@sixb/azure-ai-foundry 0.1.1 → 0.1.3

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/README.md CHANGED
@@ -26,7 +26,7 @@ The full project URL and API key are required. The provider uses the same key to
26
26
  deployments and call inference. Keys may be strings or synchronous/asynchronous functions;
27
27
  functions receive an abort signal and are evaluated for each Azure HTTP attempt.
28
28
 
29
- Model resolution always follows this sequence:
29
+ Language model resolution follows this sequence:
30
30
 
31
31
  1. Find the deployment in the Azure project and read its model name, version, and publisher.
32
32
  2. Look up the underlying model in models.dev.
@@ -40,6 +40,48 @@ If models.dev is unavailable after successful Azure discovery, resolution uses c
40
40
  facts or an explicit `definition`. Without either, it fails. Malformed catalog data still fails;
41
41
  missing capabilities and prices remain unknown. Catalog listings and refresh report outages.
42
42
 
43
+ ## Embeddings
44
+
45
+ Configure the resource endpoint separately: Foundry project endpoints do not route embeddings.
46
+
47
+ ```ts
48
+ const foundry = createAzureAIFoundry({
49
+ endpoint: "https://my-resource.services.ai.azure.com/api/projects/my-project",
50
+ apiKey: () => process.env.AZURE_AI_FOUNDRY_API_KEY,
51
+ embeddings: {
52
+ endpoint: "https://my-resource.openai.azure.com/openai/v1",
53
+ apiKey: () => process.env.AZURE_OPENAI_API_KEY,
54
+ },
55
+ })
56
+
57
+ export const productEmbedding = foundry.embedding("product-embedding", {
58
+ model: { name: "text-embedding-3-small", version: "1" },
59
+ dimensions: 1536,
60
+ })
61
+ ```
62
+
63
+ Register this binding under `models.embedding` and use it in an ObjectType's vector profile.
64
+ Set `model` to your deployment's actual model name and version. A mismatch blocks inference;
65
+ changing the declared model or version excludes previously stored vectors from search.
66
+ The deployment name must exist in the project and refer to the same deployment on the configured
67
+ resource. For connected deployments, supply the owning resource's endpoint and key.
68
+
69
+ | Deployed model | Dimensions |
70
+ | --- | --- |
71
+ | `text-embedding-3-small` | 1–1536 |
72
+ | `text-embedding-3-large` | 1–3072 |
73
+ | `text-embedding-ada-002` | Exactly 1536 |
74
+
75
+ Text batches accept up to 2048 nonempty strings; Azure enforces token limits. Other model families,
76
+ including Cohere, are rejected. Calls return vectors in input order and propagate cancellation.
77
+
78
+ Sixb object indexing and text search apply [usage accounting and limits](../../docs/models/usage-and-limits.md).
79
+ Direct `productEmbedding.embed({ texts: ["running shoes"] })` bypasses Sixb accounting.
80
+ Reference prices come from models.dev; use `rateCard` or `costEstimator` for deployment-specific
81
+ pricing. Missing pricing remains unknown and blocks cost-limited calls before inference.
82
+
83
+ See [embedding internals](./docs/embeddings.md) for resolution, transport, and failure handling.
84
+
43
85
  ## Stream a response
44
86
 
45
87
  ```ts
@@ -79,7 +121,7 @@ const cached = await model.resolve({ offline: true })
79
121
 
80
122
  | Behavior | Rule |
81
123
  | --- | --- |
82
- | Catalog listings | Supported project deployments |
124
+ | Catalog listings | Supported language deployments |
83
125
  | Azure matching | Unique case-insensitive model ID match |
84
126
  | `FW-` fallback | Unique normalized match in the models.dev Fireworks catalog |
85
127
  | Fireworks normalization | Removes namespace/`FW-`; normalizes casing and numeric `5p3` → `5.3` |
package/dist/catalog.d.ts CHANGED
@@ -6,12 +6,19 @@ export interface AzureAIFoundryCatalogOptions {
6
6
  readonly fetch?: (input: string | URL | Request, init?: RequestInit) => Promise<Response>;
7
7
  }
8
8
  export interface CatalogModel {
9
+ readonly kind: "language";
9
10
  readonly catalogProvider: "azure" | "fireworks-ai";
10
11
  readonly modelName: string;
11
12
  readonly protocol?: FoundryProtocol;
12
13
  readonly definition: ReturnType<typeof defineLanguageModel>;
13
14
  readonly rateCard?: LanguageModelRateCard;
14
15
  }
16
+ export interface CatalogEmbeddingModel {
17
+ readonly kind: "embedding";
18
+ readonly catalogProvider: "azure" | "fireworks-ai";
19
+ readonly modelName: string;
20
+ readonly rateCard?: LanguageModelRateCard;
21
+ }
15
22
  /** Provider-scoped models.dev snapshot. Construction and offline reads never perform I/O. */
16
23
  export declare class RemoteModelsDevCatalog {
17
24
  private snapshot;
@@ -21,6 +28,7 @@ export declare class RemoteModelsDevCatalog {
21
28
  constructor(options?: AzureAIFoundryCatalogOptions);
22
29
  invalidate(): void;
23
30
  get(name: string, offline?: boolean): Promise<CatalogModel | undefined>;
31
+ getEmbedding(name: string, offline?: boolean): Promise<CatalogEmbeddingModel | undefined>;
24
32
  /** Capture one snapshot so a batch of lookups cannot expire between deployments. */
25
33
  lookup(offline?: boolean): Promise<(name: string) => CatalogModel | undefined>;
