@terreno/ai 57.20.0 → 57.22.0

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.
Files changed (201) hide show
  1. package/CHANGELOG.md +81 -0
  2. package/README.md +3 -3
  3. package/biome.jsonc +3 -2
  4. package/dist/harness/agentLoop.js +1 -0
  5. package/dist/harness/commit.d.ts +4 -3
  6. package/dist/harness/commit.js +23 -4
  7. package/dist/index.d.ts +6 -9
  8. package/dist/index.js +14 -25
  9. package/dist/observability/adminScreens.js +5 -1
  10. package/dist/observability/evaluate.d.ts +8 -1
  11. package/dist/observability/evaluate.js +10 -2
  12. package/dist/observability/judgeAgreement.d.ts +53 -0
  13. package/dist/observability/judgeAgreement.js +123 -0
  14. package/dist/observability/local/datasetStore.d.ts +5 -0
  15. package/dist/observability/local/datasetStore.js +7 -2
  16. package/dist/observability/local/evaluatorStore.d.ts +17 -0
  17. package/dist/observability/local/evaluatorStore.js +73 -2
  18. package/dist/observability/local/experimentRunner.d.ts +42 -8
  19. package/dist/observability/local/experimentRunner.js +390 -11
  20. package/dist/observability/local/humanAnswers.d.ts +57 -0
  21. package/dist/observability/local/humanAnswers.js +174 -0
  22. package/dist/observability/local/judgeRun.d.ts +56 -0
  23. package/dist/observability/local/judgeRun.js +111 -0
  24. package/dist/observability/local/localPlugin.js +4 -0
  25. package/dist/observability/local/models/obsDatasetItem.js +4 -0
  26. package/dist/observability/local/models/obsEvaluator.js +4 -0
  27. package/dist/observability/local/models/obsExperiment.js +28 -0
  28. package/dist/observability/local/models/obsExperimentItem.js +4 -0
  29. package/dist/observability/local/models/obsJudgeRun.d.ts +2 -0
  30. package/dist/observability/local/models/obsJudgeRun.js +60 -0
  31. package/dist/observability/local/models/obsReviewBatch.d.ts +2 -0
  32. package/dist/observability/local/models/obsReviewBatch.js +48 -0
  33. package/dist/observability/local/models/obsReviewItem.js +10 -2
  34. package/dist/observability/local/models/obsTrace.js +4 -0
  35. package/dist/observability/local/promptReferences.d.ts +31 -0
  36. package/dist/observability/local/promptReferences.js +114 -0
  37. package/dist/observability/local/promptStore.d.ts +55 -19
  38. package/dist/observability/local/promptStore.js +171 -7
  39. package/dist/observability/local/reviewBatchStore.d.ts +106 -0
  40. package/dist/observability/local/reviewBatchStore.js +517 -0
  41. package/dist/observability/local/reviewStore.d.ts +24 -3
  42. package/dist/observability/local/reviewStore.js +98 -22
  43. package/dist/observability/local/traceStore.d.ts +14 -1
  44. package/dist/observability/local/traceStore.js +144 -11
  45. package/dist/observability/observabilityApp.js +10 -1
  46. package/dist/observability/questionFingerprint.d.ts +7 -0
  47. package/dist/observability/questionFingerprint.js +23 -0
  48. package/dist/observability/routes/experiments.js +12 -1
  49. package/dist/observability/routes/review.js +3 -1
  50. package/dist/observability/routes/reviewBatches.d.ts +11 -0
  51. package/dist/observability/routes/reviewBatches.js +78 -0
  52. package/dist/observability/routes/traces.d.ts +1 -1
  53. package/dist/observability/routes/traces.js +34 -2
  54. package/dist/observability/types.d.ts +10 -1
  55. package/dist/service/aiService.js +4 -1
  56. package/dist/service/chatTurn.js +38 -31
  57. package/dist/types/harness.d.ts +10 -4
  58. package/dist/types/index.d.ts +8 -3
  59. package/dist/types/observability.d.ts +74 -2
  60. package/e2e/smoke.ts +227 -0
  61. package/package.json +14 -28
  62. package/src/aiApp.test.ts +22 -17
  63. package/src/harness/agentLoop.test.ts +0 -3
  64. package/src/harness/agentLoop.ts +2 -1
  65. package/src/harness/agentRecovery.test.ts +30 -30
  66. package/src/harness/approvals.test.ts +11 -9
  67. package/src/harness/commit.ts +24 -4
  68. package/src/harness/events.test.ts +10 -6
  69. package/src/harness/extensions.test.ts +6 -8
  70. package/src/harness/ownership.test.ts +20 -9
  71. package/src/harness/retries.test.ts +6 -3
  72. package/src/harness/routes/conversations.test.ts +2 -2
  73. package/src/harness/routes/tasks.test.ts +2 -2
  74. package/src/harness/runners/jobsRunner.test.ts +1 -1
  75. package/src/harness/subagent.test.ts +29 -4
  76. package/src/harness/waits.test.ts +6 -4
  77. package/src/index.test.ts +3 -20
  78. package/src/index.ts +23 -26
  79. package/src/isolated/gptStream.isolated.ts +25 -28
  80. package/src/isolated/lazyProjectModel.isolated.ts +2 -2
  81. package/src/models/gptHistory.test.ts +10 -6
  82. package/src/observability/adminScreens.test.ts +7 -0
  83. package/src/observability/adminScreens.ts +6 -1
  84. package/src/observability/evaluate.test.ts +12 -10
  85. package/src/observability/evaluate.ts +12 -2
  86. package/src/observability/judgeAgreement.test.ts +111 -0
  87. package/src/observability/judgeAgreement.ts +184 -0
  88. package/src/observability/local/datasetStore.ts +30 -11
  89. package/src/observability/local/evaluatorStore.ts +96 -2
  90. package/src/observability/local/experimentRunner.test.ts +9 -12
  91. package/src/observability/local/experimentRunner.ts +497 -17
  92. package/src/observability/local/humanAnswers.ts +233 -0
  93. package/src/observability/local/judgePipeline.test.ts +444 -0
  94. package/src/observability/local/judgeRun.test.ts +173 -0
  95. package/src/observability/local/judgeRun.ts +184 -0
  96. package/src/observability/local/localPlugin.ts +4 -0
  97. package/src/observability/local/models/obsDatasetItem.ts +5 -0
  98. package/src/observability/local/models/obsEvaluator.ts +4 -0
  99. package/src/observability/local/models/obsExperiment.ts +29 -0
  100. package/src/observability/local/models/obsExperimentItem.ts +4 -0
  101. package/src/observability/local/models/obsJudgeRun.ts +61 -0
  102. package/src/observability/local/models/obsReviewBatch.ts +52 -0
  103. package/src/observability/local/models/obsReviewItem.ts +11 -2
