@terreno/ai 0.9.2 → 0.10.0

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.
@@ -0,0 +1,2 @@
1
+ [test]
2
+ root = "./src"
package/dist/aiApp.d.ts CHANGED
@@ -8,7 +8,9 @@ export interface AiAppOptions {
8
8
  /** Pre-configured AIService instance. Optional when using per-request keys or demo mode. */
9
9
  aiService?: AIService;
10
10
  /** Factory function to create a LanguageModel from a per-request API key (sent via x-ai-api-key header). */
11
- createModelFn?: (apiKey: string) => LanguageModel;
11
+ createModelFn?: (apiKey: string, modelId?: string) => LanguageModel;
12
+ /** Factory function to create a LanguageModel on the server side without a per-request key (e.g. Vertex AI with ADC). Returns undefined if no provider is configured. */
13
+ createServerModelFn?: (modelId?: string) => LanguageModel | undefined;
12
14
  /** When true and no AI service is available, routes return canned demo responses instead of failing. */
13
15
  demoMode?: boolean;
14
16
  /** File storage service for handling file uploads to GCS. */
package/dist/aiApp.js CHANGED
@@ -39,11 +39,12 @@ var AiApp = /** @class */ (function () {
39
39
  }
40
40
  AiApp.prototype.register = function (app) {
41
41
  var router = app;
42
- var _a = this.options, aiService = _a.aiService, createModelFn = _a.createModelFn, demoMode = _a.demoMode, fileStorageService = _a.fileStorageService, gcsBucket = _a.gcsBucket, maxSteps = _a.maxSteps, mcpService = _a.mcpService, openApiOptions = _a.openApiOptions, titleModelId = _a.titleModelId, toolChoice = _a.toolChoice, tools = _a.tools;
42
+ var _a = this.options, aiService = _a.aiService, createModelFn = _a.createModelFn, createServerModelFn = _a.createServerModelFn, demoMode = _a.demoMode, fileStorageService = _a.fileStorageService, gcsBucket = _a.gcsBucket, maxSteps = _a.maxSteps, mcpService = _a.mcpService, openApiOptions = _a.openApiOptions, titleModelId = _a.titleModelId, toolChoice = _a.toolChoice, tools = _a.tools;
43
43
  (0, gptHistories_1.addGptHistoryRoutes)(router, { openApiOptions: openApiOptions });
44
44
  (0, gpt_1.addGptRoutes)(router, {
45
45
  aiService: aiService,
46
46
  createModelFn: createModelFn,
47
+ createServerModelFn: createServerModelFn,
47
48
  demoMode: demoMode,
48
49
  maxSteps: maxSteps,
49
50
  mcpService: mcpService,
@@ -113,29 +113,44 @@ var sendDemoResponse = function (res, historyId) {
113
113
  };
114
114
  /**
115
115
  * Resolve the AIService for a request. Priority:
116
- * 1. Per-request API key via `x-ai-api-key` header (creates a temporary AIService)
117
- * 2. Pre-configured aiService from route options
118
- * 3. undefined (triggers demo mode response)
116
+ * 1. Per-request API key via `x-ai-api-key` header (creates a temporary AIService via createModelFn)
117
+ * 2. Specific model requested + server-side model factory (creates a temporary AIService via createServerModelFn)
118
+ * 3. Pre-configured aiService from route options
119
+ * 4. undefined (triggers demo mode response)
119
120
  */
120
121
  var resolveAiService = function (req, options, modelId) {
121
122
  var perRequestKey = req.headers["x-ai-api-key"];
122
123
  if (perRequestKey && options.createModelFn) {
123
124
  return new aiService_1.AIService({ model: options.createModelFn(perRequestKey, modelId) });
124
125
  }
126
+ if (modelId && options.createServerModelFn) {
127
+ var serverModel = options.createServerModelFn(modelId);
128
+ if (serverModel) {
129
+ return new aiService_1.AIService({ model: serverModel });
130
+ }
131
+ }
125
132
  return options.aiService;
126
133
  };
127
134
  /** Generate a short title for a conversation using a cheap model call. */
128
135
  var generateTitle = function (prompt, response, aiService, options, perRequestApiKey) { return __awaiter(void 0, void 0, void 0, function () {
129
- var titleService, conversationSnippet, title, _a;
136
+ var titleService, titleModel, conversationSnippet, title, _a;
130
137
  return __generator(this, function (_b) {
131
138
  switch (_b.label) {
132
139
  case 0:
133
140
  _b.trys.push([0, 2, , 3]);
134
141
  titleService = aiService;
135
- if (options.titleModelId && options.createModelFn && perRequestApiKey) {
136
- titleService = new aiService_1.AIService({
137
- model: options.createModelFn(perRequestApiKey, options.titleModelId),
138
- });
142
+ if (options.titleModelId) {
143
+ if (options.createModelFn && perRequestApiKey) {
144
+ titleService = new aiService_1.AIService({
