@pwshub/aisdk 0.0.3 → 0.0.4

package/README.md

@@ -2,6 +2,10 @@
 
 A thin, unified AI client for OpenAI, Anthropic, Google, DashScope, and DeepSeek with automatic parameter normalization and fallback support.
 
+[![npm version](https://badge.fury.io/js/@pwshub%2Faisdk.svg)](https://badge.fury.io/js/@pwshub%2Faisdk)
+![CodeQL](https://github.com/pwshub/aisdk/workflows/CodeQL/badge.svg)
+![CI test](https://github.com/pwshub/aisdk/workflows/ci-test/badge.svg)
+
 ## Features
 
 - **Unified API**: Single interface for multiple AI providers
@@ -241,22 +245,40 @@ const result = await ai.ask({
 
 ## Supported Models
 
-This library does not ship with a predefined list of models. Instead, it accepts **any model** from the supported providers:
+The library comes with **30 pre-configured models** from all supported providers:
 
-- **OpenAI**: Any OpenAI model
-- **Anthropic**: Any Anthropic model
-- **Google**: Any Google model
-- **DashScope**: Any DashScope model
-- **DeepSeek**: Any DeepSeek model
+- **OpenAI**: gpt-4.1-nano, gpt-4.1-mini, gpt-4.1, gpt-4o, gpt-4o-mini, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5.1, gpt-5.2, gpt-5.4, o3-mini, o4-mini
+- **Anthropic**: claude-haiku-4-5, claude-sonnet-4-6, claude-sonnet-4-5, claude-opus-4-6
+- **Google**: gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.5-pro, gemini-3.1-pro-preview, gemini-3.1-flash-lite-preview
+- **DashScope**: qwen-flash, qwen3.5-flash, qwen-plus, qwen3.5-plus, qwen-max, qwen3-max
+- **DeepSeek**: deepseek-chat, deepseek-reasoner
 
-### Loading Models
+### Managing Models
 
-Models are loaded programmatically via `setModels()` from external sources (CMS, API, or local files for evaluation):
+Models are managed via `addModels()` and `setModels()`:
 
 ```javascript
-import { createAi, setModels } from '@pwshub/aisdk'
+import { createAi, addModels, setModels, listModels } from '@pwshub/aisdk'
+
+// List all available models (30 models loaded by default)
+console.log(listModels())
+
+// Add more models to the existing list
+addModels([
+  {
+    id: 'my-custom-model',
+    name: 'my-custom-model',
+    provider: 'openai',
+    input_price: 1,
+    output_price: 2,
+    cache_price: 0.5,
+    max_in: 128000,
+    max_out: 16384,
+    enable: true,
+  },
+])
 
-// Load models from your CMS or API
+// Replace all models with your own list (e.g., from CMS)
 const modelsFromCms = await fetch('https://cms.example.com/api/models').then(r => r.json())
 setModels(modelsFromCms)
 
@@ -268,6 +290,8 @@ const result = await ai.ask({
 })
 ```
 
+> **Note:** Models are loaded automatically from `src/models.js` when the library is imported. You don't need to call `setModels()` unless you want to use a custom model list.
+
 ### Model Record Format
 
 Each model record should include:
@@ -282,8 +306,6 @@ Each model record should include:
 - `enable`: Boolean to enable/disable the model
 - `supportedParams` (optional): Array of supported parameter names
 