26
34
  private load;
@@ -1 +1 @@
1
- {"version":3,"file":"catalog.d.ts","sourceRoot":"","sources":["../src/catalog.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,mBAAmB,EAEnB,KAAK,qBAAqB,EAI3B,MAAM,mBAAmB,CAAA;AAE1B,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA;AAKlD,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAA;IACvB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAA;IAC3B,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAA;CAC1F;AACD,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,eAAe,EAAE,OAAO,GAAG,cAAc,CAAA;IAClD,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,QAAQ,CAAC,EAAE,eAAe,CAAA;IACnC,QAAQ,CAAC,UAAU,EAAE,UAAU,CAAC,OAAO,mBAAmB,CAAC,CAAA;IAC3D,QAAQ,CAAC,QAAQ,CAAC,EAAE,qBAAqB,CAAA;CAC1C;AAED,6FAA6F;AAC7F,qBAAa,sBAAsB;IACjC,OAAO,CAAC,QAAQ,CAA8B;IAC9C,OAAO,CAAC,QAAQ,CAAI;IACpB,OAAO,CAAC,OAAO,CAAC,CAAkC;IAClD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA8B;gBAC1C,OAAO,GAAE,4BAAiC;IAYtD,UAAU;IAGJ,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,UAAQ,GAAG,OAAO,CAAC,YAAY,GAAG,SAAS,CAAC;IAG3E,oFAAoF;IAC9E,MAAM,CAAC,OAAO,UAAQ,GAAG,OAAO,CAAC,CAAC,IAAI,EAAE,MAAM,KAAK,YAAY,GAAG,SAAS,CAAC;IAyBlF,OAAO,CAAC,IAAI;YAcE,WAAW;CA8F1B"}
1
+ {"version":3,"file":"catalog.d.ts","sourceRoot":"","sources":["../src/catalog.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,mBAAmB,EAEnB,KAAK,qBAAqB,EAI3B,MAAM,mBAAmB,CAAA;AAE1B,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA;AAKlD,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAA;IACvB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAA;IAC3B,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,GAAG,OAAO,EAAE,IAAI,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,QAAQ,CAAC,CAAA;CAC1F;AACD,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAA;IACzB,QAAQ,CAAC,eAAe,EAAE,OAAO,GAAG,cAAc,CAAA;IAClD,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,QAAQ,CAAC,EAAE,eAAe,CAAA;IACnC,QAAQ,CAAC,UAAU,EAAE,UAAU,CAAC,OAAO,mBAAmB,CAAC,CAAA;IAC3D,QAAQ,CAAC,QAAQ,CAAC,EAAE,qBAAqB,CAAA;CAC1C;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,IAAI,EAAE,WAAW,CAAA;IAC1B,QAAQ,CAAC,eAAe,EAAE,OAAO,GAAG,cAAc,CAAA;IAClD,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,QAAQ,CAAC,EAAE,qBAAqB,CAAA;CAC1C;AAID,6FAA6F;AAC7F,qBAAa,sBAAsB;IACjC,OAAO,CAAC,QAAQ,CAA8B;IAC9C,OAAO,CAAC,QAAQ,CAAI;IACpB,OAAO,CAAC,OAAO,CAAC,CAAkC;IAClD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA8B;gBAC1C,OAAO,GAAE,4BAAiC;IAYtD,UAAU;IAGJ,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,UAAQ,GAAG,OAAO,CAAC,YAAY,GAAG,SAAS,CAAC;IAGrE,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,UAAQ,GAAG,OAAO,CAAC,qBAAqB,GAAG,SAAS,CAAC;IAU7F,oFAAoF;IAC9E,MAAM,CAAC,OAAO,UAAQ,GAAG,OAAO,CAAC,CAAC,IAAI,EAAE,MAAM,KAAK,YAAY,GAAG,SAAS,CAAC;IA2BlF,OAAO,CAAC,IAAI;YAcE,WAAW;CA+G1B"}
@@ -0,0 +1,31 @@
1
+ import { type EmbeddingModel, type LanguageModelRateCard, type ModelCostEstimator } from "@sixb/core/models";
2
+ import type { RemoteModelsDevCatalog } from "./catalog";
3
+ import type { FoundryDeployments, ResolvedDeployment } from "./discovery";
4
+ import type { AzureAIFoundryModelMetadata } from "./provider";
5
+ import { type FoundryTransport, type TransportOptions } from "./transport";
6
+ export type AzureAIFoundryEmbeddingTransportOptions = Pick<TransportOptions, "endpoint" | "apiKey" | "headers" | "fetch">;
7
+ export interface AzureAIFoundryEmbeddingOptions {
8
+ /** Pin the actual model behind the deployment; discovery must match before inference. */
9
+ readonly model: {
10
+ readonly name: string;
11
+ readonly version: string;
12
+ };
13
+ /** Expected output size; sent to text-embedding-3 models, validated locally for ada-002. */
14
+ readonly dimensions: number;
15
+ readonly rateCard?: LanguageModelRateCard;
16
+ readonly costEstimator?: ModelCostEstimator;
17
+ }
18
+ export interface AzureAIFoundryEmbeddingModel extends EmbeddingModel {
19
+ readonly metadata: AzureAIFoundryModelMetadata;
20
+ resolve(options?: {
21
+ readonly offline?: boolean;
22
+ }): Promise<AzureAIFoundryEmbeddingModel>;
23
+ }
24
+ interface EmbeddingResolution extends ResolvedDeployment {
25
+ readonly costEstimator: ModelCostEstimator;
26
+ readonly configurableDimensions: boolean;
27
+ }
28
+ /** Discovery uses project credentials; inference uses only the explicit resource transport. */
29
+ export declare function createFoundryEmbedding(providerId: string, modelId: string, options: AzureAIFoundryEmbeddingOptions, transport: FoundryTransport, deployments: FoundryDeployments, catalog: RemoteModelsDevCatalog, resolution?: EmbeddingResolution): AzureAIFoundryEmbeddingModel;
30
+ export {};
31
