update node sdk with streaming/non-streaming return type fix (#5)

* Sync updates from stainless branch: yanxi0830/dev

* add examples

Author: yanxi0830

examples/agents.ts

@@ -2,7 +2,7 @@
 
 import { AgentConfig } from "llama-stack-client/resources/shared";
 
-const LlamaStackClient = require('llama-stack-client').default;
+import LlamaStackClient from 'llama-stack-client';
 const client = new LlamaStackClient({ baseURL: 'http://localhost:8321' });
 
 async function main() {

examples/inference.ts

@@ -1,6 +1,6 @@
 #!/usr/bin/env -S npm run tsn -T
 
-const LlamaStackClient = require('llama-stack-client').default;
+import LlamaStackClient from 'llama-stack-client';
 const client = new LlamaStackClient({ baseURL: 'http://localhost:8321' });
 
 async function main() {
@@ -22,7 +22,9 @@ async function main() {
     stream: true,
   });
   for await (const chunk of stream) {
-    process.stdout.write(chunk.event.delta.text || '');
+    if (chunk.event.delta.type === 'text') {
+      process.stdout.write(chunk.event.delta.text || '');
+    }
   }
   process.stdout.write('\n');
 }

src/index.ts

@@ -33,17 +33,16 @@ import {
   ListEvalTasksResponse,
 } from './resources/eval-tasks';
 import {
+  ChatCompletionResponseStreamChunk,
   CompletionResponse,
   EmbeddingsResponse,
   Inference,
   InferenceChatCompletionParams,
   InferenceChatCompletionParamsNonStreaming,
   InferenceChatCompletionParamsStreaming,
-  InferenceChatCompletionResponse,
   InferenceCompletionParams,
   InferenceCompletionParamsNonStreaming,
   InferenceCompletionParamsStreaming,
-  InferenceCompletionResponse,
   InferenceEmbeddingsParams,
   TokenLogProbs,
 } from './resources/inference';
@@ -407,11 +406,10 @@ export declare namespace LlamaStackClient {
 
   export {
     Inference as Inference,
+    type ChatCompletionResponseStreamChunk as ChatCompletionResponseStreamChunk,
     type CompletionResponse as CompletionResponse,
     type EmbeddingsResponse as EmbeddingsResponse,
     type TokenLogProbs as TokenLogProbs,
-    type InferenceChatCompletionResponse as InferenceChatCompletionResponse,
-    type InferenceCompletionResponse as InferenceCompletionResponse,
     type InferenceChatCompletionParams as InferenceChatCompletionParams,
     type InferenceChatCompletionParamsNonStreaming as InferenceChatCompletionParamsNonStreaming,
     type InferenceChatCompletionParamsStreaming as InferenceChatCompletionParamsStreaming,
@@ -538,6 +536,7 @@ export declare namespace LlamaStackClient {
 
   export type AgentConfig = API.AgentConfig;
   export type BatchCompletion = API.BatchCompletion;
+  export type ChatCompletionResponse = API.ChatCompletionResponse;
   export type CompletionMessage = API.CompletionMessage;
   export type ContentDelta = API.ContentDelta;
   export type Document = API.Document;

src/resources/agents/agents.ts

@@ -15,11 +15,11 @@ import * as StepsAPI from './steps';
 import { StepRetrieveResponse, Steps } from './steps';
 import * as TurnAPI from './turn';
 import {
+  AgentTurnResponseStreamChunk,
   Turn,
   TurnCreateParams,
   TurnCreateParamsNonStreaming,
   TurnCreateParamsStreaming,
-  TurnCreateResponse,
   TurnResource,
   TurnResponseEvent,
   TurnResponseEventPayload,
@@ -145,10 +145,10 @@ export declare namespace Agents {
 
   export {
     TurnResource as TurnResource,
+    type AgentTurnResponseStreamChunk as AgentTurnResponseStreamChunk,
     type Turn as Turn,
     type TurnResponseEvent as TurnResponseEvent,
     type TurnResponseEventPayload as TurnResponseEventPayload,
-    type TurnCreateResponse as TurnCreateResponse,
     type TurnCreateParams as TurnCreateParams,
     type TurnCreateParamsNonStreaming as TurnCreateParamsNonStreaming,
     type TurnCreateParamsStreaming as TurnCreateParamsStreaming,

src/resources/agents/index.ts

@@ -20,10 +20,10 @@ export {
 export { Steps, type StepRetrieveResponse } from './steps';
 export {
   TurnResource,
+  type AgentTurnResponseStreamChunk,
   type Turn,
   type TurnResponseEvent,
   type TurnResponseEventPayload,
-  type TurnCreateResponse,
   type TurnCreateParams,
   type TurnCreateParamsNonStreaming,
   type TurnCreateParamsStreaming,

src/resources/agents/turn.ts

@@ -14,31 +14,30 @@ export class TurnResource extends APIResource {
     sessionId: string,
     body: TurnCreateParamsNonStreaming,
     options?: Core.RequestOptions,
-  ): APIPromise<TurnCreateResponse>;
+  ): APIPromise<Turn>;
   create(
     agentId: string,
     sessionId: string,
     body: TurnCreateParamsStreaming,
     options?: Core.RequestOptions,
-  ): APIPromise<Stream<TurnCreateResponse>>;
+  ): APIPromise<Stream<AgentTurnResponseStreamChunk>>;
   create(
     agentId: string,
     sessionId: string,
     body: TurnCreateParamsBase,
     options?: Core.RequestOptions,
-  ): APIPromise<Stream<TurnCreateResponse> | TurnCreateResponse>;
+  ): APIPromise<Stream<AgentTurnResponseStreamChunk> | Turn>;
   create(
     agentId: string,
     sessionId: string,
     body: TurnCreateParams,
     options?: Core.RequestOptions,
-  ): APIPromise<TurnCreateResponse> | APIPromise<Stream<TurnCreateResponse>> {
+  ): APIPromise<Turn> | APIPromise<Stream<AgentTurnResponseStreamChunk>> {
     return this._client.post(`/v1/agents/${agentId}/session/${sessionId}/turn`, {
       body,
       ...options,
-      headers: { Accept: 'text/event-stream', ...options?.headers },
       stream: body.stream ?? false,
-    }) as APIPromise<TurnCreateResponse> | APIPromise<Stream<TurnCreateResponse>>;
+    }) as APIPromise<Turn> | APIPromise<Stream<AgentTurnResponseStreamChunk>>;
   }
 