  104. package/src/observability/local/models/obsTrace.ts +5 -0
  105. package/src/observability/local/obsModels.test.ts +13 -0
  106. package/src/observability/local/promptReferences.test.ts +128 -0
  107. package/src/observability/local/promptReferences.ts +149 -0
  108. package/src/observability/local/promptStore.test.ts +295 -3
  109. package/src/observability/local/promptStore.ts +243 -32
  110. package/src/observability/local/reviewBatchStore.test.ts +251 -0
  111. package/src/observability/local/reviewBatchStore.ts +656 -0
  112. package/src/observability/local/reviewStore.ts +122 -25
  113. package/src/observability/local/traceStore.test.ts +139 -1
  114. package/src/observability/local/traceStore.ts +162 -11
  115. package/src/observability/observabilityApp.test.ts +13 -10
  116. package/src/observability/observabilityApp.ts +10 -1
  117. package/src/observability/questionFingerprint.test.ts +33 -0
  118. package/src/observability/questionFingerprint.ts +24 -0
  119. package/src/observability/routes/datasets.test.ts +2 -2
  120. package/src/observability/routes/evaluators.test.ts +2 -2
  121. package/src/observability/routes/experiments.test.ts +39 -2
  122. package/src/observability/routes/experiments.ts +13 -1
  123. package/src/observability/routes/prompts.test.ts +3 -3
  124. package/src/observability/routes/review.test.ts +2 -2
  125. package/src/observability/routes/review.ts +3 -1
  126. package/src/observability/routes/reviewBatches.test.ts +209 -0
  127. package/src/observability/routes/reviewBatches.ts +138 -0
  128. package/src/observability/routes/status.test.ts +8 -8
  129. package/src/observability/routes/testMultiStage.test.ts +6 -6
  130. package/src/observability/routes/traces.test.ts +63 -2
  131. package/src/observability/routes/traces.ts +46 -4
  132. package/src/observability/types.ts +5 -2
  133. package/src/routes/files.test.ts +5 -3
  134. package/src/routes/gpt.test.ts +21 -14
  135. package/src/routes/gptActions.test.ts +7 -8
  136. package/src/routes/gptDatasets.test.ts +1 -1
  137. package/src/routes/gptFiles.test.ts +1 -1
  138. package/src/routes/gptHistories.test.ts +15 -9
  139. package/src/routes/mcp.test.ts +3 -3
  140. package/src/routes/projects.test.ts +5 -3
  141. package/src/routes/routes.test.ts +156 -115
  142. package/src/service/aiService.test.ts +22 -1
  143. package/src/service/aiService.ts +4 -1
  144. package/src/service/asks.test.ts +14 -6
  145. package/src/service/chatTurn.ts +42 -33
  146. package/src/service/fileStorage.test.ts +5 -2
  147. package/src/service/getMCPTools.test.ts +9 -5
  148. package/src/service/mcpService.test.ts +6 -3
  149. package/src/tests/bunSetup.ts +0 -54
  150. package/src/tests/chatHarness.ts +4 -4
  151. package/src/tests/fakeJudge.ts +8 -0
  152. package/src/tests/helpers.ts +26 -4
  153. package/src/types/harness.ts +11 -4
  154. package/src/types/index.ts +8 -3
  155. package/src/types/observability.ts +86 -2
  156. package/tsconfig.typecheck.json +18 -0
  157. package/dist/langfuseApp.d.ts +0 -9
  158. package/dist/langfuseApp.js +0 -53
  159. package/dist/langfuseCache.d.ts +0 -15
  160. package/dist/langfuseCache.js +0 -54
  161. package/dist/langfuseClient.d.ts +0 -6
  162. package/dist/langfuseClient.js +0 -34
  163. package/dist/langfusePrompts.d.ts +0 -12
  164. package/dist/langfusePrompts.js +0 -89
  165. package/dist/langfuseRoutesEvaluations.d.ts +0 -3
  166. package/dist/langfuseRoutesEvaluations.js +0 -32
  167. package/dist/langfuseRoutesMiddleware.d.ts +0 -2
  168. package/dist/langfuseRoutesMiddleware.js +0 -13
  169. package/dist/langfuseRoutesPlayground.d.ts +0 -2
  170. package/dist/langfuseRoutesPlayground.js +0 -36
  171. package/dist/langfuseRoutesPrompts.d.ts +0 -2
  172. package/dist/langfuseRoutesPrompts.js +0 -56
  173. package/dist/langfuseRoutesTraces.d.ts +0 -2
  174. package/dist/langfuseRoutesTraces.js +0 -30
  175. package/dist/langfuseTracing.d.ts +0 -4
  176. package/dist/langfuseTracing.js +0 -32
  177. package/dist/langfuseTypes.d.ts +0 -96
  178. package/dist/langfuseTypes.js +0 -2
  179. package/dist/langfuseVercelAi.d.ts +0 -12
  180. package/dist/langfuseVercelAi.js +0 -49
  181. package/src/isolated/langfuseClient.isolated.ts +0 -60
  182. package/src/langfuseApp.test.ts +0 -135
  183. package/src/langfuseApp.ts +0 -62
  184. package/src/langfuseCache.test.ts +0 -75
  185. package/src/langfuseCache.ts +0 -72
  186. package/src/langfuseClient.test.ts +0 -69
  187. package/src/langfuseClient.ts +0 -34
  188. package/src/langfusePrompts.test.ts +0 -188
  189. package/src/langfusePrompts.ts +0 -121
  190. package/src/langfuseRoutes.test.ts +0 -256
  191. package/src/langfuseRoutesEvaluations.ts +0 -48
  192. package/src/langfuseRoutesMiddleware.test.ts +0 -57
  193. package/src/langfuseRoutesMiddleware.ts +0 -12
  194. package/src/langfuseRoutesPlayground.ts +0 -50
  195. package/src/langfuseRoutesPrompts.ts +0 -87
  196. package/src/langfuseRoutesTraces.ts +0 -40
  197. package/src/langfuseTracing.test.ts +0 -68
  198. package/src/langfuseTracing.ts +0 -33
  199. package/src/langfuseTypes.ts +0 -103
  200. package/src/langfuseVercelAi.test.ts +0 -79
  201. package/src/langfuseVercelAi.ts +0 -72
@@ -4,11 +4,14 @@ import {DateTime, Settings} from "luxon";
4
4
  import {createLocalObservabilityPlugin} from "../observability/local/localPlugin";
5
5
  import {registerObsSpan} from "../observability/local/models/obsSpan";
6
6
  import {registerObsTrace} from "../observability/local/models/obsTrace";
7
- import type {HarnessTaskDocument, HarnessTestHooks} from "../types/harness";
7
+ import type {HarnessTaskDefinition, HarnessTaskDocument, HarnessTestHooks} from "../types/harness";
8
8
  import type {ObsSpanModel, ObsTraceModel} from "../types/observability";
9
9
  import {type AnyHarnessTaskDefinition, defineTask, Harness, InProcessRunner} from "./harness";
10
10
  import {registerHarnessTask} from "./models/harnessTask";
11
11
 
