kiku-agent 1.0.5 → 1.0.7

package/server/agent/config.js

@@ -1,4 +1,31 @@
-export const CLOUD_PROVIDERS = {
+const PROVIDER_MODEL_ALIASES = {
+  openai: {
+    'o1-mini': 'gpt-5.6-luna',
+    'o3-mini': 'gpt-5.6-luna',
+  },
+  openrouter: {
+    auto: 'openrouter/auto',
+  },
+  groq: {
+    'gpt-oss-120b': 'openai/gpt-oss-120b',
+    'gpt-oss-20b': 'openai/gpt-oss-20b',
+  },
+  deepseek: {
+    'deepseek-chat': 'deepseek-v4-flash',
+    'deepseek-reasoner': 'deepseek-v4-flash',
+  },
+  gemini: {
+    'gemini-3.1-pro': 'gemini-3.1-pro-preview',
+  },
+};
+
+export function normalizeProviderModel(providerId, modelId) {
+  const normalizedProvider = String(providerId || '').toLowerCase().trim();
+  const normalizedModel = String(modelId || '').trim();
+  return PROVIDER_MODEL_ALIASES[normalizedProvider]?.[normalizedModel] || normalizedModel;
+}
+
+export const CLOUD_PROVIDERS = {
   lmstudio: {
     id: 'lmstudio',
     name: 'LM Studio (Local)',
@@ -23,18 +50,17 @@ export const CLOUD_PROVIDERS = {
     name: 'OpenAI Cloud',
     baseUrl: process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1',
     apiKey: process.env.OPENAI_API_KEY || '',
-    defaultModel: process.env.OPENAI_MODEL || 'gpt-4o-mini',
+    defaultModel: normalizeProviderModel('openai', process.env.OPENAI_MODEL || 'gpt-5.6-luna'),
     isLocal: false,
     envKey: 'OPENAI_API_KEY',
-    docUrl: 'https://platform.openai.com/docs',
-    apiRefUrl: 'https://platform.openai.com/docs/api-reference',
-    pricingSummary: 'Pay-per-token (gpt-4o-mini: ~$0.15/1M in, $0.60/1M out)',
+    docUrl: 'https://developers.openai.com/api/docs/models',
+    apiRefUrl: 'https://developers.openai.com/api/reference',
+    pricingSummary: 'GPT-5.6 Luna: $0.20/1M in, $1.20/1M out; Terra/Sol available for higher capability',
     features: ['tools', 'streaming', 'vision', 'reasoning', 'system-prompt'],
     recommendedModels: [
-      { id: 'gpt-4o-mini', name: 'GPT-4o Mini', tier: 'ultra-cheap', cost: '$0.15 / 1M in', context: '128k', hasTools: true, description: 'Recommended default. Fast, intelligent, and highly cost-effective with tool use.' },
-      { id: 'gpt-4o', name: 'GPT-4o', tier: 'flagship', cost: '$2.50 / 1M in', context: '128k', hasTools: true, description: 'OpenAI flagship multimodal model for complex multimodal & reasoning tasks.' },
-      { id: 'o3-mini', name: 'o3 Mini', tier: 'reasoning', cost: '$1.10 / 1M in', context: '200k', hasTools: true, description: 'Fast, cost-efficient reasoning model with function calling capabilities.' },
-      { id: 'o1-mini', name: 'o1 Mini', tier: 'reasoning', cost: '$1.10 / 1M in', context: '128k', hasTools: true, description: 'Reasoning model specialized for complex STEM logic and coding.' },
+      { id: 'gpt-5.6-luna', name: 'GPT-5.6 Luna', tier: 'cost-sensitive', cost: '$0.20 / 1M in', context: '1.05M', hasTools: true, description: 'Recommended default for efficient, high-volume multilingual agent workloads.' },
+      { id: 'gpt-5.6-terra', name: 'GPT-5.6 Terra', tier: 'balanced', cost: '$2.00 / 1M in', context: '1.05M', hasTools: true, description: 'Balances intelligence and cost for production reasoning and tool use.' },
+      { id: 'gpt-5.6-sol', name: 'GPT-5.6 Sol', tier: 'flagship', cost: '$4.00 / 1M in', context: '1.05M', hasTools: true, description: 'Flagship model for complex professional reasoning and coding.' },
     ],
   },
   openrouter: {
@@ -42,62 +68,53 @@ export const CLOUD_PROVIDERS = {
     name: 'OpenRouter Cloud',
     baseUrl: process.env.OPENROUTER_BASE_URL || 'https://openrouter.ai/api/v1',