-> **Note**: The `examples/` folder includes `models.json` as a reference for running evaluation scripts.
-
 ## Error Handling
 
 ```javascript

package/index.d.ts

@@ -70,5 +70,6 @@ export interface AiClient {
 }
 
 export function createAi(opts?: AiOptions): AiClient;
+export function addModels(models: ModelRecord[]): void;
 export function setModels(models: ModelRecord[]): void;
 export function listModels(): ModelRecord[];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pwshub/aisdk",
-  "version": "0.0.3",
+  "version": "0.0.4",
   "description": "A thin, unified AI client for OpenAI, Anthropic, Google, DashScope, and DeepSeek with automatic param normalization and fallback support",
   "repository": {
     "type": "git",

package/src/index.js

@@ -38,10 +38,21 @@
  *   },
  * })
  *
+ * @example Using messages array for multi-turn conversations
+ * const result = await ai.ask({
+ *   model: 'claude-sonnet-4-20250514',
+ *   apikey: 'your-api-key',
+ *   messages: [
+ *     { role: 'user', content: 'What is the capital of Vietnam?' },
+ *     { role: 'assistant', content: 'The capital of Vietnam is Hanoi.' },
+ *     { role: 'user', content: 'What is its population?' },
+ *   ],
+ * })
+ *
  */
 
 import {
-  getModel, listModels, setModels,
+  getModel, listModels, setModels, addModels,
 } from './registry.js'
 import { normalizeConfig } from './config.js'
 import { coerceConfig } from './coerce.js'
@@ -64,8 +75,9 @@ export {
  * @typedef {Object} AskParams
  * @property {string} model                       - Model ID (must be registered via setModels())
  * @property {string} apikey                      - API key for the provider
- * @property {string} prompt                      - The user message
- * @property {string} [system]                    - Optional system prompt
+ * @property {string} [prompt]                    - The user message (alternative to messages)
+ * @property {string} [system]                    - Optional system prompt (used with prompt)
+ * @property {import('./providers.js').Message[]} [messages] - Array of messages with role and content (alternative to prompt)
  * @property {string[]} [fallbacks]               - Ordered list of fallback model IDs
  * @property {Record<string, unknown>} [providerOptions] - Provider-specific options merged into body
  * @property {number} [temperature]
@@ -152,11 +164,11 @@ const callModel = async (modelId, params, gatewayUrl) => {
   const normalizedConfig = normalizeConfig(coerced, providerId, supportedParams, modelId)
 
   const {
-    prompt, system, providerOptions = {},
+    prompt, system, messages, providerOptions = {},
   } = params
 
   /** @type {import('./providers.js').Message[]} */
-  const messages = [
+  const messageList = messages ?? [
     ...(system ? [{
       role: 'system', content: system,
     }] : []),
@@ -166,7 +178,7 @@ const callModel = async (modelId, params, gatewayUrl) => {
   ]
 
   const url = gatewayUrl ?? adapter.url(modelName, apikey)
-  const body = adapter.buildBody(modelName, messages, normalizedConfig, providerOptions)
+  const body = adapter.buildBody(modelName, messageList, normalizedConfig, providerOptions)
 
   let res
   try {
@@ -267,4 +279,4 @@ export const createAi = (opts = {}) => {
   }
 }
 
-export { setModels }
+export { addModels, setModels, listModels }

package/src/models.js

@@ -0,0 +1,345 @@
+/**
+ * @fileoverview Default model registry for @pwshub/aisdk.
+ *
+ * This module exports a default list of models that are loaded automatically
+ * when the library is imported. Users can modify this list via addModels()
+ * and setModels() from the main export.
+ */
+
+/**
+ * @typedef {import('./registry.js').ModelRecord} ModelRecord
+ */
+
+/** @type {ModelRecord[]} */
+export const DEFAULT_MODELS = [
+  {
+    id: 'claude-haiku-4-5',
+    name: 'claude-haiku-4-5',
+    provider: 'anthropic',
+    input_price: 1,
+    output_price: 5,
+    cache_price: 0,
+    max_in: 200000,
+    max_out: 64000,
+    enable: true,
+  },
+  {
+    id: 'claude-sonnet-4-6',
+    name: 'claude-sonnet-4-6',
+    provider: 'anthropic',
+    input_price: 3,
+    output_price: 15,
+    cache_price: 0,
+    max_in: 200000,
+    max_out: 64000,
+    enable: true,
+  },
+  {
+    id: 'claude-sonnet-4-5',
+    name: 'claude-sonnet-4-5',