145
+ model: options.createModelFn(perRequestApiKey, options.titleModelId),
146
+ });
147
+ }
148
+ else if (options.createServerModelFn) {
149
+ titleModel = options.createServerModelFn(options.titleModelId);
150
+ if (titleModel) {
151
+ titleService = new aiService_1.AIService({ model: titleModel });
152
+ }
153
+ }
139
154
  }
140
155
  conversationSnippet = "User: ".concat(prompt, "\nAssistant: ").concat(response.substring(0, 500));
141
156
  return [4 /*yield*/, titleService.generateText({
@@ -379,7 +394,10 @@ var addGptRoutes = function (router, options) {
379
394
  messages: messages,
380
395
  model: aiService.model,
381
396
  providerOptions: !supportsTools
382
- ? { google: { responseModalities: ["TEXT", "IMAGE"] } }
397
+ ? {
398
+ google: { responseModalities: ["TEXT", "IMAGE"] },
399
+ vertex: { responseModalities: ["TEXT", "IMAGE"] },
400
+ }
383
401
  : undefined,
384
402
  stopWhen: allTools ? (0, ai_1.stepCountIs)(maxSteps !== null && maxSteps !== void 0 ? maxSteps : 5) : (0, ai_1.stepCountIs)(1),
385
403
  system: effectiveSystemPrompt !== null && effectiveSystemPrompt !== void 0 ? effectiveSystemPrompt : undefined,
@@ -156,6 +156,8 @@ export interface GptRouteOptions {
156
156
  aiService?: import("../service/aiService").AIService;
157
157
  /** Factory to create a LanguageModel from a per-request API key (x-ai-api-key header). */
158
158
  createModelFn?: (apiKey: string, modelId?: string) => import("ai").LanguageModel;
159
+ /** Factory to create a LanguageModel on the server side without a per-request key (e.g. Vertex AI with ADC). Used for model switching when no x-ai-api-key header is present. Returns undefined if no provider is configured (falls through to demo mode). */
160
+ createServerModelFn?: (modelId?: string) => import("ai").LanguageModel | undefined;
159
161
  /** Factory to create per-request tools (e.g. tools that need the request's API key). Merged with static tools. */
160
162
  createRequestTools?: (req: import("express").Request) => Record<string, import("ai").Tool>;
161
163
  /** Return canned responses when no AI service is available. */
package/package.json CHANGED
@@ -10,7 +10,7 @@
10
10
  "@langfuse/otel": "^5.0.1",
11
11
  "@opentelemetry/api": "^1.9.0",
12
12
  "@opentelemetry/sdk-node": "^0.213.0",
13
- "@terreno/api": "0.9.2",
13
+ "@terreno/api": "0.10.0",
14
14
  "ai": "^6.0.97",
15
15
  "multer": "^1.4.5-lts.1"
16
16
  },
@@ -44,8 +44,11 @@
44
44
  "license": "Apache-2.0",
45
45
  "main": "dist/index.js",
46
46
  "name": "@terreno/ai",
47
+ "publishConfig": {
48
+ "access": "public"
49
+ },
47
50
  "peerDependencies": {
48
- "@terreno/api": "0.9.2",
51
+ "@terreno/api": "0.10.0",
49
52
  "mongoose": "^8.0.0"
50
53
  },
51
54
  "repository": {
@@ -63,5 +66,5 @@
63
66
  "test:ci": "bun test --preload ./src/tests/bunSetup.ts"
64
67
  },
65
68
  "types": "dist/index.d.ts",
66
- "version": "0.9.2"
69
+ "version": "0.10.0"
67
70
  }
package/src/aiApp.ts CHANGED
@@ -16,7 +16,9 @@ export interface AiAppOptions {
16
16
  /** Pre-configured AIService instance. Optional when using per-request keys or demo mode. */
17
17
  aiService?: AIService;
18
18
  /** Factory function to create a LanguageModel from a per-request API key (sent via x-ai-api-key header). */
19
- createModelFn?: (apiKey: string) => LanguageModel;
19
+ createModelFn?: (apiKey: string, modelId?: string) => LanguageModel;
20
+ /** Factory function to create a LanguageModel on the server side without a per-request key (e.g. Vertex AI with ADC). Returns undefined if no provider is configured. */
21
+ createServerModelFn?: (modelId?: string) => LanguageModel | undefined;
20
22
  /** When true and no AI service is available, routes return canned demo responses instead of failing. */
21
23
  demoMode?: boolean;
22
24
  /** File storage service for handling file uploads to GCS. */
@@ -75,6 +77,7 @@ export class AiApp implements TerrenoPlugin {
75
77
  const {
76
78
  aiService,
77
79
  createModelFn,
80
+ createServerModelFn,
78
81
  demoMode,
79
82
  fileStorageService,
80
83
  gcsBucket,
@@ -90,6 +93,7 @@ export class AiApp implements TerrenoPlugin {
90
93
  addGptRoutes(router, {
91
94
  aiService,
92
95
  createModelFn,
96
+ createServerModelFn,
93
97
  demoMode,
94
98
  maxSteps,
95
99
  mcpService,
@@ -1,4 +1,4 @@
1
- import {afterEach, beforeEach, describe, expect, it} from "bun:test";
1
+ import {afterEach, describe, expect, it, mock} from "bun:test";
2
2
 
