@upstash/qstash 2.5.4 → 2.5.5

package/index.d.mts

@@ -75,6 +75,107 @@ type WithCursor<T> = T & {
 };
 type BodyInit = Blob | FormData | URLSearchParams | ReadableStream<Uint8Array> | string;
 type HeadersInit = Headers | Record<string, string> | [string, string][] | IterableIterator<[string, string]>;
+type RequestOptions = RequestInit & {
+    backend?: string;
+};
+type ChatRateLimit = {
+    "limit-requests": string | null;
+    "limit-tokens": string | null;
+    "remaining-requests": string | null;
+    "remaining-tokens": string | null;
+    "reset-requests": string | null;
+    "reset-tokens": string | null;
+};
+type RateLimit = {
+    limit: string | null;
+    remaining: string | null;
+    reset: string | null;
+};
+
+type ChatCompletionMessage = {
+    role: "system" | "assistant" | "user";
+    content: string;
+};
+type ChatModel = "meta-llama/Meta-Llama-3-8B-Instruct" | "mistralai/Mistral-7B-Instruct-v0.2";
+type ChatResponseFormat = {
+    type: "text" | "json_object";
+};
+type TopLogprob = {
+    token: string;
+    bytes: number[];
+    logprob: number;
+};
+type ChatCompletionTokenLogprob = {
+    token: string;
+    bytes: number[];
+    logprob: number;
+    top_logprobs: TopLogprob[];
+};
+type ChoiceLogprobs = {
+    content: ChatCompletionTokenLogprob[];
+};
+type Choice = {
+    finish_reason: "stop" | "length";
+    index: number;
+    logprobs: ChoiceLogprobs;
+    message: ChatCompletionMessage;
+};
+type CompletionUsage = {
+    completion_tokens: number;
+    prompt_tokens: number;
+    total_tokens: number;
+};
+type ChatCompletion = {
+    id: string;
+    choices: Choice[];
+    created: number;
+    model: string;
+    object: "chat.completion";
+    system_fingerprint: string;
+    usage: CompletionUsage;
+};
+type ChunkChoice = {
+    delta: ChatCompletionMessage;
+    finish_reason: "stop" | "length";
+    index: number;
+    logprobs: ChoiceLogprobs;
+};
+type ChatCompletionChunk = {
+    id: string;
+    choices: ChunkChoice[];
+    created: number;
+    model: string;
+    object: "chat.completion.chunk";
+    system_fingerprint: string;
+    usage: CompletionUsage;
+};
+type StreamEnabled = {
+    stream: true;
+};
+type StreamDisabled = {
+    stream: false;
+} | object;
+type StreamParameter = StreamEnabled | StreamDisabled;
+type ChatRequest<TStream extends StreamParameter> = {
+    messages: ChatCompletionMessage[];
+    model: ChatModel;
+    frequency_penalty?: number;
+    logit_bias?: Record<string, number>;
+    logprobs?: boolean;
+    top_logprobs?: number;
+    max_tokens?: number;
+    n?: number;
+    presence_penalty?: number;
+    response_format?: ChatResponseFormat;
+    seed?: number;
+    stop?: string | string[];
+    temperature?: number;
+    top_p?: number;
+} & TStream;
+type PromptRequest<TStream extends StreamParameter> = Omit<ChatRequest<TStream>, "messages" | "stream"> & {
+    system: string;
+    user: string;
+} & TStream;
 
 type UpstashRequest = {
     /**
@@ -111,6 +212,7 @@ type UpstashResponse<TResult> = TResult & {
 };
 type Requester = {
     request: <TResult = unknown>(request: UpstashRequest) => Promise<UpstashResponse<TResult>>;
+    requestStream: (request: UpstashRequest) => AsyncIterable<ChatCompletionChunk>;
 };
 type RetryConfig = false | {
     /**
@@ -259,7 +361,7 @@ declare class Queue {
     /**
      * Enqueue a message to a queue, serializing the body to JSON.
      */
-    enqueueJSON<TBody = unknown>(request: PublishRequest<TBody>): Promise<PublishResponse<PublishRequest<TBody>>>;
+    enqueueJSON<TBody = unknown, TRequest extends PublishRequest<TBody> = PublishRequest<TBody>>(request: TRequest): Promise<PublishResponse<TRequest>>;
 }
 
 type Schedule = {
@@ -438,6 +540,34 @@ declare class Topics {
     delete(name: string): Promise<void>;
 }
 
