
- "@opengeni/runtime": "^0.17.1",
53
- "@opengeni/storage": "^0.2.54",
43
+ "@opengeni/codex": "^0.2.10",
44
+ "@opengeni/config": "^0.10.3",
45
+ "@opengeni/contracts": "^0.32.0",
46
+ "@opengeni/core": "^0.19.0",
47
+ "@opengeni/db": "^0.23.0",
48
+ "@opengeni/documents": "^0.2.72",
49
+ "@opengeni/events": "^0.3.62",
50
+ "@opengeni/github": "^0.4.19",
51
+ "@opengeni/observability": "^0.4.6",
52
+ "@opengeni/runtime": "^0.18.0",
53
+ "@opengeni/storage": "^0.2.56",
54
54
  "@temporalio/activity": "^1.17.0",
55
55
  "@temporalio/client": "^1.17.0",
56
56
  "@temporalio/worker": "^1.17.0",
@@ -70,6 +70,9 @@ import {
70
70
  getEnrollment,
71
71
  abandonRecordingForTurnAttempt,
72
72
  markSessionAttemptQuiesced,
73
+ getOrCreatePreferenceRegistrySnapshot,
74
+ getOrCreateWorkspaceInstructionPolicySnapshot,
75
+ PreferenceRegistryInitiatorError,
73
76
  beginConnectorActionExecution,
74
77
  completeConnectorActionExecution,
75
78
  prepareConnectorActionApproval,
@@ -93,8 +96,11 @@ import {
93
96
  sanitizeHistoryItemsForModel,
94
97
  appendPersistentSessionSettings,
95
98
  appendSessionInstructions,
99
+ appendWorkspaceGovernance,
96
100
  appendWorkspaceMemory,
97
101
  composeAgentInstructions,
102
+ hasActiveWorkspaceInstructionPolicy,
103
+ renderWorkspaceGovernanceContext,
98
104
  summarizeForCompaction,
99
105
  requestRemoteCompactionV2,
100
106
  serializedToolsForRemoteCompaction,
@@ -142,6 +148,7 @@ import {
142
148
  import { connectionTokenResolverForTurn } from "./mcp-credentials";
143
149
  import {
144
150
  assertTurnExecutionPolicyMatchesConfigV1,
151
+ calculateGatewayReportedCostMicros,
145
152
  calculateModelUsageCostMicros,
146
153
  configuredModelPricing,
147
154
  configuredStaticUsageLimits,
@@ -151,6 +158,8 @@ import {
151
158
  serviceTierForLatencyMode,
152
159
  settingsWithResolvedModelContext,
153
160
  resolveTurnExecutionPolicyV1,
161
+ OPENGENI_GATEWAY_MODELS,
162
+ OPENGENI_GATEWAY_PROVIDER_ID,
154
163
  type ModelUsageInput,
155
164
  type ModelProviderApi,
156
165
  type RegistryProviderKind,
@@ -201,6 +210,7 @@ import {
201
210
  codexRequestStorage,
202
211
  isCodexTransportError,
203
212
  withCodexRequestOverrides,
213
+ selectCodexCredentialId,
204
214
  type CodexRequestContext,
205
215
  type CodexUsageHeaderSnapshot,
206
216
  } from "@opengeni/codex";
@@ -499,14 +509,7 @@ export function selectCodexCredentialForTurn(args: {
499
509
  activeCredentialId: string | null;
500
510
  connectedIds: Set<string>;
501
511
  }): string | null {
502
- const { sessionPinnedCredentialId: pin, activeCredentialId: active, connectedIds } = args;
503
- if (pin && connectedIds.has(pin)) {
504
- return pin;
505
- }
506
- if (active && connectedIds.has(active)) {
507
- return active;
508
- }
509
- return null;
512
+ return selectCodexCredentialId(args);
510
513
  }
511
514
 
512
515
  export function filterUnmaterializedSandboxFileDownloads(
@@ -1186,6 +1189,8 @@ export async function processModelResponseUsageEvent(input: {
1186
1189
  externallyBilled: input.externallyBilled,
1187
1190
  usage: responseUsage.usage,
1188
1191
  normalizedUsage,
1192
+ gatewayManaged: input.provider === OPENGENI_GATEWAY_PROVIDER_ID,
1193
+ gatewayBilling: responseUsage.gatewayBilling,
1189
1194
  sourceKey,
1190
1195
  ...(input.latencyMode ? { latencyMode: input.latencyMode } : {}),
1191
1196
  observability: input.observability,
@@ -1312,6 +1317,8 @@ export async function processCompactionModelUsageEvent(input: {
1312
1317
  externallyBilled: input.externallyBilled,
1313
1318
  usage: input.usage.usage,
1314
1319
  normalizedUsage,
1320
+ gatewayManaged: input.provider === OPENGENI_GATEWAY_PROVIDER_ID,
1321
+ gatewayBilling: input.usage.gatewayBilling,
1315
1322
  sourceKey,
1316
1323
  observability: input.observability,
1317
1324
  });
@@ -4053,10 +4060,29 @@ export function createRunAgentTurnActivity(services: () => Promise<ActivityServi
4053
4060
  // deployment default). null means the workspace has no override, so the
4054
4061
  // runtime falls back to runSettings.agentInstructionsTemplate (the
4055
4062
  // deployment default, byte-identical to the historical preamble).