3
3
  import {getCached, invalidateCache, LangfuseCache, setCached} from "./langfuseCache";
4
4
  import type {LangfuseCachedPrompt} from "./langfuseTypes";
@@ -13,71 +13,77 @@ const samplePrompt: LangfuseCachedPrompt = {
13
13
  version: 1,
14
14
  };
15
15
 
16
- describe("cache", () => {
17
- beforeEach(async () => {
18
- await LangfuseCache.deleteMany({});
19
- });
16
+ const originalFindOne = LangfuseCache.findOne;
17
+ const originalFindOneAndUpdate = LangfuseCache.findOneAndUpdate;
18
+ const originalDeleteMany = LangfuseCache.deleteMany;
20
19
 
21
- afterEach(async () => {
22
- await LangfuseCache.deleteMany({});
20
+ describe("cache", () => {
21
+ afterEach(() => {
22
+ LangfuseCache.findOne = originalFindOne;
23
+ LangfuseCache.findOneAndUpdate = originalFindOneAndUpdate;
24
+ LangfuseCache.deleteMany = originalDeleteMany;
23
25
  });
24
26
 
25
27
  describe("getCached", () => {
26
28
  it("returns null when no entry exists", async () => {
29
+ const lean = mock(async () => null);
30
+ const findOne = mock(() => ({lean}));
31
+ LangfuseCache.findOne = findOne as unknown as typeof LangfuseCache.findOne;
32
+
27
33
  const result = await getCached("prompt:missing:production");
28
- expect(result).toBeNull();
29
- });
30
34
 
31
- it("returns null for expired entries", async () => {
32
- await setCached("prompt:test:production", samplePrompt, -1);
33
- const result = await getCached("prompt:test:production");
34
35
  expect(result).toBeNull();
36
+ expect(findOne).toHaveBeenCalledTimes(1);
37
+ const [query] = (findOne as {mock: {calls: unknown[][]}}).mock.calls[0];
38
+ expect((query as {key: string}).key).toBe("prompt:missing:production");
39
+ expect((query as {expiresAt: {$gt: Date}}).expiresAt.$gt).toBeInstanceOf(Date);
35
40
  });
36
41
 
37
- it("returns the value for valid non-expired entries", async () => {
38
- await setCached("prompt:test:production", samplePrompt, 60);
42
+ it("returns cached value when entry exists", async () => {
43
+ const lean = mock(async () => ({value: samplePrompt}));
44
+ const findOne = mock(() => ({lean}));
45
+ LangfuseCache.findOne = findOne as unknown as typeof LangfuseCache.findOne;
46
+
39
47
  const result = await getCached("prompt:test:production");
40
- expect(result).not.toBeNull();
41
- expect(result?.name).toBe("test-prompt");
42
- expect(result?.type).toBe("text");
48
+
49
+ expect(result).toEqual(samplePrompt);
50
+ expect(lean).toHaveBeenCalledTimes(1);
43
51
  });
44
52
  });
45
53
 
46
54
  describe("setCached", () => {
47
- it("stores a new entry", async () => {
48
- await setCached("prompt:test:production", samplePrompt, 60);
49
- const count = await LangfuseCache.countDocuments({key: "prompt:test:production"});
50
- expect(count).toBe(1);
51
- });
55
+ it("upserts cache entries with computed expiry", async () => {
56
+ const findOneAndUpdate = mock(async () => ({}));
57
+ LangfuseCache.findOneAndUpdate =
58
+ findOneAndUpdate as unknown as typeof LangfuseCache.findOneAndUpdate;
52
59
 