+declare class Chat {
+    private http;
+    constructor(http: Requester);
+    private static toChatRequest;
+    /**
+     * Calls the Upstash completions api given a ChatRequest.
+     *
+     * Returns a ChatCompletion or a stream of ChatCompletionChunks
+     * if stream is enabled.
+     *
+     * @param request ChatRequest with messages
+     * @returns Chat completion or stream
+     */
+    create: <TStream extends StreamParameter>(request: ChatRequest<TStream>) => Promise<TStream extends StreamEnabled ? AsyncIterable<ChatCompletionChunk> : ChatCompletion>;
+    /**
+     * Calls the Upstash completions api given a PromptRequest.
+     *
+     * Returns a ChatCompletion or a stream of ChatCompletionChunks
+     * if stream is enabled.
+     *
+     * @param request PromptRequest with system and user messages.
+     *    Note that system parameter shouldn't be passed in the case of
+     *    mistralai/Mistral-7B-Instruct-v0.2 model.
+     * @returns Chat completion or stream
+     */
+    prompt: <TStream extends StreamParameter>(request: PromptRequest<TStream>) => Promise<TStream extends StreamEnabled ? AsyncIterable<ChatCompletionChunk> : ChatCompletion>;
+}
+
 type ClientConfig = {
     /**
      * Url of the qstash api server.
@@ -560,12 +690,21 @@ type PublishRequest<TBody = BodyInit> = {
      */
     url: string;
     topic?: never;
+    api?: never;
 } | {
     url?: never;
     /**
-     * The url where the message should be sent to.
+     * The topic the message should be sent to.
      */
     topic: string;
+    api?: never;
+} | {
+    url?: never;
+    topic?: never;
+    /**
+     * The api endpoint the request should be sent to.
+     */
+    api: "llm";
 });
 type PublishJsonRequest = Omit<PublishRequest, "body"> & {
     /**
@@ -629,6 +768,12 @@ declare class Client {
      * Create, read, update or delete queues.
      */
     queue(request?: QueueRequest): Queue;
+    /**
+     * Access the Chat API
+     *
+     * Call the create or prompt methods
+     */
+    chat(): Chat;
     publish<TRequest extends PublishRequest>(request: TRequest): Promise<PublishResponse<TRequest>>;
     /**
      * publishJSON is a utility wrapper around `publish` that automatically serializes the body
@@ -664,15 +809,19 @@ declare class Client {
      */
     events(request?: EventsRequest): Promise<GetEventsResponse>;
 }
-type PublishToUrlResponse = {
+type PublishToApiResponse = {
     messageId: string;
+};
+type PublishToUrlResponse = PublishToApiResponse & {
     url: string;
     deduplicated?: boolean;
 };
 type PublishToTopicResponse = PublishToUrlResponse[];
-type PublishResponse<R> = R extends {
+type PublishResponse<TRequest> = TRequest extends {
     url: string;
-} ? PublishToUrlResponse : PublishToTopicResponse;
+} ? PublishToUrlResponse : TRequest extends {
+    topic: string;
+} ? PublishToTopicResponse : PublishToApiResponse;
 
 /**
  * Result of 500 Internal Server Error
@@ -681,7 +830,10 @@ declare class QstashError extends Error {
     constructor(message: string);
 }
 declare class QstashRatelimitError extends QstashError {
-    constructor(args: unknown);
+    constructor(args: RateLimit);
+}
+declare class QstashChatRatelimitError extends QstashError {
+    constructor(args: ChatRateLimit);
 }
 
-export { type AddEndpointsRequest, type BodyInit, Client, type CreateScheduleRequest, type Endpoint, type Event, type EventsRequest, type GetEventsResponse, type HeadersInit, type Message, Messages, type PublishBatchRequest, type PublishJsonRequest, type PublishRequest, type PublishResponse, type PublishToTopicResponse, type PublishToUrlResponse, QstashError, QstashRatelimitError, type QueueRequest, Receiver, type ReceiverConfig, type RemoveEndpointsRequest, type Schedule, Schedules, SignatureError, type State, type Topic, Topics, type VerifyRequest, type WithCursor };
+export { type AddEndpointsRequest, type BodyInit, Chat, type ChatCompletion, type ChatCompletionChunk, type ChatCompletionMessage, type ChatRateLimit, type ChatRequest, Client, type CreateScheduleRequest, type Endpoint, type Event, type EventsRequest, type GetEventsResponse, type HeadersInit, type Message, Messages, type PromptRequest, type PublishBatchRequest, type PublishJsonRequest, type PublishRequest, type PublishResponse, type PublishToApiResponse, type PublishToTopicResponse, type PublishToUrlResponse, QstashChatRatelimitError, QstashError, QstashRatelimitError, type QueueRequest, type RateLimit, Receiver, type ReceiverConfig, type RemoveEndpointsRequest, type RequestOptions, type Schedule, Schedules, SignatureError, type State, type StreamDisabled, type StreamEnabled, type StreamParameter, type Topic, Topics, type VerifyRequest, type WithCursor };