12
+ /** A task definition whose input is not checked, so tests can create it with any input. */
13
+ type TestTaskDefinition = HarnessTaskDefinition<unknown, unknown, unknown>;
14
+
12
15
  const TaskModel = registerHarnessTask();
13
16
  let SpanModel: ObsSpanModel;
14
17
  let TraceModel: ObsTraceModel;
@@ -62,7 +65,7 @@ const flakyTask = ({
62
65
  failures: number;
63
66
  name: string;
64
67
  retry: {backoffMs?: number; maxAttempts?: number; maxBackoffMs?: number};
65
- }): {attemptsSeen: number[]; definition: AnyHarnessTaskDefinition} => {
68
+ }): {attemptsSeen: number[]; definition: TestTaskDefinition} => {
66
69
  const attemptsSeen: number[] = [];
67
70
  const definition = defineTask<unknown, unknown, {ok: boolean}>({
68
71
  initial: () => ({phase: "call"}),
@@ -81,7 +84,7 @@ const flakyTask = ({
81
84
  retry,
82
85
  version: 1,
83
86
  });
84
- return {attemptsSeen, definition: definition as unknown as AnyHarnessTaskDefinition};
87
+ return {attemptsSeen, definition: definition as unknown as TestTaskDefinition};
85
88
  };
86
89
 
87
90
  describe("Harness phase retries", () => {
@@ -10,7 +10,7 @@ import {createLocalObservabilityPlugin} from "../../observability/local/localPlu
10
10
  import {registerObsSpan} from "../../observability/local/models/obsSpan";
11
11
  import {registerObsTrace} from "../../observability/local/models/obsTrace";
12
12
  import {gatedStreamingModel} from "../../tests/harnessStreaming";
13
- import {ensureTestUsers, UserModel} from "../../tests/helpers";
13
+ import {ensureTestUsers, terrenoUserModel, UserModel} from "../../tests/helpers";
14
14
  import type {ObsSpanModel, ObsTraceModel} from "../../types/observability";
15
15
  import {defineAgent, defineTool, Harness, HarnessApp, InProcessRunner} from "../harness";
16
16
  import {registerHarnessConversation} from "../models/harnessConversation";
@@ -88,7 +88,7 @@ const setup = async () => {
88
88
  });
89
89
  openHarnesses.push(harness);
90
90
  await harness.start();
91
- const app: express.Application = new TerrenoApp({skipListen: true, userModel: UserModel})
91
+ const app: express.Application = new TerrenoApp({skipListen: true, userModel: terrenoUserModel})
92
92
  .register(new HarnessApp({harness}))
93
93
  .build();
94
94
  const login = async ({email, password}: {email: string; password: string}) => {
@@ -8,7 +8,7 @@ import type TestAgent from "supertest/lib/agent";
8
8
  import {createLocalObservabilityPlugin} from "../../observability/local/localPlugin";
9
9
  import {registerObsSpan} from "../../observability/local/models/obsSpan";
10
10
  import {registerObsTrace} from "../../observability/local/models/obsTrace";
11
- import {ensureTestUsers, UserModel} from "../../tests/helpers";
11
+ import {ensureTestUsers, terrenoUserModel, UserModel} from "../../tests/helpers";
12
12
  import type {ObsSpanModel, ObsTraceModel} from "../../types/observability";
13
13
  import {defineTask, Harness, HarnessApp, InProcessRunner} from "../harness";
14
14
  import {registerHarnessEvent, registerHarnessEventStream} from "../models/harnessEvent";
@@ -55,7 +55,7 @@ const setup = async ({start = false}: {start?: boolean} = {}) => {
55
55
  if (start) {
56
56
  await harness.start();
57
57
  }
58
- const app: express.Application = new TerrenoApp({skipListen: true, userModel: UserModel})
58
+ const app: express.Application = new TerrenoApp({skipListen: true, userModel: terrenoUserModel})
59
59
  .register(new HarnessApp({harness}))
60
60
  .build();
61
61
  const login = async ({email, password}: {email: string; password: string}) => {
@@ -676,7 +676,7 @@ describe("JobsRunner", () => {
676
676
  });
677
677
 