+ //# sourceMappingURL=embedding.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"embedding.d.ts","sourceRoot":"","sources":["../src/embedding.ts"],"names":[],"mappings":"AAAA,OAAO,EAEL,KAAK,cAAc,EAOnB,KAAK,qBAAqB,EAE1B,KAAK,kBAAkB,EAGxB,MAAM,mBAAmB,CAAA;AAE1B,OAAO,KAAK,EAAyB,sBAAsB,EAAE,MAAM,WAAW,CAAA;AAC9E,OAAO,KAAK,EAA4B,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAA;AACnG,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,YAAY,CAAA;AAC7D,OAAO,EAAa,KAAK,gBAAgB,EAAa,KAAK,gBAAgB,EAAE,MAAM,aAAa,CAAA;AAGhG,MAAM,MAAM,uCAAuC,GAAG,IAAI,CACxD,gBAAgB,EAChB,UAAU,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,CAC5C,CAAA;AAED,MAAM,WAAW,8BAA8B;IAC7C,yFAAyF;IACzF,QAAQ,CAAC,KAAK,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAA;IACnE,4FAA4F;IAC5F,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAA;IAC3B,QAAQ,CAAC,QAAQ,CAAC,EAAE,qBAAqB,CAAA;IACzC,QAAQ,CAAC,aAAa,CAAC,EAAE,kBAAkB,CAAA;CAC5C;AAED,MAAM,WAAW,4BAA6B,SAAQ,cAAc;IAClE,QAAQ,CAAC,QAAQ,EAAE,2BAA2B,CAAA;IAC9C,OAAO,CAAC,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,OAAO,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,OAAO,CAAC,4BAA4B,CAAC,CAAA;CACzF;AAED,UAAU,mBAAoB,SAAQ,kBAAkB;IACtD,QAAQ,CAAC,aAAa,EAAE,kBAAkB,CAAA;IAC1C,QAAQ,CAAC,sBAAsB,EAAE,OAAO,CAAA;CACzC;AAED,+FAA+F;AAC/F,wBAAgB,sBAAsB,CACpC,UAAU,EAAE,MAAM,EAClB,OAAO,EAAE,MAAM,EACf,OAAO,EAAE,8BAA8B,EACvC,SAAS,EAAE,gBAAgB,EAC3B,WAAW,EAAE,kBAAkB,EAC/B,OAAO,EAAE,sBAAsB,EAC/B,UAAU,CAAC,EAAE,mBAAmB,GAC/B,4BAA4B,CA+J9B"}
package/dist/index.d.ts CHANGED
@@ -1,5 +1,6 @@
1
1
  export type { AzureAIFoundryCatalogOptions } from "./catalog";
2
2
  export type { AzureAIFoundryCatalog, AzureAIFoundryDeployment, AzureAIFoundryDiscoveryOptions, } from "./discovery";
3
+ export type { AzureAIFoundryEmbeddingModel, AzureAIFoundryEmbeddingOptions, AzureAIFoundryEmbeddingTransportOptions, } from "./embedding";
3
4
  export type { AzureAIFoundryChatOptions, AzureAIFoundryMessagesOptions, AzureAIFoundryModel, AzureAIFoundryModelMetadata, AzureAIFoundryModelOptions, AzureAIFoundryOptions, AzureAIFoundryProvider, } from "./provider";
4
5
  export { createAzureAIFoundry } from "./provider";
5
6
  export type { FoundryProtocol } from "./transport";
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,4BAA4B,EAAE,MAAM,WAAW,CAAA;AAC7D,YAAY,EACV,qBAAqB,EACrB,wBAAwB,EACxB,8BAA8B,GAC/B,MAAM,aAAa,CAAA;AACpB,YAAY,EACV,yBAAyB,EACzB,6BAA6B,EAC7B,mBAAmB,EACnB,2BAA2B,EAC3B,0BAA0B,EAC1B,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,YAAY,CAAA;AACnB,OAAO,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAA;AACjD,YAAY,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,4BAA4B,EAAE,MAAM,WAAW,CAAA;AAC7D,YAAY,EACV,qBAAqB,EACrB,wBAAwB,EACxB,8BAA8B,GAC/B,MAAM,aAAa,CAAA;AACpB,YAAY,EACV,4BAA4B,EAC5B,8BAA8B,EAC9B,uCAAuC,GACxC,MAAM,aAAa,CAAA;AACpB,YAAY,EACV,yBAAyB,EACzB,6BAA6B,EAC7B,mBAAmB,EACnB,2BAA2B,EAC3B,0BAA0B,EAC1B,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,YAAY,CAAA;AACnB,OAAO,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAA;AACjD,YAAY,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA"}
package/dist/index.js CHANGED
@@ -2,9 +2,9 @@
2
2
  // src/provider.ts
3
3
  import {
4
4
  assertJsonObject,
5
- ModelCatalogUnavailableError as ModelCatalogUnavailableError3,
5
+ ModelCatalogUnavailableError as ModelCatalogUnavailableError4,
6
6
  ModelProviderError as ModelProviderError4,
7
- UnsupportedModelFeatureError as UnsupportedModelFeatureError5
7
+ UnsupportedModelFeatureError as UnsupportedModelFeatureError6
8
8
  } from "@sixb/core/models";
9
9
  import { chatEvents } from "@sixb/model-protocols/chat";
10
10
  import { messagesEvents } from "@sixb/model-protocols/messages";
@@ -159,18 +159,20 @@ class FoundryTransport {
159
159
  baseUrl;
160
160
  projectUrl;
161
161
  options;
162
- constructor(options) {
162
+ constructor(options, target = "project") {
163
163
  if (!URL.canParse(options.endpoint))
164
- throw new TypeError(`${PREFIX} endpoint must be a valid project URL.`);
164
+ throw new TypeError(`${PREFIX} endpoint must be a valid ${target} URL.`);
165
165
  const url = new URL(options.endpoint);
166
166
  if (!["https:", "http:"].includes(url.protocol) || url.username || url.password || url.search || url.hash) {
167
- throw new TypeError(`${PREFIX} endpoint must be an HTTP(S) project URL without credentials, query, or fragment.`);
167
+ throw new TypeError(`${PREFIX} endpoint must be an HTTP(S) ${target} URL without credentials, query, or fragment.`);
168
168
  }
169
169
  const path = url.pathname.replace(/\/+$/, "");
170
- if (!/^\/api\/projects\/[^/]+$/.test(path))
170
+ if (target === "project" && !/^\/api\/projects\/[^/]+$/.test(path))
171
171
  throw new TypeError(`${PREFIX} endpoint must be a Foundry project URL: https://<resource>.services.ai.azure.com/api/projects/<project>.`);
172
+ if (target === "resource" && path !== "" && path !== "/openai/v1")
173
+ throw new TypeError(`${PREFIX} embeddings.endpoint must be a resource origin or /openai/v1 URL.`);
172
174
  this.projectUrl = `${url.origin}${path}`;
173
- this.baseUrl = `${this.projectUrl}/openai/v1`;
175
+ this.baseUrl = target === "project" ? `${this.projectUrl}/openai/v1` : `${url.origin}/openai/v1`;
174
176
  if (typeof options.apiKey !== "function" && (typeof options.apiKey !== "string" || !options.apiKey.trim()))
175
177
  throw new TypeError(`${PREFIX} apiKey is required.`);
176
178
  if (options.maxRetries !== undefined && (!Number.isSafeInteger(options.maxRetries) || options.maxRetries < 0)) {