package/index.d.ts

@@ -75,6 +75,107 @@ type WithCursor<T> = T & {
 };
 type BodyInit = Blob | FormData | URLSearchParams | ReadableStream<Uint8Array> | string;
 type HeadersInit = Headers | Record<string, string> | [string, string][] | IterableIterator<[string, string]>;
+type RequestOptions = RequestInit & {
+    backend?: string;
+};
+type ChatRateLimit = {
+    "limit-requests": string | null;
+    "limit-tokens": string | null;
+    "remaining-requests": string | null;
+    "remaining-tokens": string | null;
+    "reset-requests": string | null;
+    "reset-tokens": string | null;
+};
+type RateLimit = {
+    limit: string | null;
+    remaining: string | null;
+    reset: string | null;
+};
+
+type ChatCompletionMessage = {
+    role: "system" | "assistant" | "user";
+    content: string;
+};
+type ChatModel = "meta-llama/Meta-Llama-3-8B-Instruct" | "mistralai/Mistral-7B-Instruct-v0.2";
+type ChatResponseFormat = {
+    type: "text" | "json_object";
+};
+type TopLogprob = {
+    token: string;
+    bytes: number[];
+    logprob: number;
+};
+type ChatCompletionTokenLogprob = {
+    token: string;
+    bytes: number[];
+    logprob: number;
+    top_logprobs: TopLogprob[];
+};
+type ChoiceLogprobs = {
+    content: ChatCompletionTokenLogprob[];
+};
+type Choice = {
+    finish_reason: "stop" | "length";
+    index: number;
+    logprobs: ChoiceLogprobs;
+    message: ChatCompletionMessage;
+};
+type CompletionUsage = {
+    completion_tokens: number;
+    prompt_tokens: number;
+    total_tokens: number;
+};
+type ChatCompletion = {
+    id: string;
+    choices: Choice[];
+    created: number;
+    model: string;
+    object: "chat.completion";
+    system_fingerprint: string;
+    usage: CompletionUsage;
+};
+type ChunkChoice = {
+    delta: ChatCompletionMessage;
+    finish_reason: "stop" | "length";
+    index: number;
+    logprobs: ChoiceLogprobs;
+};
+type ChatCompletionChunk = {
+    id: string;
+    choices: ChunkChoice[];
+    created: number;
+    model: string;
+    object: "chat.completion.chunk";
+    system_fingerprint: string;
+    usage: CompletionUsage;
+};
+type StreamEnabled = {
+    stream: true;
+};
+type StreamDisabled = {
+    stream: false;
+} | object;
+type StreamParameter = StreamEnabled | StreamDisabled;
+type ChatRequest<TStream extends StreamParameter> = {
+    messages: ChatCompletionMessage[];
+    model: ChatModel;
+    frequency_penalty?: number;
+    logit_bias?: Record<string, number>;
+    logprobs?: boolean;
+    top_logprobs?: number;
+    max_tokens?: number;
+    n?: number;
+    presence_penalty?: number;
+    response_format?: ChatResponseFormat;
+    seed?: number;
+    stop?: string | string[];
+    temperature?: number;
+    top_p?: number;
+} & TStream;
+type PromptRequest<TStream extends StreamParameter> = Omit<ChatRequest<TStream>, "messages" | "stream"> & {
+    system: string;
+    user: string;
+} & TStream;
 
 type UpstashRequest = {
     /**
@@ -111,6 +212,7 @@ type UpstashResponse<TResult> = TResult & {
 };
 type Requester = {
     request: <TResult = unknown>(request: UpstashRequest) => Promise<UpstashResponse<TResult>>;
+    requestStream: (request: UpstashRequest) => AsyncIterable<ChatCompletionChunk>;
 };
 type RetryConfig = false | {
     /**
@@ -259,7 +361,7 @@ declare class Queue {
     /**
      * Enqueue a message to a queue, serializing the body to JSON.
      */
-    enqueueJSON<TBody = unknown>(request: PublishRequest<TBody>): Promise<PublishResponse<PublishRequest<TBody>>>;
+    enqueueJSON<TBody = unknown, TRequest extends PublishRequest<TBody> = PublishRequest<TBody>>(request: TRequest): Promise<PublishResponse<TRequest>>;
 }
 
 type Schedule = {
@@ -438,6 +540,34 @@ declare class Topics {
     delete(name: string): Promise<void>;
 }
 