678
678
  it("is not reachable from the root @terreno/ai or @terreno/ai/harness imports", () => {
679
- const srcRoot = resolve(import.meta.dir, "../..");
679
+ const srcRoot = resolve(__dirname, "../..");
680
680
  const seen = new Set<string>();
681
681
  const importsJobs: string[] = [];
682
682
  const visit = (file: string): void => {
@@ -10,6 +10,7 @@ import {
10
10
  } from "bun:test";
11
11
  import {z} from "@terreno/api";
12
12
  import type {LanguageModel} from "ai";
13
+ import type mongoose from "mongoose";
13
14
  import {createLocalObservabilityPlugin} from "../observability/local/localPlugin";
14
15
  import {registerObsSpan} from "../observability/local/models/obsSpan";
15
16
  import {registerObsTrace} from "../observability/local/models/obsTrace";
@@ -246,9 +247,6 @@ const parentTask = ({
246
247
 
247
248
  describe("rt.runAgent (subagents)", () => {
248
249
  beforeAll(async () => {
249
- // The shared preload's beforeEach imports this lazily under a 5 s hook timeout; a
250
- // cold first import can exceed it, so load it here first.
251
- await import("../langfuseClient");
252
250
  createLocalObservabilityPlugin();
253
251
  SpanModel = registerObsSpan();
254
252
  TraceModel = registerObsTrace();
@@ -377,6 +375,16 @@ describe("rt.runAgent (subagents)", () => {
377
375
  ).rejects.toThrow(
378
376
  harnessErrorMatching("invalidRequest", 'Prompt ref "plan" label must be a string')
379
377
  );
378
+ await expect(
379
+ harness.createTask(parent, {}, {prompts: [{includes: {}, name: "plan", version: 1} as never]})
380
+ ).rejects.toThrow(
381
+ harnessErrorMatching("invalidRequest", 'Prompt ref "plan" includes must be an array')
382
+ );
383
+ await expect(
384
+ harness.createTask(parent, {}, {prompts: [{includes: {}, version: 1} as never]})
385
+ ).rejects.toThrow(
386
+ harnessErrorMatching("invalidRequest", "Prompt ref includes must be an array")
387
+ );
380
388
  await expect(harness.createTask(parent, {}, {trace: {tags: ["ok", " "]}})).rejects.toThrow(
381
389
  harnessErrorMatching("invalidRequest", "trace.tags must be an array of non-empty strings")
382
390
  );
@@ -396,6 +404,23 @@ describe("rt.runAgent (subagents)", () => {
396
404
  expect(plain(trace.prompts)).toEqual([plan]);
397
405
  });
398
406
 
407
+ it("records a registry prompt's shared includes after it", async () => {
408
+ const agent = summarizerWith();
409
+ const parent = parentTask({agent, inputs: []});
410
+ const {model} = scriptedModel([]);
411
+ const {harness} = await openProcess({model, registry: [agent, parent]});
412
+
413
+ const include = {includedBy: "plan", label: "production", name: "base-rules", version: 4};
414
+ const created = await harness.createTask(
415
+ parent,
416
+ {},
417
+ {prompts: [{includes: [include], name: "plan", version: 2}]}
418
+ );
419
+
420
+ const trace = await TraceModel.findExactlyOne({_id: created.traceId});
421
+ expect(plain(trace.prompts)).toEqual([{name: "plan", version: 2}, include]);
422
+ });
423
+
399
424
  it("fails the caller at once for an instructions prompt without a body or a bad prompt ref", async () => {
400
425
  const agent = summarizerWith();
401
426
  const noBody = parentTask({
@@ -653,7 +678,7 @@ describe("rt.runAgent (subagents)", () => {
653
678
  expect(await TaskModel.countDocuments({name: AGENT_TURN_TASK_NAME})).toBe(2);
654
679
  // The first child is found again, not re-run, when the phase resumes for the second.
655
680
  expect(calls).toHaveLength(2);
656
- const firstMessages = async (conversationId: unknown) =>
681
+ const firstMessages = async (conversationId: mongoose.Types.ObjectId | string) =>
657
682
  (await MessageModel.find({conversationId}).sort({seq: 1})).map(
658
683
  ({parts}) => (parts[0] as {text: string}).text
659
684
  );
@@ -6,7 +6,7 @@ import {createLocalObservabilityPlugin} from "../observability/local/localPlugin
6
6
  import {registerObsSpan} from "../observability/local/models/obsSpan";
7
7
  import {registerObsTrace} from "../observability/local/models/obsTrace";
8
8
  import {harnessErrorMatching} from "../tests/harnessErrors";
9
- import type {HarnessTaskDocument, HarnessTestHooks} from "../types/harness";
9
+ import type {HarnessTaskDefinition, HarnessTaskDocument, HarnessTestHooks} from "../types/harness";
10
10
  import type {ObsSpanModel, ObsTraceModel} from "../types/observability";
11
11
  import {commitWaiting, type HarnessModels} from "./commit";
12
12
  import {errorMessage} from "./errors";
@@ -16,6 +16,9 @@ import {registerHarnessInboxEvent} from "./models/harnessInboxEvent";
16
16
  import {registerHarnessOwner} from "./models/harnessOwner";
17
17
  import {registerHarnessTask} from "./models/harnessTask";
18
18
 