53
- it("upserts when the same key is set twice", async () => {
60
+ const beforeCall = Date.now();
54
61
  await setCached("prompt:test:production", samplePrompt, 60);
55
- const updated: LangfuseCachedPrompt = {...samplePrompt, version: 2};
56
- await setCached("prompt:test:production", updated, 60);
62
+ const afterCall = Date.now();
57
63
 
58
- const count = await LangfuseCache.countDocuments({key: "prompt:test:production"});
59
- expect(count).toBe(1);
64
+ expect(findOneAndUpdate).toHaveBeenCalledTimes(1);
65
+ const [filter, update, options] = (findOneAndUpdate as {mock: {calls: unknown[][]}}).mock
66
+ .calls[0];
67
+ expect(filter).toEqual({key: "prompt:test:production"});
68
+ expect((update as {key: string}).key).toBe("prompt:test:production");
69
+ expect((update as {value: LangfuseCachedPrompt}).value).toEqual(samplePrompt);
70
+ expect(options).toEqual({upsert: true});
60
71
 
61
- const result = await getCached("prompt:test:production");
62
- expect(result?.version).toBe(2);
72
+ const expiresAt = (update as {expiresAt: Date}).expiresAt.getTime();
73
+ expect(expiresAt).toBeGreaterThanOrEqual(beforeCall + 59000);
74
+ expect(expiresAt).toBeLessThanOrEqual(afterCall + 61000);
63
75
  });
64
76
  });
65
77
 
66
78
  describe("invalidateCache", () => {
67
- it("removes entries matching the pattern", async () => {
68
- await setCached("prompt:test-prompt:production", samplePrompt, 60);
69
- await setCached("prompt:test-prompt:staging", {...samplePrompt, version: 2}, 60);
70
- await setCached("prompt:other-prompt:production", {...samplePrompt, name: "other"}, 60);
79
+ it("passes regex key pattern to deleteMany", async () => {
80
+ const deleteMany = mock(async () => ({}));
81
+ LangfuseCache.deleteMany = deleteMany as unknown as typeof LangfuseCache.deleteMany;
71
82
 
72
83
  await invalidateCache("prompt:test-prompt:");
73
84
 
74
- const remaining = await LangfuseCache.find({}).lean();
75
- expect(remaining.length).toBe(1);
76
- expect(remaining[0].key).toBe("prompt:other-prompt:production");
77
- });
78
-
79
- it("does not throw when no entries match", async () => {
80
- await expect(invalidateCache("prompt:nonexistent:")).resolves.toBeUndefined();
85
+ expect(deleteMany).toHaveBeenCalledTimes(1);
86
+ expect(deleteMany).toHaveBeenCalledWith({key: {$regex: "prompt:test-prompt:"}});
81
87
  });
82
88
  });
83
89
  });
@@ -1,8 +1,26 @@
1
- import {afterEach, beforeEach, describe, expect, it, mock} from "bun:test";
1
+ import {beforeEach, describe, expect, it, mock} from "bun:test";
2
2
 
3
- import {LangfuseCache} from "./langfuseCache";
4
3
  import type {LangfuseCachedPrompt} from "./langfuseTypes";
5
4
 
5
+ const cacheValues = new Map<string, LangfuseCachedPrompt>();
6
+
7
+ const mockLoggerDebug = mock(() => {});
8
+ const mockLoggerInfo = mock(() => {});
9
+ const mockGetCached = mock(async (key: string) => {
10
+ return cacheValues.get(key) ?? null;
11
+ });
12
+ const mockSetCached = mock(async (key: string, value: LangfuseCachedPrompt) => {
13
+ cacheValues.set(key, value);
14
+ });
15
+ const mockInvalidateCache = mock(async (prefix: string) => {
16
+ const keys = [...cacheValues.keys()];
17
+ for (const key of keys) {
18
+ if (key.startsWith(prefix)) {
19
+ cacheValues.delete(key);
20
+ }
21
+ }
22
+ });
23
+ const mockPromptCreate = mock(async (_params: Record<string, unknown>) => {});
6
24
  const mockPromptGet = mock(async (_name: string, _options?: Record<string, unknown>) => ({
7
25
  config: {},
8
26
  labels: ["production"],
@@ -13,11 +31,26 @@ const mockPromptGet = mock(async (_name: string, _options?: Record<string, unkno
13
31
  version: 1,
14
32
  }));
15
33
 
34
+ mock.module("@terreno/api", () => ({
35
+ logger: {
36
+ debug: mockLoggerDebug,
37
+ info: mockLoggerInfo,
38
+ },
39
+ }));
40
+
41
+ mock.module("./langfuseCache", () => ({
42
+ getCached: mockGetCached,
43
+ invalidateCache: mockInvalidateCache,
44
+ setCached: mockSetCached,
45
+ }));
46
+
16
47
  mock.module("./langfuseClient", () => ({
17
- getLangfuseClient: () => ({prompt: {get: mockPromptGet}}),
48
+ getLangfuseClient: () => ({prompt: {create: mockPromptCreate, get: mockPromptGet}}),
18
49
  }));
19
50
 