+declare class Chat {
+    private http;
+    constructor(http: Requester);
+    private static toChatRequest;
+    /**
+     * Calls the Upstash completions api given a ChatRequest.
+     *
+     * Returns a ChatCompletion or a stream of ChatCompletionChunks
+     * if stream is enabled.
+     *
+     * @param request ChatRequest with messages
+     * @returns Chat completion or stream
+     */
+    create: <TStream extends StreamParameter>(request: ChatRequest<TStream>) => Promise<TStream extends StreamEnabled ? AsyncIterable<ChatCompletionChunk> : ChatCompletion>;
+    /**
+     * Calls the Upstash completions api given a PromptRequest.
+     *
+     * Returns a ChatCompletion or a stream of ChatCompletionChunks
+     * if stream is enabled.
+     *
+     * @param request PromptRequest with system and user messages.
+     *    Note that system parameter shouldn't be passed in the case of
+     *    mistralai/Mistral-7B-Instruct-v0.2 model.
+     * @returns Chat completion or stream
+     */
+    prompt: <TStream extends StreamParameter>(request: PromptRequest<TStream>) => Promise<TStream extends StreamEnabled ? AsyncIterable<ChatCompletionChunk> : ChatCompletion>;
+}
+
 type ClientConfig = {
     /**
      * Url of the qstash api server.
@@ -560,12 +690,21 @@ type PublishRequest<TBody = BodyInit> = {
      */
     url: string;
     topic?: never;
+    api?: never;
 } | {
     url?: never;
     /**
-     * The url where the message should be sent to.
+     * The topic the message should be sent to.
      */
     topic: string;
+    api?: never;
+} | {
+    url?: never;
+    topic?: never;
+    /**
+     * The api endpoint the request should be sent to.
+     */
+    api: "llm";
 });
 type PublishJsonRequest = Omit<PublishRequest, "body"> & {
     /**
@@ -629,6 +768,12 @@ declare class Client {
      * Create, read, update or delete queues.
      */
     queue(request?: QueueRequest): Queue;
+    /**
+     * Access the Chat API
+     *
+     * Call the create or prompt methods
+     */
+    chat(): Chat;
     publish<TRequest extends PublishRequest>(request: TRequest): Promise<PublishResponse<TRequest>>;
     /**
      * publishJSON is a utility wrapper around `publish` that automatically serializes the body
@@ -664,15 +809,19 @@ declare class Client {
      */
     events(request?: EventsRequest): Promise<GetEventsResponse>;
 }
-type PublishToUrlResponse = {
+type PublishToApiResponse = {
     messageId: string;
+};
+type PublishToUrlResponse = PublishToApiResponse & {
     url: string;
     deduplicated?: boolean;
 };
 type PublishToTopicResponse = PublishToUrlResponse[];
-type PublishResponse<R> = R extends {
+type PublishResponse<TRequest> = TRequest extends {
     url: string;
-} ? PublishToUrlResponse : PublishToTopicResponse;
+} ? PublishToUrlResponse : TRequest extends {
+    topic: string;
+} ? PublishToTopicResponse : PublishToApiResponse;
 
 /**
  * Result of 500 Internal Server Error
@@ -681,7 +830,10 @@ declare class QstashError extends Error {
     constructor(message: string);
 }
 declare class QstashRatelimitError extends QstashError {
-    constructor(args: unknown);
+    constructor(args: RateLimit);
+}
+declare class QstashChatRatelimitError extends QstashError {
+    constructor(args: ChatRateLimit);
 }
 
-export { type AddEndpointsRequest, type BodyInit, Client, type CreateScheduleRequest, type Endpoint, type Event, type EventsRequest, type GetEventsResponse, type HeadersInit, type Message, Messages, type PublishBatchRequest, type PublishJsonRequest, type PublishRequest, type PublishResponse, type PublishToTopicResponse, type PublishToUrlResponse, QstashError, QstashRatelimitError, type QueueRequest, Receiver, type ReceiverConfig, type RemoveEndpointsRequest, type Schedule, Schedules, SignatureError, type State, type Topic, Topics, type VerifyRequest, type WithCursor };
+export { type AddEndpointsRequest, type BodyInit, Chat, type ChatCompletion, type ChatCompletionChunk, type ChatCompletionMessage, type ChatRateLimit, type ChatRequest, Client, type CreateScheduleRequest, type Endpoint, type Event, type EventsRequest, type GetEventsResponse, type HeadersInit, type Message, Messages, type PromptRequest, type PublishBatchRequest, type PublishJsonRequest, type PublishRequest, type PublishResponse, type PublishToApiResponse, type PublishToTopicResponse, type PublishToUrlResponse, QstashChatRatelimitError, QstashError, QstashRatelimitError, type QueueRequest, type RateLimit, Receiver, type ReceiverConfig, type RemoveEndpointsRequest, type RequestOptions, type Schedule, Schedules, SignatureError, type State, type StreamDisabled, type StreamEnabled, type StreamParameter, type Topic, Topics, type VerifyRequest, type WithCursor };

package/index.js

@@ -1,4 +1,4 @@
-"use strict";Object.defineProperty(exports, "__esModule", {value: true}); function _nullishCoalesce(lhs, rhsFn) { if (lhs != null) { return lhs; } else { return rhsFn(); } } function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; }
+"use strict";Object.defineProperty(exports, "__esModule", {value: true}); function _nullishCoalesce(lhs, rhsFn) { if (lhs != null) { return lhs; } else { return rhsFn(); } } function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; } var _class; var _class2;
 
 
 var _chunkUUR7N6E6js = require('./chunk-UUR7N6E6.js');
@@ -55,14 +55,19 @@ var QstashRatelimitError = class extends QstashError {
     super(`You have been ratelimited. ${JSON.stringify(args)} `);
   }
 };
+var QstashChatRatelimitError = class extends QstashError {
+  constructor(args) {
+    super(`You have been ratelimited. ${JSON.stringify(args)} `);
+  }
+};
 
 // src/client/http.ts
-var HttpClient = class {
+var HttpClient = (_class = class {
   
   
   