19
+ /** A task definition whose input is not checked, so tests can create it with any input. */
20
+ type TestTaskDefinition = HarnessTaskDefinition<unknown, unknown, unknown>;
21
+
19
22
  const TaskModel = registerHarnessTask();
20
23
  const OwnerModel = registerHarnessOwner();
21
24
  const InboxModel = registerHarnessInboxEvent();
@@ -58,8 +61,7 @@ const findTask = (id: unknown): Promise<HarnessTaskDocument> => TaskModel.findEx
58
61
  const statusIs = (id: unknown, status: string) => async (): Promise<boolean> =>
59
62
  (await findTask(id)).status === status;
60
63
 
61
- const asAny = (definition: unknown): AnyHarnessTaskDefinition =>
62
- definition as AnyHarnessTaskDefinition;
64
+ const asAny = (definition: unknown): TestTaskDefinition => definition as TestTaskDefinition;
63
65
 
64
66
  /** One phase that waits on `event` (with `timeout`) and completes with what it got. */
65
67
  const waitsOnce = ({
@@ -72,7 +74,7 @@ const waitsOnce = ({
72
74
  name: string;
73
75
  runs: {count: number};
74
76
  timeout?: {minutes: number};
75
- }): AnyHarnessTaskDefinition =>
77
+ }): TestTaskDefinition =>
76
78
  asAny(
77
79
  defineTask<unknown, unknown, unknown>({
78
80
  initial: () => ({phase: "await"}),
package/src/index.test.ts CHANGED
@@ -18,21 +18,10 @@ describe("@terreno/ai public exports", () => {
18
18
  "addMcpRoutes",
19
19
  "addProjectRoutes",
20
20
  "assertVertexModelsEnabled",
21
- "compilePrompt",
22
21
  "createAskTools",
23
- "createPrompt",
24
- "createTelemetryConfig",
25
22
  "createVertexProvider",
26
23
  "findAgentBlock",
27
- "getCached",
28
24
  "getMCPTools",
29
- "getLangfuseClient",
30
- "getPrompt",
31
- "initLangfuseClient",
32
- "initTracing",
33
- "invalidateCache",
34
- "invalidatePromptCache",
35
- "isLangfuseInitialized",
36
25
  "isVertexModelAllowed",
37
26
  "jsonSchema",
38
27
  "listEnabledVertexModels",
@@ -41,17 +30,13 @@ describe("@terreno/ai public exports", () => {
41
30
  "normalizeLlmJsonTextForStructuredOutput",
42
31
  "normalizeVertexModelId",
43
32
  "parseAiJson",
44
- "preparePromptForAI",
45
33
  "getObservabilityApp",
46
34
  "resetObservabilityApp",
47
35
  "resolveObservabilityControl",
48
36
  "runBufferedChatTurn",
49
37
  "validateObservabilityConfig",
50
- "setCached",
51
- "shutdownLangfuseClient",
52
- "shutdownTracing",
53
38
  "verifyVertexModelsEnabled",
54
- ] as const;
39
+ ];
55
40
 
56
41
  const expectedClasses = [
57
42
  "AIRequest",
@@ -60,14 +45,12 @@ describe("@terreno/ai public exports", () => {
60
45
  "FileAttachment",
61
46
  "FileStorageService",
62
47
  "GptHistory",
63
- "LangfuseApp",
64
- "LangfuseCache",
65
48
  "MCPService",
66
49
  "MemoryScoreSink",
67
50
  "MemoryTraceSink",
68
51
  "ObservabilityApp",
69
52
  "Project",
70
- ] as const;
53
+ ];
71
54
 
72
55
  const expectedConstants = [
73
56
  "COMPACT_SURFACE_SYSTEM_PROMPT",
@@ -85,7 +68,7 @@ describe("@terreno/ai public exports", () => {
85
68
  "toggleChecklistHostAction",
86
69
  "TRANSLATION_PROMPT",
87
70
  "TemperaturePresets",
88
- ] as const;
71
+ ];
89
72
 