20
- const {compilePrompt, getPrompt} = await import("./langfusePrompts");
51
+ const {compilePrompt, createPrompt, getPrompt, invalidatePromptCache} = await import(
52
+ "./langfusePrompts"
53
+ );
21
54
 
22
55
  const textPrompt: LangfuseCachedPrompt = {
23
56
  config: {},
@@ -42,84 +75,92 @@ const chatPrompt: LangfuseCachedPrompt = {
42
75
  version: 1,
43
76
  };
44
77
 
45
- describe("compilePrompt", () => {
46
- describe("text prompts", () => {
47
- it("replaces variables in a text prompt", () => {
48
- const result = compilePrompt(textPrompt, {age: "30", name: "Alice"});
49
- expect(result).toBe("Hello Alice, you are 30 years old.");
50
- });
78
+ beforeEach(() => {
79
+ cacheValues.clear();
80
+ mockLoggerDebug.mockClear();
81
+ mockLoggerInfo.mockClear();
82
+ mockGetCached.mockClear();
83
+ mockSetCached.mockClear();
84
+ mockInvalidateCache.mockClear();
85
+ mockPromptCreate.mockClear();
86
+ mockPromptGet.mockClear();
87
+ mockPromptGet.mockImplementation(async () => ({
88
+ config: {},
89
+ labels: ["production"],
90
+ name: "test-prompt",
91
+ prompt: "Hello world",
92
+ tags: [],
93
+ type: "text" as const,
94
+ version: 1,
95
+ }));
96
+ });
51
97
 
52
- it("leaves unreplaced variables as-is when not provided", () => {
53
- const result = compilePrompt(textPrompt, {name: "Alice"});
54
- expect(result).toBe("Hello Alice, you are {{age}} years old.");
55
- });
98
+ describe("compilePrompt", () => {
99
+ it("replaces variables in text prompts", () => {
100
+ const result = compilePrompt(textPrompt, {age: "30", name: "Alice"});
101
+ expect(result).toBe("Hello Alice, you are 30 years old.");
102
+ });
56
103
 
57
- it("returns prompt unchanged when no variables provided", () => {
58
- const noVarPrompt: LangfuseCachedPrompt = {
59
- ...textPrompt,
60
- prompt: "Hello world.",
61
- };
62
- const result = compilePrompt(noVarPrompt);
63
- expect(result).toBe("Hello world.");
64
- });
104
+ it("keeps missing text variables unreplaced", () => {
105
+ const result = compilePrompt(textPrompt, {name: "Alice"});
106
+ expect(result).toBe("Hello Alice, you are {{age}} years old.");
65
107
  });
66
108
 
67
- describe("chat prompts", () => {
68
- it("replaces variables in all chat messages", () => {
69
- const result = compilePrompt(chatPrompt, {company: "Acme", name: "Bob"});
70
- expect(Array.isArray(result)).toBe(true);
71
- const messages = result as Array<{role: string; content: string}>;
72
- expect(messages[0].content).toBe("You are a helpful assistant for Acme.");
73
- expect(messages[1].content).toBe("My name is Bob.");
74
- });
109
+ it("replaces variables in chat prompts", () => {
110
+ const result = compilePrompt(chatPrompt, {company: "Acme", name: "Bob"}) as Array<{
111
+ content: string;
112
+ role: string;
113
+ }>;
75
114
 
76
- it("preserves roles in chat messages", () => {
77
- const result = compilePrompt(chatPrompt, {company: "X", name: "Y"}) as Array<{
78
- role: string;
79
- content: string;
80
- }>;
81
- expect(result[0].role).toBe("system");
82
- expect(result[1].role).toBe("user");
83
- });
115
+ expect(result[0]).toEqual({content: "You are a helpful assistant for Acme.", role: "system"});
116
+ expect(result[1]).toEqual({content: "My name is Bob.", role: "user"});
84
117
  });
85
118
  });
86
119
 
87
120
  describe("getPrompt", () => {
88
- beforeEach(async () => {
89
- await LangfuseCache.deleteMany({});
90
- mockPromptGet.mockClear();
91
- mockPromptGet.mockImplementation(async (_name: string, _options?: Record<string, unknown>) => ({
92
- config: {},
93
- labels: ["production"],
94
- name: "test-prompt",
95
- prompt: "Hello world",
96
- tags: [],
97
- type: "text" as const,
98
- version: 1,
99
- }));
100
- });
121
+ it("returns cached prompts without calling Langfuse", async () => {
122
+ cacheValues.set("prompt:test-prompt:production", textPrompt);
101
123
 