     apiKey: process.env.OPENROUTER_API_KEY || '',
-    defaultModel: process.env.OPENROUTER_MODEL || 'google/gemini-2.5-flash',
-    extraHeaders: {
-      'HTTP-Referer': 'https://jujingyi-kikuuu.vercel.app/',
-      'X-Title': 'Kiku Agent',
-    },
+    defaultModel: normalizeProviderModel('openrouter', process.env.OPENROUTER_MODEL || 'openrouter/auto'),
+    extraHeaders: {
+      'HTTP-Referer': 'https://jujingyi-kikuuu.vercel.app/',
+      'X-OpenRouter-Title': 'Kiku Agent',
+    },
     isLocal: false,
     envKey: 'OPENROUTER_API_KEY',
-    docUrl: 'https://openrouter.ai/docs/quick-start',
-    apiRefUrl: 'https://openrouter.ai/docs/api_reference/overview',
-    pricingSummary: 'Universal aggregator with free tiers & flexible pay-as-you-go',
+    docUrl: 'https://openrouter.ai/docs/guides/routing/routers/auto-router',
+    apiRefUrl: 'https://openrouter.ai/docs/api/reference/overview',
+    pricingSummary: 'Auto Router selects a current model dynamically; usage is billed at the routed model rate',
     features: ['universal-routing', 'tools', 'streaming', 'free-tier', 'fallback'],
     recommendedModels: [
-      { id: 'google/gemini-2.5-flash', name: 'Gemini 2.5 Flash', tier: 'ultra-cheap', cost: '$0.10 / 1M in', context: '1M', hasTools: true, description: 'Super fast with huge 1M context window and native function calling.' },
-      { id: 'deepseek/deepseek-chat', name: 'DeepSeek Chat (V3)', tier: 'ultra-cheap', cost: '$0.14 / 1M in', context: '64k', hasTools: true, description: 'DeepSeek-V3 671B MoE via OpenRouter at ultra-low price.' },
-      { id: 'deepseek/deepseek-r1', name: 'DeepSeek Reasoner (R1)', tier: 'reasoning', cost: '$0.55 / 1M in', context: '64k', hasTools: false, description: 'Full DeepSeek-R1 reasoning with Chain-of-Thought output.' },
-      { id: 'meta-llama/llama-3.3-70b-instruct', name: 'Llama 3.3 70B Instruct', tier: 'balanced', cost: '$0.40 / 1M in', context: '128k', hasTools: true, description: 'State-of-the-art open-weights flagship model.' },
-      { id: 'openai/gpt-4o-mini', name: 'GPT-4o Mini', tier: 'ultra-cheap', cost: '$0.15 / 1M in', context: '128k', hasTools: true, description: 'OpenAI mini model routed via OpenRouter.' },
-      { id: 'qwen/qwen-2.5-72b-instruct', name: 'Qwen 2.5 72B Instruct', tier: 'balanced', cost: '$0.35 / 1M in', context: '128k', hasTools: true, description: 'Top-tier open-weights multilingual and coding model.' },
-      { id: 'meta-llama/llama-3.1-8b-instruct:free', name: 'Llama 3.1 8B (Free)', tier: 'free', cost: '$0.00', context: '128k', hasTools: true, description: 'Free tier Llama 3.1 8B on OpenRouter.' },
+      { id: 'openrouter/auto', name: 'OpenRouter Auto Router', tier: 'dynamic', cost: 'Routed model rate', context: 'Model-dependent', hasTools: true, description: 'Recommended default. Chooses from the current catalog so a retired third-party model does not break startup.' },
     ],
   },
-  groq: {
+  groq: {
     id: 'groq',
     name: 'Groq Cloud',
     baseUrl: process.env.GROQ_BASE_URL || 'https://api.groq.com/openai/v1',
     apiKey: process.env.GROQ_API_KEY || '',
-    defaultModel: process.env.GROQ_MODEL || 'llama-3.3-70b-versatile',
+    defaultModel: normalizeProviderModel('groq', process.env.GROQ_MODEL || 'openai/gpt-oss-120b'),
     isLocal: false,
     envKey: 'GROQ_API_KEY',
     docUrl: 'https://console.groq.com/docs/quickstart',
     apiRefUrl: 'https://console.groq.com/docs/api-reference',
-    pricingSummary: 'Ultra-fast LPU inference (llama-3.1-8b: $0.05/1M, llama-3.3-70b: $0.59/1M)',