@@ -189,7 +191,7 @@ class FoundryTransport {
189
191
  signal.throwIfAborted();
190
192
  const headers = await this.headers(signal, diagnostics, protocol === "messages" ? "x-api-key" : "api-key");
191
193
  headers.set("content-type", "application/json");
192
- headers.set("accept", "text/event-stream");
194
+ headers.set("accept", protocol === "embeddings" ? "application/json" : "text/event-stream");
193
195
  if (protocol === "messages")
194
196
  headers.set("anthropic-version", "2023-06-01");
195
197
  signal.throwIfAborted();
@@ -217,7 +219,7 @@ class FoundryTransport {
217
219
  if (response.ok)
218
220
  return response;
219
221
  const error = this.redactError(response, await httpError(response, providerId, modelId, signal));
220
- if (!error.retryable || attempt >= (this.options.maxRetries ?? 2))
222
+ if (protocol === "embeddings" || !error.retryable || attempt >= (this.options.maxRetries ?? 2))
221
223
  throw error;
222
224
  await wait(Math.min(error.retryAfterMs ?? 250 * 2 ** attempt, this.options.maxRetryDelayMs ?? 60000), signal);
223
225
  }
@@ -249,6 +251,8 @@ class FoundryTransport {
249
251
  return (this.options.fetch ?? fetch)(url, init);
250
252
  }
251
253
  url(protocol) {
254
+ if (protocol === "embeddings")
255
+ return `${this.baseUrl}/embeddings`;
252
256
  if (protocol === "responses")
253
257
  return `${this.baseUrl}/responses`;
254
258
  if (protocol === "chat")
@@ -364,8 +368,13 @@ class RemoteModelsDevCatalog {
364
368
  async get(name, offline = false) {
365
369
  return (await this.lookup(offline))(name);
366
370
  }
367
- async lookup(offline = false) {
371
+ async getEmbedding(name, offline = false) {
368
372
  const models = offline ? this.snapshot : await this.load();
373
+ const matches = models.filter((m) => m.kind === "embedding" && m.catalogProvider === "azure" && m.modelName.toLowerCase() === name.toLowerCase());
374
+ return matches.length === 1 ? matches[0] : undefined;
375
+ }
376
+ async lookup(offline = false) {
377
+ const models = (offline ? this.snapshot : await this.load()).filter((m) => m.kind === "language");
369
378
  return (name) => {
370
379
  const matches = models.filter((m) => m.catalogProvider === "azure" && m.modelName.toLowerCase() === name.toLowerCase());
371
380
  if (matches.length)
@@ -437,6 +446,16 @@ class RemoteModelsDevCatalog {
437
446
  if (!raw || raw.id !== id)
438
447
  throw new TypeError(`${PREFIX} Invalid models.dev model identity.`);
439
448
  const modalities = object(raw.modalities);
449
+ if (raw.type === "embedding" || raw.family === "text-embedding" || raw.family === "cohere-embed" || /^(text-embedding-|cohere-embed)/i.test(id) || strings(modalities?.output).includes("embedding")) {
450
+ return [
451
+ {
452
+ kind: "embedding",
453
+ catalogProvider,
454
+ modelName: id,
455
+ rateCard: rates(object(raw.cost), undefined)
456
+ }
457
+ ];
458
+ }
440
459
  if (!strings(modalities?.output).includes("text"))
441
460
  return [];
442
461
  const limits = object(raw.limit);
@@ -444,6 +463,7 @@ class RemoteModelsDevCatalog {
444
463
  const protocol = catalogProvider === "fireworks-ai" ? "chat" : provider?.npm === "@ai-sdk/anthropic" ? "messages" : provider?.shape === "completions" || provider?.npm === "@ai-sdk/openai-compatible" ? "chat" : undefined;
445
464
  return [
446
465
  {
466
+ kind: "language",
447
467
  catalogProvider,
448
468
  modelName: id,
449
469
  protocol,
@@ -1314,6 +1334,218 @@ function deploymentRecord(raw, invalid) {
1314
1334
  });
1315
1335
  }
1316
1336
 
1337
+ // src/embedding.ts
1338
+ import {
1339
+ defineModelRateCard as defineModelRateCard3,
1340
+ EmbeddingModelResponseError,
1341
+ ModelCatalogUnavailableError as ModelCatalogUnavailableError3,
1342
+ UnsupportedModelFeatureError as UnsupportedModelFeatureError3
1343
+ } from "@sixb/core/models";
1344
+ function createFoundryEmbedding(providerId, modelId, options, transport, deployments, catalog, resolution) {
1345
+ const dimensions = options.dimensions;
1346
+ const expected = options.model;
1347
+ if (!expected || [expected.name, expected.version].some((value) => typeof value !== "string" || !value || value.trim() !== value)) {
1348
+ throw new TypeError(`${PREFIX} Embedding deployments require an expected model name and version.`);
1349
+ }
1350
+ const representation = Object.freeze({ name: expected.name, version: expected.version });
1351
+ const definition = Object.freeze({
1352
+ kind: "embedding",
1353
+ providerId,
1354
+ modelId,
1355
+ dimensions,
1356
+ representation
1357
+ });
1358
+ if (!modelId || modelId.trim() !== modelId || !Number.isSafeInteger(dimensions) || dimensions < 1 || dimensions > 16000)
1359
+ throw new TypeError(`${PREFIX} Embedding models require a nonempty deployment name and dimensions between 1 and 16000.`);
1360
+ if (options.rateCard && options.costEstimator)
1361
+ throw new TypeError(`${PREFIX} Configure either rateCard or costEstimator, not both.`);
1362
+ const settings = Object.freeze({
1363
+ model: representation,
1364
+ dimensions,
1365
+ rateCard: options.rateCard && defineModelRateCard3(options.rateCard),