102
- afterEach(async () => {
103
- await LangfuseCache.deleteMany({});
124
+ const result = await getPrompt("test-prompt");
125
+
126
+ expect(result).toEqual(textPrompt);
127
+ expect(mockPromptGet).toHaveBeenCalledTimes(0);
128
+ expect(mockLoggerDebug).toHaveBeenCalledWith(
129
+ 'Langfuse prompt cache hit: "test-prompt" v1 (label: production)'
130
+ );
104
131
  });
105
132
 
106
- it("fetches from Langfuse when cache is empty", async () => {
133
+ it("fetches and caches prompts with default ttl/label", async () => {
107
134
  const result = await getPrompt("test-prompt");
108
- expect(result.name).toBe("test-prompt");
109
- expect(result.type).toBe("text");
110
- expect(mockPromptGet).toHaveBeenCalledTimes(1);
111
- });
112
135
 
113
- it("returns cached result on second call", async () => {
114
- await getPrompt("test-prompt");
115
- await getPrompt("test-prompt");
116
- expect(mockPromptGet).toHaveBeenCalledTimes(1);
136
+ expect(mockPromptGet).toHaveBeenCalledWith("test-prompt", {
137
+ cacheTtlSeconds: 0,
138
+ label: "production",
139
+ });
140
+ expect(mockSetCached).toHaveBeenCalledWith("prompt:test-prompt:production", result, 60);
141
+ expect(cacheValues.get("prompt:test-prompt:production")).toEqual(result);
117
142
  });
118
143
 
119
- it("writes fetched prompt to cache", async () => {
120
- await getPrompt("test-prompt");
121
- const countAfter = await LangfuseCache.countDocuments({key: "prompt:test-prompt:production"});
122
- expect(countAfter).toBe(1);
144
+ it("fetches and caches prompts with custom ttl/label", async () => {
145
+ await getPrompt("test-prompt", {label: "staging"}, {cache: {promptTtlSeconds: 120}});
146
+
147
+ expect(mockPromptGet).toHaveBeenCalledWith("test-prompt", {
148
+ cacheTtlSeconds: 0,
149
+ label: "staging",
150
+ });
151
+ expect(mockSetCached).toHaveBeenCalledWith(
152
+ "prompt:test-prompt:staging",
153
+ {
154
+ config: {},
155
+ labels: ["production"],
156
+ name: "test-prompt",
157
+ prompt: "Hello world",
158
+ tags: [],
159
+ type: "text",
160
+ version: 1,
161
+ },
162
+ 120
163
+ );
123
164
  });
124
165
 