+    pricingSummary: 'Ultra-fast LPU inference (GPT-OSS 20B: $0.075/1M in, GPT-OSS 120B: $0.15/1M in)',
     features: ['lpu-speed', 'tools', 'streaming', 'openai-compatible'],
-    recommendedModels: [
-      { id: 'llama-3.3-70b-versatile', name: 'Llama 3.3 70B Versatile', tier: 'balanced', cost: '$0.59 / 1M in', context: '128k', hasTools: true, description: 'Recommended default. 70B flagship quality with fast LPU speed and tool use.' },
-      { id: 'llama-3.1-8b-instant', name: 'Llama 3.1 8B Instant', tier: 'ultra-cheap', cost: '$0.05 / 1M in', context: '128k', hasTools: true, description: 'Blazing fast (~800 tokens/s) and ultra-low cost for quick inquiries.' },
-      { id: 'gpt-oss-120b', name: 'GPT-OSS 120B', tier: 'balanced', cost: '$0.59 / 1M in', context: '128k', hasTools: true, description: 'Open-weights 120B flagship model running with ultra-fast Groq LPU inference.' },
-      { id: 'mixtral-8x7b-32768', name: 'Mixtral 8x7B', tier: 'cheap', cost: '$0.24 / 1M in', context: '32k', hasTools: true, description: 'Mistral MoE architecture with 32k context.' },
-      { id: 'gemma2-9b-it', name: 'Gemma 2 9B IT', tier: 'cheap', cost: '$0.20 / 1M in', context: '8k', hasTools: true, description: 'Google Gemma 2 on Groq LPU hardware.' },
-    ],
+    recommendedModels: [
+      { id: 'openai/gpt-oss-120b', name: 'GPT-OSS 120B', tier: 'flagship', cost: '$0.15 / 1M in', context: '131k', hasTools: true, description: 'Recommended default. Production-grade 120B reasoning model with tool use on Groq.' },
+      { id: 'openai/gpt-oss-20b', name: 'GPT-OSS 20B', tier: 'ultra-cheap', cost: '$0.075 / 1M in', context: '131k', hasTools: true, description: 'Fast production model for lower-cost, high-volume requests.' },
+    ],
   },
   deepseek: {
     id: 'deepseek',
     name: 'DeepSeek Cloud',
-    baseUrl: process.env.DEEPSEEK_BASE_URL || 'https://api.deepseek.com/v1',
+    baseUrl: process.env.DEEPSEEK_BASE_URL || 'https://api.deepseek.com',
     apiKey: process.env.DEEPSEEK_API_KEY || '',
-    defaultModel: process.env.DEEPSEEK_MODEL || 'deepseek-chat',
+    defaultModel: normalizeProviderModel('deepseek', process.env.DEEPSEEK_MODEL || 'deepseek-v4-flash'),
     isLocal: false,
     envKey: 'DEEPSEEK_API_KEY',
     docUrl: 'https://api-docs.deepseek.com',
     apiRefUrl: 'https://api-docs.deepseek.com/api/create-chat-completion',
-    pricingSummary: 'Ultra-low cost (deepseek-chat: $0.14 cached / $0.27 uncached / 1M)',
-    features: ['deepseek-v3', 'deepseek-r1-reasoning', 'tools', 'streaming', 'context-cache'],
+    pricingSummary: 'V4 Flash: $0.22-$0.44/1M uncached input; V4 Pro: $0.66-$1.32/1M (off-peak/peak)',
+    features: ['deepseek-v4', 'thinking', 'tools', 'streaming', 'context-cache', '1m-context'],
     recommendedModels: [
-      { id: 'deepseek-chat', name: 'DeepSeek Chat (V3)', tier: 'ultra-cheap', cost: '$0.14-$0.27 / 1M in', context: '64k', hasTools: true, description: 'Recommended default. DeepSeek-V3 671B MoE, top-tier reasoning, tool calling, fast & ultra-cheap.' },
-      { id: 'deepseek-reasoner', name: 'DeepSeek Reasoner (R1)', tier: 'reasoning', cost: '$0.55 / 1M in', context: '64k', hasTools: false, description: 'DeepSeek-R1 full Chain-of-Thought reasoning with native thinking output.' },
+      { id: 'deepseek-v4-flash', name: 'DeepSeek V4 Flash', tier: 'balanced', cost: '$0.22-$0.44 / 1M uncached in', context: '1M', hasTools: true, description: 'Recommended default. Current V4 model with tool calls and thinking/non-thinking modes.' },