1366
+ costEstimator: options.costEstimator
1367
+ });
1368
+ const deployment = resolution?.deployment;
1369
+ const metadata = Object.freeze(deployment ? {
1370
+ deployment,
1371
+ modelName: deployment.modelName,
1372
+ modelVersion: deployment.modelVersion,
1373
+ publisher: deployment.modelPublisher,
1374
+ sku: deployment.sku.name,
1375
+ discoveredAt: resolution.discoveredAt
1376
+ } : {});
1377
+ let pending;
1378
+ const binding = Object.freeze({
1379
+ providerId,
1380
+ modelId,
1381
+ definition,
1382
+ batching: Object.freeze({ maxInputs: 2048, maxInputBytes: 8191, maxTotalInputBytes: 300000 }),
1383
+ metadata,
1384
+ costEstimator: resolution?.costEstimator,
1385
+ async resolve(input) {
1386
+ if (resolution)
1387
+ return binding;
1388
+ const resolved = await deployments.resolve(modelId, input?.offline === true);
1389
+ if (resolved.deployment.modelName !== representation.name || resolved.deployment.modelVersion !== representation.version) {
1390
+ throw new TypeError(`${PREFIX} Embedding deployment '${modelId}' resolved to ${resolved.deployment.modelName}@${resolved.deployment.modelVersion}; expected ${representation.name}@${representation.version}.`);
1391
+ }
1392
+ const configurableDimensions = validateDeployment(resolved, dimensions);
1393
+ let entry;
1394
+ try {
1395
+ entry = await catalog.getEmbedding(resolved.deployment.modelName, input?.offline);
1396
+ } catch (error) {
1397
+ if (!(error instanceof ModelCatalogUnavailableError3))
1398
+ throw error;
1399
+ entry = await catalog.getEmbedding(resolved.deployment.modelName, true);
1400
+ console.warn(`${PREFIX} Embedding pricing catalog unavailable; using cached or explicit pricing when available.`);
1401
+ }
1402
+ const costEstimator = settings.costEstimator ?? tokenEstimator(settings.rateCard ?? entry?.rateCard, undefined, [], (raw) => !raw || Object.keys(raw).every((key) => ["prompt_tokens", "total_tokens"].includes(key) && counter(raw[key]) !== undefined), resolved.deployment.modelName, resolved.deployment.modelVersion, [
1403
+ modelId,
1404
+ resolved.deployment.modelName,
1405
+ `${resolved.deployment.modelName}-${resolved.deployment.modelVersion}`
1406
+ ]);
1407
+ return createFoundryEmbedding(providerId, modelId, settings, transport, deployments, catalog, {
1408
+ ...resolved,
1409
+ configurableDimensions,
1410
+ costEstimator
1411
+ });
1412
+ },
1413
+ async embed(input) {
1414
+ const signal = input.signal ?? AbortSignal.timeout(30000);
1415
+ signal.throwIfAborted();
1416
+ if (!Array.isArray(input.texts) || input.texts.length > 2048 || Array.from(input.texts).some((text) => typeof text !== "string" || !text.trim()))
1417
+ throw new TypeError(`${PREFIX} Embedding inputs must contain at most 2048 nonempty strings.`);
1418
+ const texts = [...input.texts];
1419
+ if (!texts.length)
1420
+ return { vectors: [] };
1421
+ if (!resolution) {
1422
+ pending ??= binding.resolve().catch((error) => {
1423
+ pending = undefined;
1424
+ throw error;
1425
+ });
1426
+ const resolved = await abortable(() => pending, signal);
1427
+ return resolved.embed({ texts, purpose: input.purpose, signal });
1428
+ }
1429
+ const response = await transport.post(JSON.stringify({
1430
+ model: modelId,
1431
+ input: texts,
1432
+ encoding_format: "float",
1433
+ ...resolution.configurableDimensions ? { dimensions } : {}
1434
+ }), signal, providerId, modelId, "embeddings");
1435
+ return readEmbeddingResult(response, transport, signal, {
1436
+ ...definition,
1437
+ count: texts.length,
1438
+ deployment: resolution.deployment
1439
+ });
1440
+ }
1441
+ });
1442
+ return binding;
1443
+ }
1444
+ function validateDeployment(resolved, dimensions) {
1445
+ const deployment = resolved.deployment;
1446
+ const name = deployment.modelName.toLowerCase();
1447
+ const maxDimensions = name === "text-embedding-3-small" ? 1536 : name === "text-embedding-3-large" ? 3072 : name === "text-embedding-ada-002" ? 1536 : undefined;
1448
+ if (deployment.modelPublisher.toLowerCase() !== "openai" || !maxDimensions || deployment.capabilities.embeddings === "false" || deployment.capabilities.embedding === "false")
1449
+ throw new UnsupportedModelFeatureError3(`${PREFIX} This embedding deployment is not supported. Use an OpenAI text-embedding-3-small, text-embedding-3-large, or text-embedding-ada-002 deployment.`);
1450
+ const configurable = name !== "text-embedding-ada-002";
1451
+ if (dimensions > maxDimensions || !configurable && dimensions !== maxDimensions)
1452
+ throw new TypeError(`${PREFIX} ${name} requires ${configurable ? `1\u2013${maxDimensions}` : maxDimensions} dimensions.`);
1453
+ return configurable;
1454
+ }
1455
+ function parseVectors(data, count, dimensions, invalid) {
1456
+ if (!Array.isArray(data) || data.length !== count)
1457
+ throw invalid();
1458
+ const vectors = new Array(count);
1459
+ for (const value of data) {
1460
+ const entry = record(value);
1461
+ const index = counter(entry?.index);
1462
+ const embedding = entry?.embedding;
1463