125
166
  it("throws when Langfuse fetch fails", async () => {
@@ -130,3 +171,72 @@ describe("getPrompt", () => {
130
171
  await expect(getPrompt("missing-prompt")).rejects.toThrow("Prompt not found");
131
172
  });
132
173
  });
174
+
175
+ describe("createPrompt", () => {
176
+ it("creates text prompts with default labels/tags", async () => {
177
+ const result = await createPrompt({
178
+ name: "created-prompt",
179
+ prompt: "Text prompt body",
180
+ type: "text",
181
+ });
182
+
183
+ expect(mockPromptCreate).toHaveBeenCalledWith({
184
+ config: undefined,
185
+ labels: ["production"],
186
+ name: "created-prompt",
187
+ prompt: "Text prompt body",
188
+ tags: [],
189
+ type: "text",
190
+ });
191
+ expect(mockInvalidateCache).toHaveBeenCalledWith("prompt:created-prompt:");
192
+ expect(mockPromptGet).toHaveBeenCalledWith("created-prompt", {
193
+ cacheTtlSeconds: 0,
194
+ label: "production",
195
+ });
196
+ expect(result.name).toBe("test-prompt");
197
+ });
198
+
199
+ it("creates chat prompts preserving role and content", async () => {
200
+ await createPrompt({
201
+ config: {temperature: 0.3},
202
+ labels: ["staging"],
203
+ name: "chat-prompt",
204
+ prompt: [
205
+ {content: "System says hello", role: "system"},
206
+ {content: "User asks question", role: "user"},
207
+ ],
208
+ tags: ["support"],
209
+ type: "chat",
210
+ });
211
+
212
+ expect(mockPromptCreate).toHaveBeenCalledWith({
213
+ config: {temperature: 0.3},
214
+ labels: ["staging"],
215
+ name: "chat-prompt",
216
+ prompt: [
217
+ {content: "System says hello", role: "system"},
218
+ {content: "User asks question", role: "user"},
219
+ ],
220
+ tags: ["support"],
221
+ type: "chat",
222
+ });
223
+ });
224
+ });
225
+
226
+ describe("invalidatePromptCache", () => {
227
+ it("invalidates cache entries for a prompt prefix", async () => {
228
+ cacheValues.set("prompt:target-prompt:production", textPrompt);
229
+ cacheValues.set("prompt:target-prompt:staging", textPrompt);
230
+ cacheValues.set("prompt:other-prompt:production", textPrompt);
231
+
232
+ await invalidatePromptCache("target-prompt");
233
+
234
+ expect(mockInvalidateCache).toHaveBeenCalledWith("prompt:target-prompt:");
235
+ expect(cacheValues.has("prompt:target-prompt:production")).toBe(false);
236
+ expect(cacheValues.has("prompt:target-prompt:staging")).toBe(false);
237
+ expect(cacheValues.has("prompt:other-prompt:production")).toBe(true);
238
+ expect(mockLoggerInfo).toHaveBeenCalledWith(
239
+ "Langfuse prompt cache invalidated for: target-prompt"
240
+ );
241
+ });
242
+ });
@@ -0,0 +1,57 @@
1
+ import {describe, expect, it, mock} from "bun:test";
2
+ import type {NextFunction, Request, Response} from "express";
3
+
4
+ import {requireAdmin} from "./langfuseRoutesMiddleware";
5
+
6
+ const buildRequest = (user: unknown): Request => {
7
+ return {user} as Request;
8
+ };
9
+
10
+ const buildResponse = (): Response => {
11
+ return {} as Response;
12
+ };
13
+
14
+ describe("requireAdmin", () => {
15
+ it("calls next when user.isAdmin returns true", () => {
16
+ const req = buildRequest({
17
+ admin: false,
18
+ isAdmin: () => true,
19
+ });
20
+ const next = mock(() => {}) as NextFunction;
21
+
22
+ requireAdmin(req, buildResponse(), next);
23
+
24
+ expect(next).toHaveBeenCalledTimes(1);
25
+ });
26
+
27
+ it("uses Permissions.IsAdmin fallback when isAdmin is missing", () => {
28
+ const req = buildRequest({admin: true});
29
+ const next = mock(() => {}) as NextFunction;
30
+
31
+ requireAdmin(req, buildResponse(), next);
32
+
33
+ expect(next).toHaveBeenCalledTimes(1);
34
+ });
35
+
36
+ it("throws APIError with 403 when user is not admin", () => {
37
+ const req = buildRequest({admin: false});
38
+ const next = mock(() => {}) as NextFunction;
39
+
40
+ try {
41
+ requireAdmin(req, buildResponse(), next);
42
+ throw new Error("Expected requireAdmin to throw");
43
+ } catch (error) {
44
+ const apiError = error as {
45
+ disableExternalErrorTracking?: boolean;
46
+ name?: string;
47
+ status?: number;
48
+ title?: string;
49
+ };
50
+ expect(apiError.name).toBe("APIError");
51
+ expect(apiError.status).toBe(403);
52
+ expect(apiError.title).toBe("Forbidden");
53
+ expect(apiError.disableExternalErrorTracking).toBe(true);
54
+ expect(next).toHaveBeenCalledTimes(0);
55
+ }
56
+ });
57
+ });
package/src/routes/gpt.ts CHANGED
@@ -33,9 +33,10 @@ const sendDemoResponse = (res: express.Response, historyId?: string): void => {
33
33
 
34
34
  /**
35
35
  * Resolve the AIService for a request. Priority:
36
- * 1. Per-request API key via `x-ai-api-key` header (creates a temporary AIService)
37
- * 2. Pre-configured aiService from route options
38
- * 3. undefined (triggers demo mode response)
36
+ * 1. Per-request API key via `x-ai-api-key` header (creates a temporary AIService via createModelFn)
37
+ * 2. Specific model requested + server-side model factory (creates a temporary AIService via createServerModelFn)
38
+ * 3. Pre-configured aiService from route options
39
+ * 4. undefined (triggers demo mode response)
39
40
  */
40
41
  const resolveAiService = (
41
42
  req: express.Request,
@@ -46,6 +47,12 @@ const resolveAiService = (
46
47
  if (perRequestKey && options.createModelFn) {
47
48
  return new AIService({model: options.createModelFn(perRequestKey, modelId)});
48
49
  }
50
+ if (modelId && options.createServerModelFn) {
51
+ const serverModel = options.createServerModelFn(modelId);
52
+ if (serverModel) {
53
+ return new AIService({model: serverModel});
54
+ }
55
+ }
49
56
  return options.aiService;
50
57
  };
51
58
 