90
73
  it.each(expectedFunctions)("exports %s as a function", (name) => {
91
74
  expect(typeof exported[name]).toBe("function");
package/src/index.ts CHANGED
@@ -12,44 +12,40 @@ export type {AIAdminAppOptions} from "./aiAdminApp";
12
12
  export {AIAdminApp} from "./aiAdminApp";
13
13
  export type {AiAppOptions} from "./aiApp";
14
14
  export {AiApp} from "./aiApp";
15
- export {LangfuseApp} from "./langfuseApp";
16
- export {getCached, invalidateCache, LangfuseCache, setCached} from "./langfuseCache";
17
- export {
18
- getLangfuseClient,
19
- initLangfuseClient,
20
- isLangfuseInitialized,
21
- shutdownLangfuseClient,
22
- } from "./langfuseClient";
23
- export {compilePrompt, createPrompt, getPrompt, invalidatePromptCache} from "./langfusePrompts";
24
- export {initTracing, shutdownTracing} from "./langfuseTracing";
25
- export type {
26
- ChatMessage,
27
- GetPromptOptions,
28
- LangfuseAppOptions,
29
- LangfuseCachedPrompt,
30
- PaginatedResult,
31
- PreparePromptResult,
32
- PromptListItem,
33
- ScoreSubmission,
34
- ScoringFunction,
35
- TelemetrySettings,
36
- TraceListItem,
37
- } from "./langfuseTypes";
38
- export {createTelemetryConfig, preparePromptForAI} from "./langfuseVercelAi";
39
15
  export {AIDataset} from "./models/aiDataset";
40
16
  export {AIRequest} from "./models/aiRequest";
41
17
  export {FileAttachment} from "./models/fileAttachment";
42
18
  export {GptHistory} from "./models/gptHistory";
43
19
  export {getProjectModel, Project} from "./models/project";
44
20
  export {AI_OBSERVABILITY_GROUP, observabilityAdminScreens} from "./observability/adminScreens";
21
+ export {
22
+ answersAgree,
23
+ computeJudgeAgreement,
24
+ type JudgeAgreement,
25
+ } from "./observability/judgeAgreement";
45
26
  export {LocalDatasetStore} from "./observability/local/datasetStore";
46
- export {LocalEvaluatorStore} from "./observability/local/evaluatorStore";
47
- export {LocalExperimentRunner} from "./observability/local/experimentRunner";
27
+ export {type EvaluatorCalibration, LocalEvaluatorStore} from "./observability/local/evaluatorStore";
28
+ export {
29
+ DEFAULT_TARGET_AGREEMENT,
30
+ type JudgeExperimentCreateInput,
31
+ LocalExperimentRunner,
32
+ } from "./observability/local/experimentRunner";
33
+ export {
34
+ type DatasetJudgmentSummary,
35
+ loadCurrentHumanAnswers,
36
+ } from "./observability/local/humanAnswers";
48
37
  export {
49
38
  createLocalObservabilityBundle,
50
39
  createLocalObservabilityPlugin,
51
40
  } from "./observability/local/localPlugin";
52
41
  export {LocalPromptStore} from "./observability/local/promptStore";
42
+ export {
43
+ DEFAULT_REVIEW_BATCH_SIZE,
44
+ LocalReviewBatchStore,
45
+ MAX_REVIEW_BATCH_SIZE,
46
+ type ReviewBatchDetail,
47
+ type ReviewBatchListItem,
48
+ } from "./observability/local/reviewBatchStore";
53
49
  export {LocalReviewStore} from "./observability/local/reviewStore";
54
50
  export {
55
51
  LocalScoreSink,
@@ -63,6 +59,7 @@ export {
63
59
  ObservabilityApp,
64
60
  resetObservabilityApp,
65
61
  } from "./observability/observabilityApp";
62
+ export {questionSetFingerprint} from "./observability/questionFingerprint";
66
63
  export type {
67
64
  ObservabilityPluginStatus,
68
65
  ObservabilityStatus,
@@ -4,7 +4,7 @@ import type {LanguageModel} from "ai";
4
4
  import type express from "express";
5
5
 
6
6
  import {GptHistory} from "../models/gptHistory";
7
- import {authAsUser, ensureTestUsers, UserModel} from "../tests/helpers";
7
+ import {authAsUser, ensureTestUsers, terrenoUserModel} from "../tests/helpers";
8
8
 
9
9
  type StreamPart = {type: string; [key: string]: unknown};
10
10
 
@@ -13,32 +13,29 @@ type StreamPart = {type: string; [key: string]: unknown};
13
13
  let streamParts: StreamPart[] = [];
14
14
  let streamFiles: Array<{base64: string; mediaType: string}> = [];
15
15
 
16
- // Mock langfuseVercelAi to avoid transitive langfuse SDK imports.
17
- const realLangfuseVercelAi = await import("../langfuseVercelAi");
18
- mock.module("../langfuseVercelAi", () => ({
19
- ...realLangfuseVercelAi,
20
- createTelemetryConfig: () => ({functionId: "test", isEnabled: false}),
21
- preparePromptForAI: async () => ({config: {}, prompt: "test", telemetry: {isEnabled: false}}),
22
- }));
23
-
24
- // Mock the AI SDK so we control exactly which fullStream parts the GPT prompt
25
- // handler iterates over. All other exports are preserved so AIService and the
26
- // route helpers keep working.
27
- const realAi = await import("ai");
28
- mock.module("ai", () => ({
29
- ...realAi,
30
- streamText: () => ({
31
- files: Promise.resolve(streamFiles),
32
- fullStream: (async function* fullStream(): AsyncGenerator<StreamPart> {
33
- for (const part of streamParts) {
34
- yield part;
35
- }
36
- })(),
37
- }),
38
- }));
39
-
40
- const {addGptRoutes} = await import("../routes/gpt");
41
- const {AIService} = await import("../service/aiService");
16
+ let addGptRoutes: typeof import("../routes/gpt.js").addGptRoutes;
17
+ let AIService: typeof import("../service/aiService.js").AIService;
18
+
19
+ beforeAll(async () => {
20
+ // Mock the AI SDK so we control exactly which fullStream parts the GPT prompt
21
+ // handler iterates over. All other exports are preserved so AIService and the
22
+ // route helpers keep working.
23
+ const realAi = await import("ai");
24
+ mock.module("ai", () => ({
25
+ ...realAi,
26
+ streamText: () => ({
27
+ files: Promise.resolve(streamFiles),
28
+ fullStream: (async function* fullStream(): AsyncGenerator<StreamPart> {
29
+ for (const part of streamParts) {
30
+ yield part;
31
+ }
32
+ })(),
33
+ }),
34
+ }));
35
+
36
+ ({addGptRoutes} = await import("../routes/gpt.js"));
37
+ ({AIService} = await import("../service/aiService.js"));
38
+ });
42
39
 
