@push.rocks/smartai 2.2.0 → 2.3.0

package/dist_ts/smartai.interfaces.d.ts

@@ -1,7 +1,49 @@
-import type { JSONObject, JSONValue, 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' | 'openai' | 'google' | 'groq' | 'mistral' | 'xai' | '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;
@@ -40,6 +82,11 @@ 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.
@@ -56,7 +103,7 @@ export interface ISmartAiOptions {
      * Enable Anthropic prompt caching on system + recent messages.
      * Only used when provider === 'anthropic'. Default: true.
      */
-    promptCaching?: boolean;
+    promptCaching?: boolean | ISmartAiCacheOptions;
 }
 /**
  * Ollama model runtime options passed in the request body `options` field.
@@ -79,4 +126,4 @@ export interface IOllamaModelOptions {
      */
     think?: boolean;
 }
-export type { LanguageModelV3 };
+export type { LanguageModelV3, LanguageModelV3Prompt };

package/dist_ts/smartai.middleware.anthropic.d.ts

@@ -1,7 +1,8 @@
 import type { LanguageModelV3Middleware } from '@ai-sdk/provider';
+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 declare function createAnthropicCachingMiddleware(): LanguageModelV3Middleware;
+export declare function createAnthropicCachingMiddleware(options?: ISmartAiCacheOptions): LanguageModelV3Middleware;

package/dist_ts/smartai.middleware.anthropic.js

@@ -1,36 +1,10 @@
+import { createSmartAiCachingMiddleware } 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() {
-    return {
-        specificationVersion: 'v3',
-        transformParams: async ({ params }) => {
-            const messages = [...params.prompt];
-            // 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 || {}),
-                    anthropic: {
-                        ...(msg.providerOptions?.anthropic || {}),
-                        cacheControl: { type: 'ephemeral' },
-                    },
-                };
-                messages[idx] = msg;
-            }
-            return { ...params, prompt: messages };
-        },
-    };
+export function createAnthropicCachingMiddleware(options = {}) {
+    return createSmartAiCachingMiddleware({ ...options, provider: 'anthropic' });
 }
-//# sourceMappingURL=data:application/json;base64,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
+//# sourceMappingURL=data:application/json;base64,eyJ2ZXJzaW9uIjozLCJmaWxlIjoic21hcnRhaS5taWRkbGV3YXJlLmFudGhyb3BpYy5qcyIsInNvdXJjZVJvb3QiOiIiLCJzb3VyY2VzIjpbIi4uL3RzL3NtYXJ0YWkubWlkZGxld2FyZS5hbnRocm9waWMudHMiXSwibmFtZXMiOltdLCJtYXBwaW5ncyI6IkFBQ0EsT0FBTyxFQUFFLDhCQUE4QixFQUFFLE1BQU0sb0JBQW9CLENBQUM7QUFHcEU7Ozs7R0FJRztBQUNILE1BQU0sVUFBVSxnQ0FBZ0MsQ0FBQyxVQUFnQyxFQUFFO0lBQ2pGLE9BQU8sOEJBQThCLENBQUMsRUFBRSxHQUFHLE9BQU8sRUFBRSxRQUFRLEVBQUUsV0FBVyxFQUFFLENBQUMsQ0FBQztBQUMvRSxDQUFDIn0=