@@ -59,10 +66,17 @@ const generateTitle = async (
59
66
  ): Promise<string | undefined> => {
60
67
  try {
61
68
  let titleService = aiService;
62
- if (options.titleModelId && options.createModelFn && perRequestApiKey) {
63
- titleService = new AIService({
64
- model: options.createModelFn(perRequestApiKey, options.titleModelId),
65
- });
69
+ if (options.titleModelId) {
70
+ if (options.createModelFn && perRequestApiKey) {
71
+ titleService = new AIService({
72
+ model: options.createModelFn(perRequestApiKey, options.titleModelId),
73
+ });
74
+ } else if (options.createServerModelFn) {
75
+ const titleModel = options.createServerModelFn(options.titleModelId);
76
+ if (titleModel) {
77
+ titleService = new AIService({model: titleModel});
78
+ }
79
+ }
66
80
  }
67
81
  const conversationSnippet = `User: ${prompt}\nAssistant: ${response.substring(0, 500)}`;
68
82
  const title = await titleService.generateText({
@@ -299,7 +313,10 @@ export const addGptRoutes = (router: any, options: GptRouteOptions): void => {
299
313
  messages,
300
314
  model: (aiService as any).model,
301
315
  providerOptions: !supportsTools
302
- ? {google: {responseModalities: ["TEXT", "IMAGE"]}}
316
+ ? {
317
+ google: {responseModalities: ["TEXT", "IMAGE"]},
318
+ vertex: {responseModalities: ["TEXT", "IMAGE"]},
319
+ }
303
320
  : undefined,
304
321
  stopWhen: allTools ? stepCountIs(maxSteps ?? 5) : stepCountIs(1),
305
322
  system: effectiveSystemPrompt ?? undefined,
@@ -27,7 +27,7 @@ export const addProjectRoutes = (
27
27
  read: [Permissions.IsOwner],
28
28
  update: [Permissions.IsOwner],
29
29
  },
30
- preCreate: (body: Record<string, unknown>, req: express.Request) =>
30
+ preCreate: (body, req: express.Request) =>
31
31
  ({
32
32
  ...body,
33
33
  userId: (req as any).user?._id,
@@ -4,19 +4,25 @@ import {setupEnvironment, winstonLogger} from "@terreno/api";
4
4
  import mongoose from "mongoose";
5
5
  import winston from "winston";
6
6
 
7
+ const shouldConnectToTestDb = process.env.BUN_TEST_DISABLE_DB !== "true";
8
+
7
9
  // Connect to MongoDB once for all tests
8
- beforeAll(async () => {
9
- await mongoose
10
- .connect("mongodb://127.0.0.1/terreno-ai-test?&connectTimeoutMS=360000")
11
- .catch((err) => {
12
- console.error("Failed to connect to MongoDB:", err);
13
- });
14
- });
10
+ if (shouldConnectToTestDb) {
11
+ beforeAll(async () => {
12
+ await mongoose
13
+ .connect("mongodb://127.0.0.1/terreno-ai-test?&connectTimeoutMS=360000")
14
+ .catch((err) => {
15
+ console.error("Failed to connect to MongoDB:", err);
16
+ });
17
+ });
18
+ }
15
19
 
16
20
  // Close MongoDB connection after all tests
17
- afterAll(async () => {
18
- await mongoose.connection.close();
19
- });
21
+ if (shouldConnectToTestDb) {
22
+ afterAll(async () => {
23
+ await mongoose.connection.close();
24
+ });
25
+ }
20
26
 
21
27
  let logs: string[] = [];
22
28
 
@@ -212,6 +212,8 @@ export interface GptRouteOptions {
212
212
  aiService?: import("../service/aiService").AIService;
213
213
  /** Factory to create a LanguageModel from a per-request API key (x-ai-api-key header). */
214
214
  createModelFn?: (apiKey: string, modelId?: string) => import("ai").LanguageModel;
215
+ /** Factory to create a LanguageModel on the server side without a per-request key (e.g. Vertex AI with ADC). Used for model switching when no x-ai-api-key header is present. Returns undefined if no provider is configured (falls through to demo mode). */
216
+ createServerModelFn?: (modelId?: string) => import("ai").LanguageModel | undefined;
215
217
  /** Factory to create per-request tools (e.g. tools that need the request's API key). Merged with static tools. */
216
218
  createRequestTools?: (req: import("express").Request) => Record<string, import("ai").Tool>;
217
219
  /** Return canned responses when no AI service is available. */