   
-  constructor(config) {
+  constructor(config) {;_class.prototype.__init.call(this);_class.prototype.__init2.call(this);
     this.baseUrl = config.baseUrl.replace(/\/$/, "");
     this.authorization = config.authorization;
     this.retry = // eslint-disable-next-line @typescript-eslint/no-unnecessary-condition
@@ -75,22 +80,44 @@ var HttpClient = class {
     };
   }
   async request(request) {
-    const headers = new Headers(request.headers);
-    headers.set("Authorization", this.authorization);
-    const requestOptions = {
-      method: request.method,
-      headers,
-      body: request.body,
-      keepalive: request.keepalive
-    };
-    const url = new URL([this.baseUrl, ...request.path].join("/"));
-    if (request.query) {
-      for (const [key, value] of Object.entries(request.query)) {
-        if (value !== void 0) {
-          url.searchParams.set(key, value.toString());
+    const { response } = await this.requestWithBackoff(request);
+    if (request.parseResponseAsJson === false) {
+      return void 0;
+    }
+    return await response.json();
+  }
+  async *requestStream(request) {
+    const { response } = await this.requestWithBackoff(request);
+    if (!response.body) {
+      throw new Error("No response body");
+    }
+    const body = response.body;
+    const reader = body.getReader();
+    const decoder = new TextDecoder();
+    try {
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) {
+          break;
+        }
+        const chunkText = decoder.decode(value, { stream: true });
+        const chunks = chunkText.split("\n").filter(Boolean);
+        for (const chunk of chunks) {
+          if (chunk.startsWith("data: ")) {
+            const data = chunk.slice(6);
+            if (data === "[DONE]") {
+              break;
+            }
+            yield JSON.parse(data);
+          }
         }
       }
+    } finally {
+      await reader.cancel();
     }
+  }
+  __init() {this.requestWithBackoff = async (request) => {
+    const [url, requestOptions] = this.processRequest(request);
     let response = void 0;
     let error = void 0;
     for (let index = 0; index < this.retry.attempts; index++) {
@@ -105,7 +132,43 @@ var HttpClient = class {
     if (!response) {
       throw _nullishCoalesce(error, () => ( new Error("Exhausted all retries")));
     }
+    await this.checkResponse(response);
+    return {
+      response,
+      error
+    };
+  }}
+  __init2() {this.processRequest = (request) => {
+    const headers = new Headers(request.headers);
+    headers.set("Authorization", this.authorization);
+    const requestOptions = {
+      method: request.method,
+      headers,
+      body: request.body,
+      keepalive: request.keepalive
+    };
+    const url = new URL([this.baseUrl, ...request.path].join("/"));
+    if (request.query) {
+      for (const [key, value] of Object.entries(request.query)) {
+        if (value !== void 0) {
+          url.searchParams.set(key, value.toString());
+        }
+      }
+    }
+    return [url.toString(), requestOptions];
+  }}
+  async checkResponse(response) {
     if (response.status === 429) {
+      if (response.headers.get("x-ratelimit-limit-requests")) {
+        throw new QstashChatRatelimitError({
+          "limit-requests": response.headers.get("x-ratelimit-limit-requests"),
+          "limit-tokens": response.headers.get("x-ratelimit-limit-tokens"),
+          "remaining-requests": response.headers.get("x-ratelimit-remaining-requests"),
+          "remaining-tokens": response.headers.get("x-ratelimit-remaining-tokens"),
+          "reset-requests": response.headers.get("x-ratelimit-reset-requests"),
+          "reset-tokens": response.headers.get("x-ratelimit-reset-tokens")
+        });
+      }
       throw new QstashRatelimitError({
         limit: response.headers.get("Burst-RateLimit-Limit"),
         remaining: response.headers.get("Burst-RateLimit-Remaining"),
@@ -116,12 +179,8 @@ var HttpClient = class {
       const body = await response.text();
       throw new QstashError(body.length > 0 ? body : `Error: status=${response.status}`);
     }
-    if (request.parseResponseAsJson === false) {
-      return void 0;
-    }
-    return await response.json();
   }
-};
+}, _class);
 
 // src/client/messages.ts
 var Messages = class {
@@ -192,6 +251,9 @@ function processHeaders(request) {
   }
   return headers;
 }
+function getRequestPath(request) {
+  return _nullishCoalesce(_nullishCoalesce(request.url, () => ( request.topic)), () => ( `api/${request.api}`));
+}
 
 // src/client/queue.ts
 var Queue = class {
@@ -264,7 +326,7 @@ var Queue = class {
       throw new Error("Please provide a queue name to the Queue constructor");
     }
     const headers = processHeaders(request);
-    const destination = _nullishCoalesce(request.url, () => ( request.topic));
+    const destination = getRequestPath(request);
     const response = await this.http.request({
       path: ["v2", "enqueue", this.queueName, destination],
       body: request.body,
@@ -415,6 +477,69 @@ var Topics = class {
   }
 };
 