+    provider: 'anthropic',
+    input_price: 3,
+    output_price: 15,
+    cache_price: 0,
+    max_in: 200000,
+    max_out: 1000000,
+    enable: true,
+  },
+  {
+    id: 'claude-opus-4-6',
+    name: 'claude-opus-4-6',
+    provider: 'anthropic',
+    input_price: 5,
+    output_price: 25,
+    cache_price: 0,
+    max_in: 200000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gemini-2.5-flash',
+    name: 'gemini-2.5-flash',
+    provider: 'google',
+    input_price: 0.3,
+    output_price: 2.5,
+    cache_price: 0.03,
+    max_in: 1048576,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'gemini-2.5-flash-lite',
+    name: 'gemini-2.5-flash-lite',
+    provider: 'google',
+    input_price: 0.1,
+    output_price: 0.4,
+    cache_price: 0.01,
+    max_in: 1048576,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'gemini-2.5-pro',
+    name: 'gemini-2.5-pro',
+    provider: 'google',
+    input_price: 1.25,
+    output_price: 10,
+    cache_price: 0.125,
+    max_in: 1048576,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'gemini-3.1-pro-preview',
+    name: 'gemini-3.1-pro-preview',
+    provider: 'google',
+    input_price: 2,
+    output_price: 12,
+    cache_price: 0.2,
+    max_in: 1048576,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'gemini-3.1-flash-lite-preview',
+    name: 'gemini-3.1-flash-lite-preview',
+    provider: 'google',
+    input_price: 0.25,
+    output_price: 1.5,
+    cache_price: 0.025,
+    max_in: 1048576,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'gpt-4.1-nano',
+    name: 'gpt-4.1-nano',
+    provider: 'openai',
+    input_price: 0.1,
+    output_price: 0.4,
+    cache_price: 0.025,
+    max_in: 1047576,
+    max_out: 32768,
+    enable: true,
+  },
+  {
+    id: 'gpt-4.1-mini',
+    name: 'gpt-4.1-mini',
+    provider: 'openai',
+    input_price: 0.4,
+    output_price: 1.6,
+    cache_price: 0.1,
+    max_in: 1047576,
+    max_out: 32768,
+    enable: true,
+  },
+  {
+    id: 'gpt-4.1',
+    name: 'gpt-4.1',
+    provider: 'openai',
+    input_price: 2,
+    output_price: 8,
+    cache_price: 0.5,
+    max_in: 1047576,
+    max_out: 32768,
+    enable: true,
+  },
+  {
+    id: 'gpt-4o',
+    name: 'gpt-4o',
+    provider: 'openai',
+    input_price: 2.5,
+    output_price: 10,
+    cache_price: 1.25,
+    max_in: 128000,
+    max_out: 16384,
+    enable: true,
+  },
+  {
+    id: 'gpt-4o-mini',
+    name: 'gpt-4o-mini',
+    provider: 'openai',
+    input_price: 0.15,
+    output_price: 0.6,
+    cache_price: 0.075,
+    max_in: 128000,
+    max_out: 16384,
+    enable: true,
+  },
+  {
+    id: 'gpt-5',
+    name: 'gpt-5',
+    provider: 'openai',
+    input_price: 1.25,
+    output_price: 10,
+    cache_price: 0.125,
+    max_in: 400000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gpt-5-mini',
+    name: 'gpt-5-mini',
+    provider: 'openai',
+    input_price: 0.25,
+    output_price: 2,
+    cache_price: 0.025,
+    max_in: 400000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gpt-5-nano',
+    name: 'gpt-5-nano',
+    provider: 'openai',
+    input_price: 0.05,
+    output_price: 0.4,
+    cache_price: 0.005,
+    max_in: 400000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gpt-5.1',
+    name: 'gpt-5.1',
+    provider: 'openai',
+    input_price: 1.25,
+    output_price: 10,
+    cache_price: 0.125,
+    max_in: 400000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gpt-5.2',
+    name: 'gpt-5.2',
+    provider: 'openai',
+    input_price: 1.75,
+    output_price: 14,
+    cache_price: 0.175,
+    max_in: 400000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'gpt-5.4',
+    name: 'gpt-5.4',
+    provider: 'openai',
+    input_price: 2.5,
+    output_price: 15,
+    cache_price: 0.25,
+    max_in: 1050000,
+    max_out: 128000,
+    enable: true,
+  },
+  {
+    id: 'o3-mini',
+    name: 'o3-mini',
+    provider: 'openai',
+    input_price: 1.1,
+    output_price: 4.4,
+    cache_price: 0.55,
+    max_in: 200000,
+    max_out: 100000,
+    enable: true,
+  },
+  {
+    id: 'o4-mini',
+    name: 'o4-mini',
+    provider: 'openai',
+    input_price: 1.1,
+    output_price: 4.4,
+    cache_price: 0.275,
+    max_in: 200000,
+    max_out: 100000,
+    enable: true,
+  },
+  {
+    id: 'deepseek-chat',
+    name: 'deepseek-chat',