4056
- const workspaceAgentInstructions = await resolveWorkspaceAgentInstructions(
4057
- db,
4058
- input.workspaceId,
4059
- );
4063
+ const governanceClaims = {
4064
+ accountId: input.accountId,
4065
+ workspaceId: input.workspaceId,
4066
+ sessionId: input.sessionId,
4067
+ turnId: turn.id,
4068
+ attemptId: input.attemptId,
4069
+ executionGeneration: turn.executionGeneration,
4070
+ };
4071
+ const [workspaceAgentInstructions, instructionPolicySnapshot, preferenceSnapshot] =
4072
+ await Promise.all([
4073
+ resolveWorkspaceAgentInstructions(db, input.workspaceId),
4074
+ getOrCreateWorkspaceInstructionPolicySnapshot(db, governanceClaims),
4075
+ getOrCreatePreferenceRegistrySnapshot(db, governanceClaims).catch((error) => {
4076
+ if (error instanceof PreferenceRegistryInitiatorError) return null;
4077
+ throw error;
4078
+ }),
4079
+ ]);
4080
+ const workspaceGovernance = renderWorkspaceGovernanceContext({
4081
+ instructionPolicy: instructionPolicySnapshot,
4082
+ preferences: preferenceSnapshot,
4083
+ });
4084
+ const structuredWorkspacePolicyActive =
4085
+ hasActiveWorkspaceInstructionPolicy(instructionPolicySnapshot);
4060
4086
  const workspaceMemory = await resolveWorkspaceMemoryBlock(db, input.workspaceId);
4061
4087
  const baseRunSettings = {
4062
4088
  // IMAGE PRECEDENCE (M3): rig > pack > deployment. settingsWithRigImage runs
@@ -4431,19 +4457,26 @@ export function createRunAgentTurnActivity(services: () => Promise<ActivityServi
4431
4457
  const persistentSessionSettings = {
4432
4458
  titleIsSet: Boolean(session.title?.trim()),
4433
4459
  };
4434
- const compactionInstructions = appendPersistentSessionSettings(
4435
- appendSessionInstructions(
4436
- appendWorkspaceMemory(
4437
- composeAgentInstructions(
4438
- workspaceAgentInstructions ?? modelRunSettings.agentInstructionsTemplate,
4439
- undefined,
4440
- rigVersion && rigName ? { name: rigName, version: rigVersion.version } : undefined,
4460
+ const compactionInstructions = appendWorkspaceMemory(
4461
+ appendPersistentSessionSettings(
4462
+ appendSessionInstructions(
4463
+ appendWorkspaceGovernance(
4464
+ composeAgentInstructions(
4465
+ structuredWorkspacePolicyActive
4466
+ ? modelRunSettings.agentInstructionsTemplate
4467
+ : (workspaceAgentInstructions ?? modelRunSettings.agentInstructionsTemplate),
4468
+ undefined,
4469
+ rigVersion && rigName
4470
+ ? { name: rigName, version: rigVersion.version }
4471
+ : undefined,
4472
+ ),
4473
+ workspaceGovernance ?? undefined,
4441
4474
  ),
4442
- workspaceMemory ?? undefined,
4475
+ session.instructions ?? undefined,
4443
4476
  ),
4444
- session.instructions ?? undefined,
4477
+ persistentSessionSettings,
4445
4478
  ),
4446
- persistentSessionSettings,
4479
+ workspaceMemory ?? undefined,
4447
4480
  );
4448
4481
  const requested = await isSessionCompactionRequested(
4449
4482
  db,
@@ -5708,7 +5741,10 @@ export function createRunAgentTurnActivity(services: () => Promise<ActivityServi
5708
5741
  skillLibrarySelections: skillLibraryRuntime.skillLibrarySelections,
5709
5742
  }
5710
5743
  : {}),
5711
- ...(workspaceAgentInstructions ? { instructionsTemplate: workspaceAgentInstructions } : {}),
5744
+ ...(!structuredWorkspacePolicyActive && workspaceAgentInstructions
5745
+ ? { instructionsTemplate: workspaceAgentInstructions }
5746
+ : {}),
5747
+ ...(workspaceGovernance ? { workspaceGovernance } : {}),
5712
5748
  ...(workspaceMemory ? { workspaceMemory } : {}),
5713
5749
  // Per-session persona tier (session > workspace > deployment default).
5714
5750
  // Composed system-level AFTER the workspace persona so it refines it for
@@ -9211,6 +9247,7 @@ export type ModelUsageBillingRecord = {
9211
9247
  /** Same quantity written to usage_events.model.cost when present; else 0. */
9212
9248
  pricedCostMicros: number;
9213
9249
  normalizedUsage: ModelCallUsageNormalization;
9250
+ upstreamProvider?: string;
9214
9251
  };
9215
9252
 
9216
9253
  // Exported for unit testing the external-billing bypass; not part of the activity surface.