   retrieve(
@@ -51,6 +50,10 @@ export class TurnResource extends APIResource {
   }
 }
 
+export interface AgentTurnResponseStreamChunk {
+  event: TurnResponseEvent;
+}
+
 export interface Turn {
   input_messages: Array<Shared.UserMessage | Shared.ToolResponseMessage>;
 
@@ -88,22 +91,44 @@ export namespace Turn {
 
   export namespace OutputAttachment {
     export interface ImageContentItem {
+      /**
+       * Image as a base64 encoded string or an URL
+       */
       image: ImageContentItem.Image;
 
+      /**
+       * Discriminator type of the content item. Always "image"
+       */
       type: 'image';
     }
 
     export namespace ImageContentItem {
+      /**
+       * Image as a base64 encoded string or an URL
+       */
       export interface Image {
+        /**
+         * base64 encoded image data as string
+         */
         data?: string;
 
+        /**
+         * A URL of the image or data URL in the format of data:image/{type};base64,{data}.
+         * Note that URL could have length limits.
+         */
         url?: Shared.URL;
       }
     }
 
     export interface TextContentItem {
+      /**
+       * Text content
+       */
       text: string;
 
+      /**
+       * Discriminator type of the content item. Always "text"
+       */
       type: 'text';
     }
   }
@@ -168,14 +193,6 @@ export namespace TurnResponseEventPayload {
   }
 }
 
-export type TurnCreateResponse = Turn | TurnCreateResponse.AgentTurnResponseStreamChunk;
-
-export namespace TurnCreateResponse {
-  export interface AgentTurnResponseStreamChunk {
-    event: TurnAPI.TurnResponseEvent;
-  }
-}
-
 export type TurnCreateParams = TurnCreateParamsNonStreaming | TurnCreateParamsStreaming;
 
 export interface TurnCreateParamsBase {
@@ -202,22 +219,44 @@ export namespace TurnCreateParams {
 
   export namespace Document {
     export interface ImageContentItem {
+      /**
+       * Image as a base64 encoded string or an URL
+       */
       image: ImageContentItem.Image;
 
+      /**
+       * Discriminator type of the content item. Always "image"
+       */
       type: 'image';
     }
 
     export namespace ImageContentItem {
+      /**
+       * Image as a base64 encoded string or an URL
+       */
       export interface Image {
+        /**
+         * base64 encoded image data as string
+         */
         data?: string;
 
+        /**
+         * A URL of the image or data URL in the format of data:image/{type};base64,{data}.
+         * Note that URL could have length limits.
+         */
         url?: Shared.URL;
       }
     }
 
     export interface TextContentItem {
+      /**
+       * Text content
+       */
       text: string;
 
+      /**
+       * Discriminator type of the content item. Always "text"
+       */
       type: 'text';
     }
   }
@@ -242,10 +281,10 @@ export interface TurnCreateParamsStreaming extends TurnCreateParamsBase {
 
 export declare namespace TurnResource {
   export {
+    type AgentTurnResponseStreamChunk as AgentTurnResponseStreamChunk,
     type Turn as Turn,
     type TurnResponseEvent as TurnResponseEvent,
     type TurnResponseEventPayload as TurnResponseEventPayload,
-    type TurnCreateResponse as TurnCreateResponse,
     type TurnCreateParams as TurnCreateParams,
     type TurnCreateParamsNonStreaming as TurnCreateParamsNonStreaming,
     type TurnCreateParamsStreaming as TurnCreateParamsStreaming,

src/resources/batch-inference.ts

@@ -2,7 +2,6 @@
 
 import { APIResource } from '../resource';
 import * as Core from '../core';
-import * as InferenceAPI from './inference';
 import * as Shared from './shared';
 
 export class BatchInference extends APIResource {
@@ -22,21 +21,7 @@ export class BatchInference extends APIResource {
 }
 
 export interface BatchInferenceChatCompletionResponse {
-  batch: Array<BatchInferenceChatCompletionResponse.Batch>;
-}
-
-export namespace BatchInferenceChatCompletionResponse {
-  export interface Batch {
-    /**
-     * The complete response message
-     */
-    completion_message: Shared.CompletionMessage;
-
-    /**
-     * Optional log probabilities for generated tokens
-     */
-    logprobs?: Array<InferenceAPI.TokenLogProbs>;
-  }
+  batch: Array<Shared.ChatCompletionResponse>;
 }
 
 export interface BatchInferenceChatCompletionParams {

src/resources/index.ts

@@ -48,11 +48,10 @@ export {
 } from './eval-tasks';
 export {
   Inference,
+  type ChatCompletionResponseStreamChunk,
   type CompletionResponse,
   type EmbeddingsResponse,
   type TokenLogProbs,
-  type InferenceChatCompletionResponse,
-  type InferenceCompletionResponse,
   type InferenceChatCompletionParams,
   type InferenceChatCompletionParamsNonStreaming,
   type InferenceChatCompletionParamsStreaming,