+// src/client/llm/chat.ts
+var Chat = (_class2 = class _Chat {
+  
+  constructor(http) {;_class2.prototype.__init3.call(this);_class2.prototype.__init4.call(this);
+    this.http = http;
+  }
+  static toChatRequest(request) {
+    const messages = [];
+    messages.push(
+      { role: "system", content: request.system },
+      { role: "user", content: request.user }
+    );
+    const chatRequest = { ...request, messages };
+    return chatRequest;
+  }
+  /**
+   * Calls the Upstash completions api given a ChatRequest.
+   *
+   * Returns a ChatCompletion or a stream of ChatCompletionChunks
+   * if stream is enabled.
+   *
+   * @param request ChatRequest with messages
+   * @returns Chat completion or stream
+   */
+  __init3() {this.create = async (request) => {
+    const body = JSON.stringify(request);
+    if ("stream" in request && request.stream) {
+      return this.http.requestStream({
+        path: ["llm", "v1", "chat", "completions"],
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          Connection: "keep-alive",
+          Accept: "text/event-stream",
+          "Cache-Control": "no-cache"
+        },
+        body
+      });
+    }
+    return this.http.request({
+      path: ["llm", "v1", "chat", "completions"],
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body
+    });
+  }}
+  /**
+   * Calls the Upstash completions api given a PromptRequest.
+   *
+   * Returns a ChatCompletion or a stream of ChatCompletionChunks
+   * if stream is enabled.
+   *
+   * @param request PromptRequest with system and user messages.
+   *    Note that system parameter shouldn't be passed in the case of
+   *    mistralai/Mistral-7B-Instruct-v0.2 model.
+   * @returns Chat completion or stream
+   */
+  __init4() {this.prompt = async (request) => {
+    const chatRequest = _Chat.toChatRequest(request);
+    return this.create(chatRequest);
+  }}
+}, _class2);
+
 // src/client/client.ts
 var Client = class {
   
@@ -465,10 +590,18 @@ var Client = class {
   queue(request) {
     return new Queue(this.http, _optionalChain([request, 'optionalAccess', _6 => _6.queueName]));
   }
+  /**
+   * Access the Chat API
+   *
+   * Call the create or prompt methods
+   */
+  chat() {
+    return new Chat(this.http);
+  }
   async publish(request) {
     const headers = processHeaders(request);
     const response = await this.http.request({
-      path: ["v2", "publish", _nullishCoalesce(request.url, () => ( request.topic))],
+      path: ["v2", "publish", getRequestPath(request)],
       body: request.body,
       headers,
       method: "POST"
@@ -498,7 +631,7 @@ var Client = class {
       const headers = processHeaders(message);
       const headerEntries = Object.fromEntries(headers.entries());
       messages.push({
-        destination: _nullishCoalesce(message.url, () => ( message.topic)),
+        destination: getRequestPath(message),
         headers: headerEntries,
         body: message.body,
         ...message.queueName && { queue: message.queueName }
@@ -577,4 +710,6 @@ var Client = class {
 
 
 
-exports.Client = Client; exports.Messages = Messages; exports.QstashError = QstashError; exports.QstashRatelimitError = QstashRatelimitError; exports.Receiver = _chunkUUR7N6E6js.Receiver; exports.Schedules = Schedules; exports.SignatureError = _chunkUUR7N6E6js.SignatureError; exports.Topics = Topics;
+
+
+exports.Chat = Chat; exports.Client = Client; exports.Messages = Messages; exports.QstashChatRatelimitError = QstashChatRatelimitError; exports.QstashError = QstashError; exports.QstashRatelimitError = QstashRatelimitError; exports.Receiver = _chunkUUR7N6E6js.Receiver; exports.Schedules = Schedules; exports.SignatureError = _chunkUUR7N6E6js.SignatureError; exports.Topics = Topics;

package/index.mjs

@@ -55,6 +55,11 @@ var QstashRatelimitError = class extends QstashError {
     super(`You have been ratelimited. ${JSON.stringify(args)} `);
   }
 };
+var QstashChatRatelimitError = class extends QstashError {
+  constructor(args) {
+    super(`You have been ratelimited. ${JSON.stringify(args)} `);
+  }
+};
 