43
40
  const createMockModel = () => ({
44
41
  doGenerate: mock(async () => ({
@@ -87,7 +84,7 @@ describe("GPT prompt streaming edge cases", () => {
87
84
  addGptRoutes(router, {aiService, openApiOptions: options});
88
85
  },
89
86
  skipListen: true,
90
- userModel: UserModel,
87
+ userModel: terrenoUserModel,
91
88
  }).build();
92
89
  });
93
90
 
@@ -3,7 +3,7 @@ import mongoose from "mongoose";
3
3
 
4
4
  describe("@terreno/ai root entry", () => {
5
5
  it("does not register the Project model until it is used", async () => {
6
- const ai = await import("../index");
6
+ const ai = await import("../index.js");
7
7
  expect(mongoose.models.Project).toBeUndefined();
8
8
 
9
9
  expect(ai.Project.modelName).toBe("Project");
@@ -11,7 +11,7 @@ describe("@terreno/ai root entry", () => {
11
11
  });
12
12
 
13
13
  it("forwards statics, construction and instanceof through the deprecated Project export", async () => {
14
- const {Project} = await import("../index");
14
+ const {Project} = await import("../index.js");
15
15
  const userId = new mongoose.Types.ObjectId();
16
16
 
17
17
  const project = new Project({name: "Roadmap", userId});
@@ -1,6 +1,7 @@
1
1
  import {afterEach, beforeEach, describe, expect, it} from "bun:test";
2
2
  import mongoose from "mongoose";
3
3
 
4
+ import type {GptHistoryDocument, GptHistoryPendingAsk, GptHistoryPrompt} from "../types/index";
4
5
  import {GptHistory} from "./gptHistory";
5
6
 
6
7
  describe("GptHistory Model", () => {
@@ -217,9 +218,9 @@ describe("GptHistory Model", () => {
217
218
  {id: "team", label: "Team"},
218
219
  ],
219
220
  prompt: "Which plan should I set up?",
220
- select: "one",
221
+ select: "one" as const,
221
222
  };
222
- const pendingAsk = {
223
+ const pendingAsk: GptHistoryPendingAsk = {
223
224
  created: new Date("2026-09-27T12:00:00.000Z"),
224
225
  input: planAsk,
225
226
  kind: "choice",
@@ -256,7 +257,10 @@ describe("GptHistory Model", () => {
256
257
  },
257
258
  toolCallId: "call_plan",
258
259
  };
259
- const askRow = {
260
+ /** Create from deliberately invalid input, so the schema rather than the compiler rejects it. */
261
+ const createUnchecked = (doc: Record<string, unknown>) =>
262
+ GptHistory.create(doc as Partial<GptHistoryDocument>);
263
+ const askRow: GptHistoryPrompt = {
260
264
  args: planAsk,
261
265
  ask: {kind: "choice", status: "pending"},
262
266
  text: "Tool call: ask_choice",
@@ -309,7 +313,7 @@ describe("GptHistory Model", () => {
309
313
  const userId = new mongoose.Types.ObjectId();
310
314
  const history = await GptHistory.create({prompts: [], userId});
311
315
 
312
- history.pendingAsk = pendingAsk as unknown as NonNullable<typeof history.pendingAsk>;
316
+ history.pendingAsk = pendingAsk;
313
317
  await history.save();
314
318
 
315
319
  const stored = await GptHistory.collection.findOne({_id: history._id});
@@ -334,7 +338,7 @@ describe("GptHistory Model", () => {
334
338
  },
335
339
  ])("rejects an unknown key in $field", async ({expected, history}) => {
336
340
  await expect(
337
- GptHistory.create({...history, userId: new mongoose.Types.ObjectId()})
341
+ createUnchecked({...history, userId: new mongoose.Types.ObjectId()})
338
342
  ).rejects.toThrow(expected);
339
343
  });
340
344
 
@@ -361,7 +365,7 @@ describe("GptHistory Model", () => {
361
365
  },
362
366
  ])("rejects $label", async ({expected, history}) => {
363
367
  await expect(
364
- GptHistory.create({...history, userId: new mongoose.Types.ObjectId()})
368
+ createUnchecked({...history, userId: new mongoose.Types.ObjectId()})
365
369
  ).rejects.toThrow(expected);
366
370
  });
367
371
  });
@@ -24,6 +24,7 @@ describe("observabilityAdminScreens", () => {
24
24
  "ai-datasets",
25
25
  "ai-experiments",
26
26
  "ai-review",
27
+ "ai-review-batches",
27
28
  ]);
28
29
  });
29
30
 