+ if (index === undefined || index >= count || vectors[index] !== undefined || !Array.isArray(embedding) || embedding.length !== dimensions)
1464
+ throw invalid();
1465
+ const vector = [];
1466
+ for (const n of embedding) {
1467
+ if (typeof n !== "number" || !Number.isFinite(Math.fround(n)))
1468
+ throw invalid();
1469
+ vector.push(n);
1470
+ }
1471
+ if (!vector.some((n) => Math.fround(n) !== 0))
1472
+ throw invalid();
1473
+ vectors[index] = vector;
1474
+ }
1475
+ return vectors;
1476
+ }
1477
+ function invalidResponse(providerId, modelId, metadata) {
1478
+ return new EmbeddingModelResponseError(`${PREFIX} Invalid embedding response: expected the resolved model and one finite, nonzero vector of the configured dimension per input, with unique indices.`, providerId, modelId, metadata);
1479
+ }
1480
+ async function readResponse(response, signal) {
1481
+ const reader = response.body?.getReader();
1482
+ if (!reader)
1483
+ throw new Error("Empty embedding response");
1484
+ const chunks = [];
1485
+ let bytes = 0;
1486
+ try {
1487
+ for (;; ) {
1488
+ const chunk = await abortable(() => reader.read(), signal);
1489
+ if (chunk.done)
1490
+ break;
1491
+ bytes += chunk.value.length;
1492
+ if (bytes > 128 * 1024 * 1024)
1493
+ throw new Error("Embedding response exceeds 128 MiB");
1494
+ chunks.push(chunk.value);
1495
+ }
1496
+ return JSON.parse(Buffer.concat(chunks).toString("utf8"));
1497
+ } finally {
1498
+ reader.cancel().catch(() => {});
1499
+ reader.releaseLock();
1500
+ }
1501
+ }
1502
+ function embeddingUsage(raw) {
1503
+ const inputTokens = counter(raw?.prompt_tokens);
1504
+ const totalTokens = counter(raw?.total_tokens);
1505
+ const knownInput = inputTokens !== undefined && (raw?.total_tokens === undefined || totalTokens === inputTokens);
1506
+ const unsupported = raw && Object.entries(raw).some(([key, value]) => !["prompt_tokens", "total_tokens"].includes(key) || counter(value) === undefined);
1507
+ return {
1508
+ ...knownInput ? { inputTokens, uncachedInputTokens: inputTokens } : {},
1509
+ outputTokens: 0,
1510
+ ...raw ? {
1511
+ raw: {
1512
+ ...inputTokens === undefined ? {} : { prompt_tokens: inputTokens },
1513
+ ...totalTokens === undefined ? {} : { total_tokens: totalTokens },
1514
+ ...unsupported ? { unsupported: true } : {}
1515
+ }
1516
+ } : {}
1517
+ };
1518
+ }
1519
+ function record(value) {
1520
+ return value !== null && typeof value === "object" && !Array.isArray(value) ? value : undefined;
1521
+ }
1522
+ async function readEmbeddingResult(response, transport, signal, expected) {
1523
+ const { providerId, modelId, dimensions } = expected;
1524
+ const facts = {
1525
+ providerIds: { requestId: transport.redactText(response, requestId(response)) }
1526
+ };
1527
+ let body;
1528
+ try {
1529
+ body = record(await readResponse(response, signal));
1530
+ } catch {
1531
+ if (signal.aborted)
1532
+ throw signal.reason;
1533
+ throw invalidResponse(providerId, modelId, facts);
1534
+ }
1535
+ const resultMetadata = {
1536
+ ...facts,
1537
+ usage: embeddingUsage(object(body?.usage)),
1538
+ responseModelId: typeof body?.model === "string" ? transport.redactText(response, body.model) : undefined
1539
+ };
1540
+ const name = expected.deployment.modelName;
1541
+ if (!body || ![modelId, name, `${name}-${expected.deployment.modelVersion}`].includes(String(body.model)))
1542
+ throw invalidResponse(providerId, modelId, resultMetadata);
1543
+ return {
1544
+ ...resultMetadata,
1545
+ vectors: parseVectors(body.data, expected.count, dimensions, () => invalidResponse(providerId, modelId, resultMetadata))
1546
+ };
1547
+ }
1548
+
1317
1549
  // src/messages-accounting.ts
1318
1550
  function foundryMessagesUsage(raw) {
1319
1551
  const uncached = counter(raw.input_tokens);
@@ -1382,7 +1614,7 @@ function tokenMeters2(raw) {
1382
1614
  // src/messages-request.ts
1383
1615
  import {
1384
1616
  modelReasoningSupportIssue as modelReasoningSupportIssue3,
1385
- UnsupportedModelFeatureError as UnsupportedModelFeatureError3
1617
+ UnsupportedModelFeatureError as UnsupportedModelFeatureError4
1386
1618
  } from "@sixb/core/models";
1387
1619
  import { messagesInput } from "@sixb/model-protocols/messages";
1388
1620
 
@@ -1498,7 +1730,7 @@ function foundryMessagesRequest(request, definition, options, scope) {
1498
1730
  let conversation = false;
1499
1731
  for (const message of request.messages) {
1500
1732
  if (message.role === "system" && conversation)
1501
- throw new UnsupportedModelFeatureError3(`${PREFIX} Messages system instructions must precede the conversation.`);
1733
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Messages system instructions must precede the conversation.`);
1502
1734
  if (message.role !== "system")
1503
1735
  conversation = true;
1504
1736
  }
@@ -1506,13 +1738,13 @@ function foundryMessagesRequest(request, definition, options, scope) {
1506
1738
  throw new TypeError(`${PREFIX} Native Messages requires maxOutputTokens in the definition, binding, or request.`);
1507
1739
  const caps = definition.capabilities;
1508
1740