 // src/client/http.ts
 var HttpClient = class {
@@ -75,22 +80,44 @@ var HttpClient = class {
     };
   }
   async request(request) {
-    const headers = new Headers(request.headers);
-    headers.set("Authorization", this.authorization);
-    const requestOptions = {
-      method: request.method,
-      headers,
-      body: request.body,
-      keepalive: request.keepalive
-    };
-    const url = new URL([this.baseUrl, ...request.path].join("/"));
-    if (request.query) {
-      for (const [key, value] of Object.entries(request.query)) {
-        if (value !== void 0) {
-          url.searchParams.set(key, value.toString());
+    const { response } = await this.requestWithBackoff(request);
+    if (request.parseResponseAsJson === false) {
+      return void 0;
+    }
+    return await response.json();
+  }
+  async *requestStream(request) {
+    const { response } = await this.requestWithBackoff(request);
+    if (!response.body) {
+      throw new Error("No response body");
+    }
+    const body = response.body;
+    const reader = body.getReader();
+    const decoder = new TextDecoder();
+    try {
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) {
+          break;
+        }
+        const chunkText = decoder.decode(value, { stream: true });
+        const chunks = chunkText.split("\n").filter(Boolean);
+        for (const chunk of chunks) {
+          if (chunk.startsWith("data: ")) {
+            const data = chunk.slice(6);
+            if (data === "[DONE]") {
+              break;
+            }
+            yield JSON.parse(data);
+          }
         }
       }
+    } finally {
+      await reader.cancel();
     }
+  }
+  requestWithBackoff = async (request) => {
+    const [url, requestOptions] = this.processRequest(request);
     let response = void 0;
     let error = void 0;
     for (let index = 0; index < this.retry.attempts; index++) {
@@ -105,7 +132,43 @@ var HttpClient = class {
     if (!response) {
       throw error ?? new Error("Exhausted all retries");
     }
+    await this.checkResponse(response);
+    return {
+      response,
+      error
+    };
+  };
+  processRequest = (request) => {
+    const headers = new Headers(request.headers);
+    headers.set("Authorization", this.authorization);
+    const requestOptions = {
+      method: request.method,
+      headers,
+      body: request.body,
+      keepalive: request.keepalive
+    };
+    const url = new URL([this.baseUrl, ...request.path].join("/"));
+    if (request.query) {
+      for (const [key, value] of Object.entries(request.query)) {
+        if (value !== void 0) {
+          url.searchParams.set(key, value.toString());
+        }
+      }
+    }
+    return [url.toString(), requestOptions];
+  };
+  async checkResponse(response) {
     if (response.status === 429) {
+      if (response.headers.get("x-ratelimit-limit-requests")) {
+        throw new QstashChatRatelimitError({
+          "limit-requests": response.headers.get("x-ratelimit-limit-requests"),
+          "limit-tokens": response.headers.get("x-ratelimit-limit-tokens"),
+          "remaining-requests": response.headers.get("x-ratelimit-remaining-requests"),
+          "remaining-tokens": response.headers.get("x-ratelimit-remaining-tokens"),
+          "reset-requests": response.headers.get("x-ratelimit-reset-requests"),
+          "reset-tokens": response.headers.get("x-ratelimit-reset-tokens")
+        });
+      }
       throw new QstashRatelimitError({
         limit: response.headers.get("Burst-RateLimit-Limit"),
         remaining: response.headers.get("Burst-RateLimit-Remaining"),
@@ -116,10 +179,6 @@ var HttpClient = class {
       const body = await response.text();
       throw new QstashError(body.length > 0 ? body : `Error: status=${response.status}`);
     }
-    if (request.parseResponseAsJson === false) {
-      return void 0;
-    }
-    return await response.json();
   }
 };
 
@@ -192,6 +251,9 @@ function processHeaders(request) {
   }
   return headers;
 }
+function getRequestPath(request) {
+  return request.url ?? request.topic ?? `api/${request.api}`;
+}
 
 // src/client/queue.ts
 var Queue = class {
@@ -264,7 +326,7 @@ var Queue = class {
       throw new Error("Please provide a queue name to the Queue constructor");
     }
     const headers = processHeaders(request);
-    const destination = request.url ?? request.topic;
+    const destination = getRequestPath(request);
     const response = await this.http.request({
       path: ["v2", "enqueue", this.queueName, destination],
       body: request.body,
@@ -415,6 +477,69 @@ var Topics = class {
   }
 };
 
+// src/client/llm/chat.ts
+var Chat = class _Chat {
+  http;
+  constructor(http) {
+    this.http = http;
+  }
+  static toChatRequest(request) {
+    const messages = [];
+    messages.push(
+      { role: "system", content: request.system },
+      { role: "user", content: request.user }
+    );
+    const chatRequest = { ...request, messages };
+    return chatRequest;
+  }
+  /**
+   * Calls the Upstash completions api given a ChatRequest.
+   *
+   * Returns a ChatCompletion or a stream of ChatCompletionChunks
+   * if stream is enabled.
+   *
+   * @param request ChatRequest with messages
+   * @returns Chat completion or stream
+   */
+  create = async (request) => {
+    const body = JSON.stringify(request);
+    if ("stream" in request && request.stream) {
+      return this.http.requestStream({
+        path: ["llm", "v1", "chat", "completions"],
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          Connection: "keep-alive",
+          Accept: "text/event-stream",
+          "Cache-Control": "no-cache"
+        },
+        body
+      });
+    }
+    return this.http.request({
+      path: ["llm", "v1", "chat", "completions"],
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body
+    });
+  };
+  /**
+   * Calls the Upstash completions api given a PromptRequest.
+   *
+   * Returns a ChatCompletion or a stream of ChatCompletionChunks
+   * if stream is enabled.
+   *
+   * @param request PromptRequest with system and user messages.
+   *    Note that system parameter shouldn't be passed in the case of
+   *    mistralai/Mistral-7B-Instruct-v0.2 model.
+   * @returns Chat completion or stream
+   */
+  prompt = async (request) => {
+    const chatRequest = _Chat.toChatRequest(request);
+    return this.create(chatRequest);
+  };
+};
+
 // src/client/client.ts
 var Client = class {
   http;
@@ -465,10 +590,18 @@ var Client = class {
   queue(request) {
     return new Queue(this.http, request?.queueName);
   }
+  /**
+   * Access the Chat API
+   *
+   * Call the create or prompt methods
+   */
+  chat() {
+    return new Chat(this.http);
+  }
   async publish(request) {
     const headers = processHeaders(request);
     const response = await this.http.request({
-      path: ["v2", "publish", request.url ?? request.topic],
+      path: ["v2", "publish", getRequestPath(request)],
       body: request.body,
       headers,
       method: "POST"
@@ -498,7 +631,7 @@ var Client = class {
       const headers = processHeaders(message);
       const headerEntries = Object.fromEntries(headers.entries());
       messages.push({
-        destination: message.url ?? message.topic,
+        destination: getRequestPath(message),
         headers: headerEntries,
         body: message.body,
         ...message.queueName && { queue: message.queueName }
@@ -569,8 +702,10 @@ var Client = class {
   }
 };
 export {
+  Chat,
   Client,
   Messages,
+  QstashChatRatelimitError,
   QstashError,
   QstashRatelimitError,
   Receiver,