+    provider: 'deepseek',
+    input_price: 0.28,
+    output_price: 0.42,
+    cache_price: 0.028,
+    max_in: 128000,
+    max_out: 8000,
+    enable: true,
+  },
+  {
+    id: 'deepseek-reasoner',
+    name: 'deepseek-reasoner',
+    provider: 'deepseek',
+    input_price: 0.28,
+    output_price: 0.42,
+    cache_price: 0.028,
+    max_in: 128000,
+    max_out: 64000,
+    enable: true,
+  },
+  {
+    id: 'qwen-flash',
+    name: 'qwen-flash',
+    provider: 'dashscope',
+    input_price: 0.05,
+    output_price: 0.4,
+    cache_price: 0,
+    max_in: 995904,
+    max_out: 32768,
+    enable: true,
+  },
+  {
+    id: 'qwen3.5-flash',
+    name: 'qwen3.5-flash',
+    provider: 'dashscope',
+    input_price: 0.1,
+    output_price: 0.4,
+    cache_price: 0,
+    max_in: 983616,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'qwen-plus',
+    name: 'qwen-plus',
+    provider: 'dashscope',
+    input_price: 0.4,
+    output_price: 1.2,
+    cache_price: 0,
+    max_in: 997952,
+    max_out: 32768,
+    enable: true,
+  },
+  {
+    id: 'qwen3.5-plus',
+    name: 'qwen3.5-plus',
+    provider: 'dashscope',
+    input_price: 0.4,
+    output_price: 2.4,
+    cache_price: 0,
+    max_in: 991808,
+    max_out: 65536,
+    enable: true,
+  },
+  {
+    id: 'qwen-max',
+    name: 'qwen-max',
+    provider: 'dashscope',
+    input_price: 1.6,
+    output_price: 6.4,
+    cache_price: 0,
+    max_in: 30720,
+    max_out: 8192,
+    enable: true,
+  },
+  {
+    id: 'qwen3-max',
+    name: 'qwen3-max',
+    provider: 'dashscope',
+    input_price: 1.2,
+    output_price: 6,
+    cache_price: 0,
+    max_in: 258048,
+    max_out: 65536,
+    enable: true,
+  },
+]

package/src/providers.js

@@ -139,10 +139,19 @@ const google = {
         role: m.role === 'assistant' ? 'model' : 'user',
         parts: [{ text: m.content }],
       }))
+    
+    // Thinking models (e.g., gemini-2.5-pro) need more tokens for reasoning
+    // Set a higher default maxOutputTokens if not specified
+    const hasMaxTokens = config.generationConfig?.maxOutputTokens !== undefined
+    const defaultGenerationConfig = hasMaxTokens ? {} : { maxOutputTokens: 8192 }
+    
     return {
       contents,
       ...(system && { systemInstruction: { parts: [{ text: system }] } }),
-      ...config, // includes nested generationConfig
+      generationConfig: {
+        ...defaultGenerationConfig,
+        ...config.generationConfig,
+      },
       ...providerOptions, // safetySettings, thinkingConfig, etc.
     }
   },
@@ -164,21 +173,40 @@ const google = {
       throw new Error('Google response missing content')
     }
 
-    // Gemini 2.5 Pro may return parts as array or direct text
+    // Gemini 2.5 Pro (thinking model) may return content without parts
+    // when all tokens were used for reasoning
+    if (!content.parts || (Array.isArray(content.parts) && content.parts.length === 0)) {
+      const thoughts = data.usageMetadata?.thoughtsTokenCount ?? 0
+      const totalTokens = data.usageMetadata?.totalTokenCount ?? 0
+      
+      if (finishReason === 'MAX_TOKENS' && thoughts > 0) {
+        throw new Error(
+          `Google model used ${thoughts}/${totalTokens} tokens for internal reasoning and has no tokens left for output. ` +
+          `Increase maxTokens to allow room for both thinking and response.`
+        )
+      }
+      
+      throw new Error('Google response has no content parts')
+    }
+
+    // Gemini may return parts as array or direct text
     if (Array.isArray(content.parts)) {
-      const text = content.parts[0]?.text
-      if (!text) {
-        // Model may have used all tokens for reasoning (thoughtsTokenCount)
+      // Concatenate all text parts (model may return multiple text blocks)
+      const texts = content.parts.filter((p) => p.text).map((p) => p.text)
+      if (texts.length === 0) {
         const thoughts = data.usageMetadata?.thoughtsTokenCount ?? 0
         if (finishReason === 'MAX_TOKENS' && thoughts > 0) {
-          throw new Error(`Google response missing content (used ${thoughts} tokens for reasoning, maxTokens may be too low)`)
+          throw new Error(
+            `Google model used ${thoughts}/${data.usageMetadata?.totalTokenCount ?? 0} tokens for internal reasoning and has no tokens left for output. ` +