  if (request.tools.length && caps.localTools !== true)
1509
- throw new UnsupportedModelFeatureError3(`${PREFIX} Configure localTools capability before using Messages tools.`);
1741
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Configure localTools capability before using Messages tools.`);
1510
1742
  const schema = request.responseFormat && messagesOutputSchema(request.responseFormat.schema);
1511
1743
  if (request.responseFormat && (caps.nativeStructuredOutput !== true || !schema))
1512
- throw new UnsupportedModelFeatureError3(`${PREFIX} Messages structured output requires a declared capability and a supported closed-object Claude schema.`);
1744
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Messages structured output requires a declared capability and a supported closed-object Claude schema.`);
1513
1745
  const thinking = reasoningRequest(request, definition, options, max);
1514
1746
  if (thinking.thinking && object(thinking.thinking)?.type !== "disabled" && ["temperature", "top_p", "top_k"].some((key) => options.request?.[key] !== undefined))
1515
- throw new UnsupportedModelFeatureError3(`${PREFIX} Explicit sampling controls cannot be combined with Messages thinking.`);
1747
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Explicit sampling controls cannot be combined with Messages thinking.`);
1516
1748
  const mapped = messagesInput(request.messages, definition.providerId, PREFIX);
1517
1749
  const output = {
1518
1750
  ...thinking.effort === undefined ? {} : { effort: thinking.effort },
@@ -1545,7 +1777,7 @@ function reasoningRequest(request, definition, options, max) {
1545
1777
  const caps = definition.capabilities.reasoning;
1546
1778
  const issue = modelReasoningSupportIssue3(caps, reasoning);
1547
1779
  if (caps === undefined || issue)
1548
- throw new UnsupportedModelFeatureError3(`${PREFIX} ${issue ?? "Configure reasoning capabilities for this deployment"}.`);
1780
+ throw new UnsupportedModelFeatureError4(`${PREFIX} ${issue ?? "Configure reasoning capabilities for this deployment"}.`);
1549
1781
  if (reasoning === "none") {
1550
1782
  return {
1551
1783
  thinking: { type: "disabled" }
@@ -1553,21 +1785,21 @@ function reasoningRequest(request, definition, options, max) {
1553
1785
  }
1554
1786
  if (typeof reasoning === "string") {
1555
1787
  if (reasoning === "minimal" || options.thinkingMode === "manual")
1556
- throw new UnsupportedModelFeatureError3(`${PREFIX} Named efforts require an adaptive-thinking Claude offering.`);
1788
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Named efforts require an adaptive-thinking Claude offering.`);
1557
1789
  return { thinking: { type: "adaptive" }, effort: reasoning };
1558
1790
  }
1559
1791
  if (options.thinkingMode === "adaptive")
1560
- throw new UnsupportedModelFeatureError3(`${PREFIX} This Claude offering only supports adaptive thinking, not manual budgets.`);
1792
+ throw new UnsupportedModelFeatureError4(`${PREFIX} This Claude offering only supports adaptive thinking, not manual budgets.`);
1561
1793
  if (reasoning.budgetTokens < 1024 || reasoning.budgetTokens >= max)
1562
- throw new UnsupportedModelFeatureError3(`${PREFIX} Messages reasoning budget must be at least 1024 and below maxOutputTokens (${max}).`);
1794
+ throw new UnsupportedModelFeatureError4(`${PREFIX} Messages reasoning budget must be at least 1024 and below maxOutputTokens (${max}).`);
1563
1795
  return { thinking: { type: "enabled", budget_tokens: reasoning.budgetTokens } };
1564
1796
  }
1565
1797
 
1566
1798
  // src/resolution.ts
1567
1799
  import {
1568
1800
  defineLanguageModel as defineLanguageModel2,
1569
- defineModelRateCard as defineModelRateCard3,
1570
- UnsupportedModelFeatureError as UnsupportedModelFeatureError4
1801
+ defineModelRateCard as defineModelRateCard4,
1802
+ UnsupportedModelFeatureError as UnsupportedModelFeatureError5
1571
1803
  } from "@sixb/core/models";
1572
1804
  function resolveModel(input) {
1573
1805
  const { deployment, options } = input;
@@ -1584,10 +1816,10 @@ function resolveModel(input) {
1584
1816
  const publisher = deploymentPublisher?.id === "fireworks-ai" ? entry?.definition.publisher ?? deploymentPublisher : deploymentPublisher ?? entry?.definition.publisher;
1585
1817
  const protocol = input.protocol ?? entry?.protocol ?? (deployment?.capabilities.responses === "true" ? "responses" : deployment?.capabilities.chatCompletion === "true" || deployment?.capabilities.chat_completion === "true" ? "chat" : deployment?.capabilities.messages === "true" ? "messages" : "responses");
1586
1818
  if (protocol === "messages" && deployment?.connectionName)
1587
- throw new UnsupportedModelFeatureError4(`${PREFIX} Native Messages cannot route a deployment from a project connection; use the owning resource's project URL and API key.`);
1819
+ throw new UnsupportedModelFeatureError5(`${PREFIX} Native Messages cannot route a deployment from a project connection; use the owning resource's project URL and API key.`);
1588
1820
