@push.rocks/smartai 2.0.1 → 2.3.0

package/ts/smartai.interfaces.ts

@@ -1,4 +1,5 @@
-import type { LanguageModelV3 } from '@ai-sdk/provider';
+import type { JSONObject, JSONValue, LanguageModelV3, LanguageModelV3Prompt } from '@ai-sdk/provider';
+import type { ISmartAiCacheOptions } from './smartai.cache.js';
 
 export type TProvider =
   | 'anthropic'
@@ -10,10 +11,109 @@ export type TProvider =
   | 'perplexity'
   | 'ollama';
 
+export type TOpenAiReasoningEffort = 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh';
+
+export type TOpenAiTextVerbosity = 'low' | 'medium' | 'high';
+
+export interface IOpenAiMaxIdTokenInfo {
+  email?: string;
+  chatgptPlanType?: string;
+  chatgptUserId?: string;
+  chatgptAccountId?: string;
+  chatgptAccountIsFedramp: boolean;
+  expiresAt?: string;
+  rawJwt: string;
+}
+
+export interface IOpenAiMaxAuthCredentials {
+  accessToken: string;
+  refreshToken?: string;
+  idToken?: string;
+  accountId?: string;
+  idTokenInfo?: IOpenAiMaxIdTokenInfo;
+  baseUrl?: string;
+  originator?: string;
+}
+
+export interface IOpenAiMaxTokenData extends IOpenAiMaxAuthCredentials {
+  refreshToken: string;
+  idToken: string;
+  idTokenInfo: IOpenAiMaxIdTokenInfo;
+}
+
+export interface IOpenAiMaxDeviceCode {
+  verificationUrl: string;
+  userCode: string;
+  deviceAuthId: string;
+  intervalSeconds: number;
+}
+
+export interface IOpenAiMaxAuthOptions {
+  issuer?: string;
+  clientId?: string;
+  fetch?: typeof fetch;
+}
+
+export interface IOpenAiMaxDeviceCodePollOptions extends IOpenAiMaxAuthOptions {
+  timeoutMs?: number;
+  sleep?: (ms: number) => Promise<void>;
+}
+
+export interface IOpenAiMaxCompleteDeviceCodeOptions extends IOpenAiMaxDeviceCodePollOptions {
+  forcedChatGptWorkspaceId?: string;
+}
+
+export interface IOpenAiProviderOptions extends JSONObject {
+  conversation?: string | null;
+  include?: string[] | null;
+  instructions?: string | null;
+  logitBias?: Record<string, number>;
+  logprobs?: boolean | number | null;
+  maxCompletionTokens?: number;
+  maxToolCalls?: number | null;
+  metadata?: JSONObject | null;
+  parallelToolCalls?: boolean | null;
+  previousResponseId?: string | null;
+  prediction?: JSONObject;
+  promptCacheKey?: string | null;
+  promptCacheRetention?: 'in_memory' | '24h' | null;
+  reasoningEffort?: TOpenAiReasoningEffort | null;
+  reasoningSummary?: string | null;
+  safetyIdentifier?: string | null;
+  serviceTier?: 'auto' | 'flex' | 'priority' | 'default' | null;
+  store?: boolean | null;
+  strictJsonSchema?: boolean | null;
+  systemMessageMode?: 'remove' | 'system' | 'developer';
+  textVerbosity?: TOpenAiTextVerbosity | null;
+  truncation?: 'auto' | 'disabled' | null;
+  user?: string | null;
+  forceReasoning?: boolean;
+  [key: string]: JSONValue | undefined;
+}
+
+export type TSmartAiProviderOptions = Record<string, JSONObject> & {
+  openai?: IOpenAiProviderOptions;
+};
+
+export interface ISmartAiModelSetup {
+  model: LanguageModelV3;
+  providerOptions?: TSmartAiProviderOptions;
+}
+
 export interface ISmartAiOptions {
   provider: TProvider;
   model: string;
   apiKey?: string;
+  /**
+   * OpenAI ChatGPT/Codex subscription credentials from the device-code auth flow.
+   * Only used when provider === 'openai'.
+   */
+  openAiMaxAuth?: IOpenAiMaxAuthCredentials;
+  /**
+   * Provider-specific AI SDK generation options.
+   * Pass this to generateText()/streamText() alongside the model.
+   */
+  providerOptions?: TSmartAiProviderOptions;
   /** For Ollama: base URL of the local server. Default: http://localhost:11434 */
   baseUrl?: string;
   /**
@@ -25,7 +125,7 @@ export interface ISmartAiOptions {
    * Enable Anthropic prompt caching on system + recent messages.
    * Only used when provider === 'anthropic'. Default: true.
    */
-  promptCaching?: boolean;
+  promptCaching?: boolean | ISmartAiCacheOptions;
 }
 
 /**
@@ -50,4 +150,4 @@ export interface IOllamaModelOptions {
   think?: boolean;
 }
 
-export type { LanguageModelV3 };
+export type { LanguageModelV3, LanguageModelV3Prompt };

package/ts/smartai.middleware.anthropic.ts

@@ -1,38 +1,12 @@
-import type { LanguageModelV3Middleware, LanguageModelV3Prompt } from '@ai-sdk/provider';
+import type { LanguageModelV3Middleware } from '@ai-sdk/provider';
+import { createSmartAiCachingMiddleware } from './smartai.cache.js';
+import type { ISmartAiCacheOptions } from './smartai.cache.js';
 
 /**
  * Creates middleware that adds Anthropic prompt caching directives.
  * Marks the last system message and last user message with ephemeral cache control,
  * reducing input token cost and latency on repeated calls.
  */
-export function createAnthropicCachingMiddleware(): LanguageModelV3Middleware {
-  return {
-    specificationVersion: 'v3',
-    transformParams: async ({ params }) => {
-      const messages = [...params.prompt] as Array<Record<string, unknown>>;
-
-      // Find the last system message and last user message
-      let lastSystemIdx = -1;
-      let lastUserIdx = -1;
-      for (let i = 0; i < messages.length; i++) {
-        if (messages[i].role === 'system') lastSystemIdx = i;
-        if (messages[i].role === 'user') lastUserIdx = i;
-      }
-
-      const targets = [lastSystemIdx, lastUserIdx].filter(i => i >= 0);
-      for (const idx of targets) {
-        const msg = { ...messages[idx] };
-        msg.providerOptions = {
-          ...(msg.providerOptions as Record<string, unknown> || {}),
-          anthropic: {
-            ...((msg.providerOptions as Record<string, unknown>)?.anthropic as Record<string, unknown> || {}),
-            cacheControl: { type: 'ephemeral' },
-          },
-        };
-        messages[idx] = msg;
-      }
-
-      return { ...params, prompt: messages as unknown as LanguageModelV3Prompt };
-    },
-  };
+export function createAnthropicCachingMiddleware(options: ISmartAiCacheOptions = {}): LanguageModelV3Middleware {
+  return createSmartAiCachingMiddleware({ ...options, provider: 'anthropic' });
 }