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@push.rocks/smartai",
-  "version": "2.2.0",
+  "version": "2.3.0",
   "private": false,
   "description": "Provider registry and capability utilities for ai-sdk (Vercel AI SDK). Core export returns LanguageModel; subpath exports provide vision, audio, image, document and research capabilities.",
   "main": "dist_ts/index.js",
@@ -34,37 +34,31 @@
   },
   "author": "Task Venture Capital GmbH",
   "license": "MIT",
-  "scripts": {
-    "test": "(tstest test/ --verbose --logfile)",
-    "typecheck": "tsbuild check",
-    "build": "(tsbuild tsfolders)",
-    "buildDocs": "(tsdoc)"
-  },
   "devDependencies": {
-    "@git.zone/tsbuild": "^4.4.0",
-    "@git.zone/tsrun": "^2.0.3",
-    "@git.zone/tstest": "^3.6.3",
+    "@git.zone/tsbuild": "^4.4.1",
+    "@git.zone/tsrun": "^2.0.4",
+    "@git.zone/tstest": "^3.6.6",
     "@push.rocks/qenv": "^6.1.4",
     "@types/json-schema": "^7.0.15",
     "@types/lodash.clonedeep": "^4.5.9",
-    "@types/node": "^25.6.0",
+    "@types/node": "^25.7.0",
     "@types/pngjs": "^6.0.5",
     "typescript": "^6.0.3",
     "undici-types": "^8.2.0"
   },
   "dependencies": {
-    "@ai-sdk/anthropic": "^3.0.75",
-    "@ai-sdk/google": "^3.0.67",
-    "@ai-sdk/groq": "^3.0.38",
-    "@ai-sdk/mistral": "^3.0.35",
-    "@ai-sdk/openai": "^3.0.62",
-    "@ai-sdk/perplexity": "^3.0.32",
+    "@ai-sdk/anthropic": "^3.0.77",
+    "@ai-sdk/google": "^3.0.73",
+    "@ai-sdk/groq": "^3.0.39",
+    "@ai-sdk/mistral": "^3.0.36",
+    "@ai-sdk/openai": "^3.0.63",
+    "@ai-sdk/perplexity": "^3.0.33",
     "@ai-sdk/provider": "^3.0.10",
-    "@ai-sdk/xai": "^3.0.88",
-    "@anthropic-ai/sdk": "0.95.0",
+    "@ai-sdk/xai": "^3.0.89",
+    "@anthropic-ai/sdk": "0.95.2",
     "@push.rocks/smartpdf": "^4.2.2",
-    "ai": "^6.0.175",
-    "openai": "^6.36.0"
+    "ai": "^6.0.180",
+    "openai": "^6.37.0"
   },
   "repository": {
     "type": "git",
@@ -115,11 +109,10 @@
     "AI toolkit",
     "provider switching"
   ],
-  "pnpm": {
-    "onlyBuiltDependencies": [
-      "esbuild",
-      "puppeteer"
-    ]
-  },
-  "packageManager": "pnpm@10.28.2"
-}
+  "scripts": {
+    "test": "(tstest test/ --verbose --logfile)",
+    "typecheck": "tsbuild check",
+    "build": "(tsbuild tsfolders)",
+    "buildDocs": "(tsdoc)"
+  }
+}

package/readme.hints.md

@@ -10,6 +10,7 @@ The package is a **provider registry** built on the Vercel AI SDK (`ai` v6). The
 - Providers: anthropic, openai, google, groq, mistral, xai, perplexity, ollama
 - Anthropic prompt caching via `wrapLanguageModel` middleware (enabled by default)
 - Custom Ollama provider implementing `LanguageModelV3` directly (for think, num_ctx support)
+- OpenAI ChatGPT/Max device-code auth in `smartai.auth.openai.ts`; `openAiMaxAuth` routes OpenAI models to the ChatGPT Codex backend
 
 ### Subpath Exports
 - `@push.rocks/smartai/vision` — `analyzeImage()` using `generateText` with image content
@@ -43,6 +44,7 @@ The package is a **provider registry** built on the Vercel AI SDK (`ai` v6). The
 - `qenv.getEnvVarOnDemand()` returns a Promise — must be awaited in tests
 - OpenAI reasoning options belong in AI SDK `providerOptions`, not model construction options
 - SmartAI accepts OpenAI model IDs as plain strings, including `gpt-5.5`
+- ChatGPT-authenticated OpenAI model calls use `https://chatgpt.com/backend-api/codex` plus bearer access token and `ChatGPT-Account-ID`, not `https://api.openai.com/v1`
 