  const flag = deployment?.capabilities[protocol === "chat" ? "chatCompletion" : protocol] ?? (protocol === "chat" ? deployment?.capabilities.chat_completion : undefined);
1589
1821
  if (flag === "false")
1590
- throw new UnsupportedModelFeatureError4(`${PREFIX} Deployment '${input.modelId}' reports that ${protocol === "chat" ? "Chat" : protocol === "responses" ? "Responses" : "Messages"} is unsupported.`);
1822
+ throw new UnsupportedModelFeatureError5(`${PREFIX} Deployment '${input.modelId}' reports that ${protocol === "chat" ? "Chat" : protocol === "responses" ? "Responses" : "Messages"} is unsupported.`);
1591
1823
  const definition = defineLanguageModel2({
1592
1824
  ...entry?.definition,
1593
1825
  publisher,
@@ -1622,7 +1854,7 @@ function resolveModel(input) {
1622
1854
  })
1623
1855
  } : {}
1624
1856
  }),
1625
- rateCard: options.rateCard ? defineModelRateCard3(options.rateCard) : entry?.rateCard,
1857
+ rateCard: options.rateCard ? defineModelRateCard4(options.rateCard) : entry?.rateCard,
1626
1858
  aliases: Object.freeze([
1627
1859
  ...new Set([
1628
1860
  input.modelId,
@@ -1679,6 +1911,12 @@ function createAzureAIFoundry(options) {
1679
1911
  throw new TypeError(`${PREFIX} providerId must not be empty.`);
1680
1912
  const deployments = new FoundryDeployments(providerId, transport, options.discovery);
1681
1913
  const modelCatalog = new RemoteModelsDevCatalog(options.catalog);
1914
+ const embeddingTransport = options.embeddings ? new FoundryTransport(options.embeddings, "resource") : undefined;
1915
+ const embedding = (name, input) => {
1916
+ if (!embeddingTransport)
1917
+ throw new TypeError(`${PREFIX} Configure embeddings.endpoint and embeddings.apiKey before declaring an embedding model.`);
1918
+ return createFoundryEmbedding(providerId, name, input, embeddingTransport, deployments, modelCatalog);
1919
+ };
1682
1920
  const model = (protocol, deploymentName, modelOptions) => {
1683
1921
  if (!deploymentName.trim() || deploymentName !== deploymentName.trim())
1684
1922
  throw new TypeError(`${PREFIX} deploymentName must be nonempty without surrounding whitespace.`);
@@ -1695,6 +1933,8 @@ function createAzureAIFoundry(options) {
1695
1933
  const results = [];
1696
1934
  for (const deployment of records) {
1697
1935
  const name = deployment.name;
1936
+ if (isEmbeddingDeployment(deployment) || await modelCatalog.getEmbedding(deployment.modelName, true))
1937
+ continue;
1698
1938
  try {
1699
1939
  const profile = resolveModel({
1700
1940
  providerId,
@@ -1709,7 +1949,7 @@ function createAzureAIFoundry(options) {
1709
1949
  if (supported)
1710
1950
  results.push(profile.definition);
1711
1951
  } catch (error) {
1712
- if (!(error instanceof UnsupportedModelFeatureError5))
1952
+ if (!(error instanceof UnsupportedModelFeatureError6))
1713
1953
  throw error;
1714
1954
  }
1715
1955
  }
@@ -1722,7 +1962,7 @@ function createAzureAIFoundry(options) {
1722
1962
  deployments: () => deployments.list()
1723
1963
  };
1724
1964
  const auto = (name, options = {}) => model(undefined, name, options);
1725
- return Object.assign(auto, { providerId, catalog, responses, messages, chat });
1965
+ return Object.assign(auto, { providerId, catalog, responses, messages, chat, embedding });
1726
1966
  }
1727
1967
 
1728
1968
  class FoundryModel {
@@ -1798,16 +2038,20 @@ class FoundryModel {
1798
2038
  if (this.resolution)
1799
2039
  return this;
1800
2040
  const resolution = await this.deployments.resolve(this.modelId, options?.offline === true);
2041
+ if (isEmbeddingDeployment(resolution.deployment))
2042
+ throw new UnsupportedModelFeatureError6(`${PREFIX} Use foundry.embedding() for an embedding deployment.`);
1801
2043
  let catalogModel;
1802
2044
  try {
1803
2045
  catalogModel = await this.modelCatalog.get(resolution.deployment.modelName, options?.offline);
1804
2046
  } catch (error) {
1805
- if (!(error instanceof ModelCatalogUnavailableError3))
2047
+ if (!(error instanceof ModelCatalogUnavailableError4))
1806
2048
  throw error;
1807
2049
  catalogModel = await this.modelCatalog.get(resolution.deployment.modelName, true);
1808
2050
  if (!catalogModel && !this.options.definition)
1809
2051
  throw error;
1810
2052
  }
2053
+ if (await this.modelCatalog.getEmbedding(resolution.deployment.modelName, true))
2054
+ throw new UnsupportedModelFeatureError6(`${PREFIX} Use foundry.embedding() for an embedding deployment.`);
1811
2055
  return new FoundryModel(this.requestedProtocol, this.transport, this.deployments, this.providerId, this.modelId, this.options, this.modelCatalog, {
1812
2056
  ...resolution,
1813
2057
  catalogModel
@@ -1891,8 +2135,11 @@ class FoundryModel {
1891
2135
  }
1892
2136
  }
1893
2137
  }
2138
+ function isEmbeddingDeployment(deployment) {
2139
+ return deployment.capabilities.embeddings === "true" || deployment.capabilities.embedding === "true" || /^(text-embedding-|cohere-embed)/i.test(deployment.modelName);
2140
+ }
1894
2141
  export {
1895
2142
  createAzureAIFoundry
1896
2143
  };
1897
2144
 
1898
- //# debugId=6D364F59E216D99A64756E2164756E21
2145
+ //# debugId=8F8ECB2AEAE14C3D64756E2164756E21