+            `Increase maxTokens to allow room for both thinking and response.`
+          )
         }
-        throw new Error('Google response missing content')
+        throw new Error('Google response has no text content')
       }
-      return text
+      return texts.join('')
     }
 
-    // Some models may return content directly
+    // Some models may return content directly as string
     if (typeof content.parts === 'string') {
       return content.parts
     }

package/src/registry.js

@@ -1,9 +1,10 @@
 /**
  * @fileoverview Model registry — in-memory store for model records.
  *
- * Models are loaded programmatically via setModels() from external sources
- * (CMS, API, or local files for evaluation). This module provides O(1) lookups
- * at runtime via a Map indexed by model ID.
+ * Default models are loaded automatically from ./models.js at import time.
+ * Users can modify the registry via addModels() and setModels().
+ *
+ * This module provides O(1) lookups at runtime via a Map indexed by model ID.
  *
  * `supportedParams` is optional per record. When absent, the provider's
  * default param set is used.
@@ -11,6 +12,8 @@
  * @typedef {'openai'|'anthropic'|'google'|'dashscope'|'deepseek'} ProviderId
  */
 
+import { DEFAULT_MODELS } from './models.js'
+
 /**
  * Mirrors the Directus collection schema exactly.
  * `supportedParams` is optional — added later via Directus field.
@@ -48,6 +51,17 @@ const VALID_PROVIDERS = ['openai', 'anthropic', 'google', 'dashscope', 'deepseek
 /** @type {Map<string, ModelRecord>} */
 let REGISTRY = new Map()
 
+/**
+ * Initializes the registry with default models.
+ * Called automatically at module import.
+ */
+const initRegistry = () => {
+  REGISTRY = new Map(DEFAULT_MODELS.map((model) => [model.id, model]))
+}
+
+// Initialize with default models on import
+initRegistry()
+
 /**
  * Validates a single model record structure and types.
  *
@@ -143,11 +157,33 @@ export const listModels = () =>
   [...REGISTRY.values()].filter((m) => m.enable)
 
 /**
- * Programmatically sets the model registry from an array of model records.
- * Use this when loading models from a CMS or other external source instead of
- * the built-in models.json file.
+ * Adds one or more models to the registry.
+ * Existing models with the same ID are overwritten.
+ *
+ * @param {ModelRecord[]} models - Array of model records to add
+ * @throws {Error} When models is not an array or contains invalid records
+ */
+export const addModels = (models) => {
+  if (!Array.isArray(models)) {
+    throw new Error(`addModels expects an array. Got: ${typeof models}`)
+  }
+
+  // Validate each model record
+  models.forEach((model, index) => {
+    validateModelRecord(model, index)
+  })
+
+  // Add models to the registry
+  models.forEach((model) => {
+    REGISTRY.set(model.id, model)
+  })
+}
+
+/**
+ * Replaces the entire model registry with a new list of models.
+ * Use this to load models from a CMS or other external source.
  *
- * @param {ModelRecord[]} models - Array of model records (same format as models.json)
+ * @param {ModelRecord[]} models - Array of model records
  * @throws {Error} When models is not an array or contains invalid records
  */
 export const setModels = (models) => {

package/src/validation.js

@@ -9,7 +9,7 @@
  * @typedef {Object} AskParams
  * @property {string} model
  * @property {string} apikey
- * @property {string} prompt
+ * @property {string} [prompt]
  * @property {string} [system]
  * @property {import('../index.js').Message[]} [messages]
  * @property {number} [temperature]
@@ -42,8 +42,14 @@ export const validateAskOptions = (params) => {
     errors.push('"apikey" must be a non-empty string')
   }
 
-  if (!params.prompt || typeof params.prompt !== 'string') {
-    errors.push('"prompt" must be a non-empty string')
+  // Either prompt or messages must be provided (but not both required)
+  if (params.prompt === undefined && params.messages === undefined) {
+    errors.push('either "prompt" or "messages" must be provided')
+  }
+
+  // When using messages, system can still be provided (will be prepended)
+  if (params.prompt !== undefined && typeof params.prompt !== 'string') {
+    errors.push('"prompt" must be a string')
   }
 
   // Optional string fields