@@ -66,6 +67,12 @@ describe("observabilityAdminScreens", () => {
66
67
  group: AI_OBSERVABILITY_GROUP,
67
68
  name: "ai-review",
68
69
  },
70
+ {
71
+ adminAccess: {action: "list", resource: "aiReview"},
72
+ displayName: "Review batches",
73
+ group: AI_OBSERVABILITY_GROUP,
74
+ name: "ai-review-batches",
75
+ },
69
76
  ]);
70
77
  });
71
78
  });
@@ -35,7 +35,12 @@ export const observabilityAdminScreens = ({localOn}: {localOn: boolean}): AdminC
35
35
  ];
36
36
  if (localOn) {
37
37
  screens.push(
38
- observabilityScreen({displayName: "Review queue", name: "ai-review", resource: "aiReview"})
38
+ observabilityScreen({displayName: "Review queue", name: "ai-review", resource: "aiReview"}),
39
+ observabilityScreen({
40
+ displayName: "Review batches",
41
+ name: "ai-review-batches",
42
+ resource: "aiReview",
43
+ })
39
44
  );
40
45
  }
41
46
  return screens;
@@ -1,5 +1,6 @@
1
1
  import {afterEach, beforeEach, describe, it} from "bun:test";
2
2
  import {assert} from "chai";
3
+ import {judgeAnswering} from "../tests/fakeJudge";
3
4
  import {
4
5
  judgeSchemaMissingDimensions,
5
6
  runEvaluator,
@@ -95,9 +96,7 @@ describe("evaluate", () => {
95
96
  output: "4",
96
97
  },
97
98
  {
98
- generateJsonObject: async () => {
99
- return {confidence: 0.95, correct: true};
100
- },
99
+ generateJsonObject: judgeAnswering(() => ({confidence: 0.95, correct: true})),
101
100
  },
102
101
  {
103
102
  properties: {
@@ -152,7 +151,7 @@ describe("evaluate", () => {
152
151
  const base = {
153
152
  evaluator: {
154
153
  assertion: {constraint: "exists", path: "answer"},
155
- dimensions: [{dataType: "boolean", key: "matches", required: true}],
154
+ dimensions: [{dataType: "boolean" as const, key: "matches", required: true}],
156
155
  name: "path-assert",
157
156
  type: "json-assert" as const,
158
157
  },
@@ -389,7 +388,7 @@ describe("evaluate", () => {
389
388
  input: {},
390
389
  output: "4",
391
390
  },
392
- {generateJsonObject: async () => ({correct: true})},
391
+ {generateJsonObject: judgeAnswering(() => ({correct: true}))},
393
392
  {properties: {correct: {type: "boolean"}}, type: "object"}
394
393
  );
395
394
  assert.equal(missingJudgePrompt.error, "llm-judge evaluator is missing judgePromptName");
@@ -408,7 +407,7 @@ describe("evaluate", () => {
408
407
  },
409
408
  {
410
409
  ai: {
411
- generateJsonObject: async () => ({confidence: "not-a-number", correct: 1}),
410
+ generateJsonObject: judgeAnswering(() => ({confidence: "not-a-number", correct: 1})),
412
411
  },
413
412
  judgeOutputSchema: {
414
413
  properties: {
@@ -461,9 +460,12 @@ describe("evaluate", () => {
461
460
  output: "4",
462
461
  },
463
462
  {
464
- generateJsonObject: async () => {
465
- return {confidence: "0.8", correct: "true", helpfulness: "not-a-number", tone: 1};
466
- },
463
+ generateJsonObject: judgeAnswering(() => ({
464
+ confidence: "0.8",
465
+ correct: "true",
466
+ helpfulness: "not-a-number",
467
+ tone: 1,
468
+ })),
467
469
  },
468
470
  {
469
471
  properties: {
@@ -489,7 +491,7 @@ describe("LocalEvaluatorStore phase 2", () => {
489
491
  beforeEach(async () => {
490
492
  new ObservabilityApp({plugins: [createLocalObservabilityBundle().plugin]});
491
493
  promptStore = new LocalPromptStore();
492
- store = new (await import("./local/evaluatorStore")).LocalEvaluatorStore(promptStore);
494
+ store = new (await import("./local/evaluatorStore.js")).LocalEvaluatorStore(promptStore);
493
495
  await registerObsEvaluator().deleteMany({});
494
496
  await clearObsPrompts();
495
497
  await promptStore.create({
@@ -170,10 +170,16 @@ export const runJsonAssertEvaluator = (
170
170
  };
171
171
  };
172
172
 
173
+ /**
174
+ * Run an llm-judge evaluator. By default the AI client resolves the judge prompt's production
175
+ * version by name. Pass `judgeSystemPrompt` (a specific version's resolved body) to run that
176
+ * version instead, as judge calibration experiments do.
177
+ */
173
178
  export const runLlmJudgeEvaluator = async (
174
179
  context: EvaluatorExecutionContext,
175
180
  ai: EvaluatorAiClient,
176
- judgeOutputSchema: Record<string, unknown>
181
+ judgeOutputSchema: Record<string, unknown>,
182
+ options: {judgeSystemPrompt?: string} = {}
177
183
  ): Promise<EvaluatorExecutionResult> => {
178
184
  const judgePromptName = context.evaluator.judgePromptName;
179
185
  if (!judgePromptName) {
@@ -181,9 +187,13 @@ export const runLlmJudgeEvaluator = async (
181
187
  }
182
188
 
183
189
  try {
190
+ const promptSource =
191
+ options.judgeSystemPrompt === undefined
192
+ ? {promptName: judgePromptName}
193
+ : {systemPrompt: options.judgeSystemPrompt};
184
194
  const payload = (await ai.generateJsonObject({
185
195
  prompt: buildJudgePrompt(context),
186
- promptName: judgePromptName,
196
+ ...promptSource,
187
197
  schema: jsonSchema(judgeOutputSchema) as FlexibleSchema<Record<string, unknown>>,
188
198
  schemaName: `${context.evaluator.name}-judge`,
189
199
  skipTrace: true,