@@ -9225,6 +9262,8 @@ export async function recordModelUsageAndDebitCredits(
9225
9262
  turnAttemptId: string;
9226
9263
  model: string;
9227
9264
  externallyBilled: boolean;
9265
+ gatewayManaged?: boolean;
9266
+ gatewayBilling?: ModelResponseUsage["gatewayBilling"];
9228
9267
  usage?: ModelUsageInput | ModelCallUsageInput | null;
9229
9268
  normalizedUsage?: ModelCallUsageNormalization;
9230
9269
  sourceKey: string;
@@ -9240,6 +9279,20 @@ export async function recordModelUsageAndDebitCredits(
9240
9279
  const inputTokens = sanitizedUsage.inputTokens ?? 0;
9241
9280
  const outputTokens = sanitizedUsage.outputTokens ?? 0;
9242
9281
  const totalTokens = sanitizedUsage.totalTokens ?? 0;
9282
+ const gatewayBilling = input.gatewayManaged ? input.gatewayBilling : undefined;
9283
+ if (gatewayBilling) {
9284
+ const gatewayModel = Object.values(OPENGENI_GATEWAY_MODELS).find(
9285
+ (candidate) => candidate.productId === input.model,
9286
+ );
9287
+ if (
9288
+ !gatewayModel ||
9289
+ !(gatewayModel.providers as readonly string[]).includes(gatewayBilling.finalProvider)
9290
+ ) {
9291
+ throw new Error(
9292
+ `AI Gateway reported unapproved provider ${gatewayBilling.finalProvider} for ${input.model}`,
9293
+ );
9294
+ }
9295
+ }
9243
9296
  // An externally billed turn is paid outside OpenGeni, so it
9244
9297
  // consumes ZERO OpenGeni credits and must never feed an OpenGeni cap. A
9245
9298
  // codex/<slug> model has no entry in configuredModelPricing, so the normal path
@@ -9285,19 +9338,24 @@ export async function recordModelUsageAndDebitCredits(
9285
9338
  });
9286
9339
  }
9287
9340
  const shouldDebit = settings.billingMode === "stripe" || settings.usageLimitsMode === "managed";
9288
- if (!shouldDebit || totalTokens === 0) {
9341
+ if (!shouldDebit || (totalTokens === 0 && !gatewayBilling)) {
9289
9342
  return {
9290
9343
  billingPath: "opengeni_credits",
9291
9344
  pricedCostMicros: 0,
9292
9345
  normalizedUsage,
9346
+ ...(gatewayBilling ? { upstreamProvider: gatewayBilling.finalProvider } : {}),
9293
9347
  };
9294
9348
  }
9295
9349
  if (!configuredModelPricing(settings)[input.model]) {
9296
9350
  throw new Error(`Missing model pricing for ${input.model}`);
9297
9351
  }
9298
- const costMicros = calculateModelUsageCostMicros(settings, input.model, sanitizedUsage, {
9299
- latencyMode: input.latencyMode ?? "standard",
9300
- });
9352
+ const costMicros = gatewayBilling
9353
+ ? calculateGatewayReportedCostMicros(settings, input.model, gatewayBilling.inferenceCostUsd, {
9354
+ inputTokens,
9355
+ })
9356
+ : calculateModelUsageCostMicros(settings, input.model, sanitizedUsage, {
9357
+ latencyMode: input.latencyMode ?? "standard",
9358
+ });
9301
9359
  await recordUsageEvent(db, {
9302
9360
  accountId: input.accountId,
9303
9361
  workspaceId: input.workspaceId,
@@ -9333,6 +9391,7 @@ export async function recordModelUsageAndDebitCredits(
9333
9391
  // per-call debit record carries cache efficiency alongside the token
9334
9392
  // counts. 0 when the provider did not report cached tokens.
9335
9393
  cachedTokens: normalizedUsage.telemetry.cachedTokens ?? 0,
9394
+ ...(gatewayBilling ? { gatewayProvider: gatewayBilling.finalProvider } : {}),
9336
9395
  },
9337
9396
  });
9338
9397
  recordCreditMicros(input.observability, "usage", result.debitedMicros);
@@ -9341,6 +9400,7 @@ export async function recordModelUsageAndDebitCredits(
9341
9400
  billingPath: "opengeni_credits",
9342
9401
  pricedCostMicros: costMicros,
9343
9402
  normalizedUsage,
9403
+ ...(gatewayBilling ? { upstreamProvider: gatewayBilling.finalProvider } : {}),
9344
9404
  };
9345
9405
  }
9346
9406
 
@@ -9368,7 +9428,7 @@ export async function recordAuthoritativeModelCallFact(input: {
9368
9428
  turnId: input.turnId,
9369
9429
  turnAttemptId: input.turnAttemptId,
9370
9430
  sourceKey: input.sourceKey,
9371
- provider: input.provider,
9431
+ provider: input.billing.upstreamProvider ?? input.provider,
9372
9432
  providerApi: input.providerApi,
9373
9433
  model: input.model,
9374
9434
  billingPath: input.billing.billingPath,