 ## Testing
 

package/readme.md

@@ -17,7 +17,7 @@ For reporting bugs, issues, or security vulnerabilities, please visit [community
 - **🔌 One function, eight providers** — `getModel()` returns a standard `LanguageModelV3`. Switch providers by changing a string.
 - **🧱 Built on Vercel AI SDK** — Uses `ai` v6 under the hood. Your model works with `generateText()`, `streamText()`, tool calling, structured output, and everything else in the AI SDK ecosystem.
 - **🏠 Custom Ollama provider** — A full `LanguageModelV3` implementation for Ollama with support for `think` mode, `num_ctx`, auto-tuned temperature for Qwen models, and native tool calling.
-- **💰 Anthropic prompt caching** — Automatic `cacheControl` middleware reduces cost and latency on repeated calls. Enabled by default, opt out with `promptCaching: false`.
+- **💰 Prompt caching** — Anthropic cache-control middleware is enabled by default; provider cache helpers are available for agent/session integrations.
 - **📦 Modular subpath exports** — Vision, audio, image, document, and research capabilities ship as separate imports. Only import what you need.
 - **⚡ Zero lock-in** — Your code uses standard AI SDK types. Swap providers without touching application logic.
 
@@ -107,6 +107,30 @@ console.log(result.text);
 
 OpenAI `reasoningEffort` supports `'none'`, `'minimal'`, `'low'`, `'medium'`, `'high'`, and `'xhigh'`. Model IDs are accepted as strings, so new IDs like `'gpt-5.5'` can be used before upstream model unions are updated.
 
+### OpenAI Max / ChatGPT Auth
+
+SmartAI can request ChatGPT subscription-backed Codex credentials with OpenAI's device-code flow. The returned credentials are passed to `getModel()` through `openAiMaxAuth`; SmartAI then routes OpenAI model calls through the ChatGPT Codex backend with the required account headers.
+
+```typescript
+import {
+  completeOpenAiMaxDeviceCodeLogin,
+  getModel,
+  requestOpenAiMaxDeviceCode,
+} from '@push.rocks/smartai';
+
+const deviceCode = await requestOpenAiMaxDeviceCode();
+console.log(`Open ${deviceCode.verificationUrl} and enter ${deviceCode.userCode}`);
+
+const openAiMaxAuth = await completeOpenAiMaxDeviceCodeLogin(deviceCode);
+const model = getModel({
+  provider: 'openai',
+  model: 'gpt-5.5',
+  openAiMaxAuth,
+});
+```
+
+Use `refreshOpenAiMaxTokenData(openAiMaxAuth)` before stored credentials expire, or after receiving an unauthorized response.
+
 ### Re-exported AI SDK Functions
 
 SmartAI re-exports the most commonly used functions from `ai` for convenience:
@@ -250,9 +274,9 @@ console.log(result.text);
 - **Streaming with reasoning** — `doStream()` emits proper `reasoning-start`, `reasoning-delta`, `reasoning-end` parts alongside text.
 - **All Ollama options** — `num_ctx`, `top_k`, `top_p`, `repeat_penalty`, `num_predict`, `stop`, `seed`.
 
-## 💰 Anthropic Prompt Caching
+## 💰 Prompt Caching
 
-When using the Anthropic provider, SmartAI automatically wraps the model with caching middleware that adds `cacheControl: { type: 'ephemeral' }` to the last system message and last user message. This can significantly reduce cost and latency for repeated calls with the same system prompt.
+When using the Anthropic provider, SmartAI automatically wraps the model with caching middleware. The middleware follows the same breakpoint strategy used by opencode: cache the first two system messages and the two most recent non-system messages. This can significantly reduce cost and latency for repeated agent calls with stable system/tool context.
 
 ```typescript
 // Caching enabled by default
@@ -271,6 +295,17 @@ const modelNoCaching = getModel({
 });
 ```
 