package/package.json

@@ -1 +1 @@
-{"version":"v2.5.4","name":"@upstash/qstash","description":"Official Typescript client for QStash","author":"Andreas Thomas <dev@chronark.com>","license":"MIT","homepage":"https://github.com/upstash/sdk-qstash-ts#readme","repository":{"type":"git","url":"git+https://github.com/upstash/sdk-qstash-ts.git"},"bugs":{"url":"https://github.com/upstash/sdk-qstash-ts/issues"},"main":"./index.js","module":"./index.mjs","types":"./index.d.ts","files":["./**"],"exports":{".":{"import":"./index.mjs","require":"./index.js"},"./nextjs":{"import":"./nextjs.mjs","require":"./nextjs.js"},"./dist/nextjs":{"import":"./nextjs.mjs","require":"./nextjs.js"}},"typesVersions":{"*":{"nextjs":["./nextjs.d.ts"]}},"keywords":["qstash","queue","events","serverless","upstash"],"scripts":{"build":"tsup && cp README.md ./dist/ && cp package.json ./dist/ && cp LICENSE ./dist/","test":"bun test","fmt":"prettier --write .","lint":"tsc && eslint \"src/**/*.{js,ts,tsx}\" --quiet --fix"},"devDependencies":{"@commitlint/cli":"^19.2.2","@commitlint/config-conventional":"^19.2.2","@types/bun":"^1.1.1","@types/crypto-js":"^4.2.0","@typescript-eslint/eslint-plugin":"^7.0.1","@typescript-eslint/parser":"^7.0.1","bun-types":"^1.1.7","eslint":"^8","eslint-plugin-unicorn":"^51.0.1","husky":"^9.0.10","next":"^14.0.2","prettier":"^3.2.5","tsup":"latest","typescript":"^5.4.5","undici-types":"^6.16.0","vitest":"latest"},"dependencies":{"crypto-js":">=4.2.0","jose":"^ 5.2.3"}}
+{"version":"v2.5.5","name":"@upstash/qstash","description":"Official Typescript client for QStash","author":"Andreas Thomas <dev@chronark.com>","license":"MIT","homepage":"https://github.com/upstash/sdk-qstash-ts#readme","repository":{"type":"git","url":"git+https://github.com/upstash/sdk-qstash-ts.git"},"bugs":{"url":"https://github.com/upstash/sdk-qstash-ts/issues"},"main":"./index.js","module":"./index.mjs","types":"./index.d.ts","files":["./**"],"exports":{".":{"import":"./index.mjs","require":"./index.js"},"./nextjs":{"import":"./nextjs.js","require":"./nextjs.js"},"./dist/nextjs":{"import":"./nextjs.js","require":"./nextjs.js"}},"typesVersions":{"*":{"nextjs":["./nextjs.d.ts"]}},"keywords":["qstash","queue","events","serverless","upstash"],"scripts":{"build":"tsup && cp README.md ./dist/ && cp package.json ./dist/ && cp LICENSE ./dist/","test":"bun test","fmt":"prettier --write .","lint":"tsc && eslint \"src/**/*.{js,ts,tsx}\" --quiet --fix"},"devDependencies":{"@commitlint/cli":"^19.2.2","@commitlint/config-conventional":"^19.2.2","@types/bun":"^1.1.1","@types/crypto-js":"^4.2.0","@typescript-eslint/eslint-plugin":"^7.0.1","@typescript-eslint/parser":"^7.0.1","ai":"^3.1.28","bun-types":"^1.1.7","eslint":"^8","eslint-plugin-unicorn":"^51.0.1","husky":"^9.0.10","next":"^14.0.2","prettier":"^3.2.5","tsup":"latest","typescript":"^5.4.5","undici-types":"^6.16.0","vitest":"latest"},"dependencies":{"crypto-js":">=4.2.0","jose":"^ 5.2.3"}}