+Longer Anthropic cache TTL is opt-in:
+
+```typescript
+const modelWithOneHourCache = getModel({
+  provider: 'anthropic',
+  model: 'claude-sonnet-4-5-20250929',
+  apiKey: process.env.ANTHROPIC_TOKEN,
+  promptCaching: { retention: '1h' },
+});
+```
+
 You can also use the middleware directly:
 
 ```typescript
@@ -281,6 +316,23 @@ const middleware = createAnthropicCachingMiddleware();
 const cachedModel = wrapLanguageModel({ model: baseModel, middleware });
 ```
 
+For agent frameworks, SmartAI exports lower-level helpers:
+
+```typescript
+import {
+  applySmartAiCacheProviderOptions,
+  createSmartAiCachingMiddleware,
+} from '@push.rocks/smartai';
+
+const providerOptions = applySmartAiCacheProviderOptions({
+  provider: 'openai',
+  sessionId: 'stable-session-id',
+  cache: 'auto',
+});
+```
+
+OpenAI request-level cache affinity is only added when a stable `sessionId` or explicit cache `key` is provided. Extended OpenAI retention (`'24h'`) is opt-in.
+
 ## 📦 Subpath Exports
 
 SmartAI provides specialized capabilities as separate subpath imports. Each one is a focused utility that takes a model (or API key) and does one thing well.

package/ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@push.rocks/smartai',
-  version: '2.2.0',
+  version: '2.3.0',
   description: 'Provider registry and capability utilities for ai-sdk (Vercel AI SDK). Core export returns LanguageModel; subpath exports provide vision, audio, image, document and research capabilities.'
 }

package/ts/index.ts

@@ -1,6 +1,13 @@
 export { getModel, getModelSetup } from './smartai.classes.smartai.js';
 export type {
   IOpenAiProviderOptions,
+  IOpenAiMaxAuthCredentials,
+  IOpenAiMaxAuthOptions,
+  IOpenAiMaxCompleteDeviceCodeOptions,
+  IOpenAiMaxDeviceCode,
+  IOpenAiMaxDeviceCodePollOptions,
+  IOpenAiMaxIdTokenInfo,
+  IOpenAiMaxTokenData,
   ISmartAiModelSetup,
   ISmartAiOptions,
   TOpenAiReasoningEffort,
@@ -9,9 +16,41 @@ export type {
   TSmartAiProviderOptions,
   IOllamaModelOptions,
   LanguageModelV3,
+  LanguageModelV3Prompt,
 } from './smartai.interfaces.js';
 export { createAnthropicCachingMiddleware } from './smartai.middleware.anthropic.js';
+export {
+  applySmartAiCacheProviderOptions,
+  applySmartAiPromptCaching,
+  createSmartAiCachingMiddleware,
+  getSmartAiCacheProviderOptions,
+  getSmartAiMessageCacheProviderOptions,
+  mergeSmartAiProviderOptions,
+  resolveSmartAiCacheProvider,
+} from './smartai.cache.js';
+export type {
+  ISmartAiCacheOptions,
+  TSmartAiCacheRetention,
+  TSmartAiCacheSetting,
+  TSmartAiMessageCacheProvider,
+} from './smartai.cache.js';
 export { createOllamaModel } from './smartai.provider.ollama.js';
+export {
+  OPENAI_MAX_AUTH_ISSUER,
+  OPENAI_MAX_CLIENT_ID,
+  OPENAI_MAX_CODEX_BASE_URL,
+  OPENAI_MAX_DEFAULT_ORIGINATOR,
+  OpenAiMaxAuthError,
+  completeOpenAiMaxDeviceCodeLogin,
+  createOpenAiMaxProviderSettings,
+  ensureOpenAiMaxWorkspaceAllowed,
+  exchangeOpenAiMaxAuthorizationCode,
+  parseOpenAiMaxIdToken,
+  pollOpenAiMaxDeviceCode,
+  refreshOpenAiMaxTokenData,
+  requestOpenAiMaxDeviceCode,
+} from './smartai.auth.openai.js';
+export type { IOpenAiMaxAuthorizationCode } from './smartai.auth.openai.js';
 
 // Re-export commonly used ai-sdk functions for consumer convenience
 export { generateText, streamText, tool, jsonSchema } from 'ai';