+      { id: 'deepseek-v4-pro', name: 'DeepSeek V4 Pro', tier: 'flagship', cost: '$0.66-$1.32 / 1M uncached in', context: '1M', hasTools: true, description: 'Higher-capability V4 model with tool calls and thinking mode.' },
     ],
   },
   gemini: {
@@ -105,18 +122,17 @@ export const CLOUD_PROVIDERS = {
     name: 'Google Gemini Cloud',
     baseUrl: process.env.GEMINI_BASE_URL || 'https://generativelanguage.googleapis.com/v1beta/openai/',
     apiKey: process.env.GEMINI_API_KEY || '',
-    defaultModel: process.env.GEMINI_MODEL || 'gemini-2.5-flash',
+    defaultModel: normalizeProviderModel('gemini', process.env.GEMINI_MODEL || 'gemini-3.7-flash'),
     isLocal: false,
     envKey: 'GEMINI_API_KEY',
-    docUrl: 'https://ai.google.dev/docs',
-    apiRefUrl: 'https://ai.google.dev/api',
-    pricingSummary: 'Ultra-cheap (gemini-2.5-flash: $0.10/1M in, $0.40/1M out)',
+    docUrl: 'https://ai.google.dev/gemini-api/docs',
+    apiRefUrl: 'https://ai.google.dev/api/generate-content',
+    pricingSummary: 'Gemini 3.7 Flash: free tier available; paid standard $0.75/1M input through 2026',
     features: ['1m-context', 'multimodal', 'tools', 'streaming', 'thinking'],
     recommendedModels: [
-      { id: 'gemini-2.5-flash', name: 'Gemini 2.5 Flash', tier: 'ultra-cheap', cost: '$0.10 / 1M in', context: '1M', hasTools: true, description: 'Recommended default. Advanced reasoning and thinking capabilities with 1M context.' },
-      { id: 'gemini-2.5-flash-lite', name: 'Gemini 2.5 Flash Lite', tier: 'lowest-cost', cost: '$0.075 / 1M in', context: '1M', hasTools: true, description: 'Cost-optimized 2.5 lightweight model for high volume queries.' },
-      { id: 'gemini-2.5-pro', name: 'Gemini 2.5 Pro', tier: 'flagship', cost: '$1.25 / 1M in', context: '2M', hasTools: true, description: 'Deep reasoning and 2M context for complex multimodal analysis.' },
-      { id: 'gemini-3.1-pro', name: 'Gemini 3.1 Pro', tier: 'flagship', cost: '$1.50 / 1M in', context: '2M', hasTools: true, description: 'Next-gen Gemini 3.1 flagship with state-of-the-art coding and multimodal reasoning.' },
+      { id: 'gemini-3.7-flash', name: 'Gemini 3.7 Flash', tier: 'latest-stable', cost: '$0.75 / 1M in (2026)', context: '1M', hasTools: true, description: 'Recommended default. Latest stable Flash model for agentic workflows and multimodal reasoning.' },
+      { id: 'gemini-3.5-flash-lite', name: 'Gemini 3.5 Flash-Lite', tier: 'lowest-cost', cost: '$0.30 / 1M in', context: '1M', hasTools: true, description: 'Stable, cost-efficient model for high-volume translation and simple processing.' },
+      { id: 'gemini-3.6-flash', name: 'Gemini 3.6 Flash', tier: 'stable', cost: '$0.75 / 1M in (2026)', context: '1M', hasTools: true, description: 'Stable previous-generation Flash fallback with function calling and thinking.' },
     ],
   },
   custom: {

package/server/agent/doctor.js

@@ -26,10 +26,10 @@ function commandAvailable(command, args = ['--version']) {
   });
 }
 
-async function checkDataDirectory() {
-  const probe = path.join(KIKU_DATA_DIR, `.write-test-${process.pid}`);
+async function checkDataDirectory(dataDir = KIKU_DATA_DIR) {
+  const probe = path.join(dataDir, `.write-test-${process.pid}`);
   try {
-    await fs.mkdir(KIKU_DATA_DIR, { recursive: true, mode: 0o700 });
+    await fs.mkdir(dataDir, { recursive: true, mode: 0o700 });
     await fs.writeFile(probe, 'ok', { mode: 0o600 });
     await fs.unlink(probe);
     return true;
@@ -42,8 +42,9 @@ import { getGpuMetrics } from './gpuMetrics.js';
 
 export async function runDoctor(options = {}) {
   const client = options.client;
+  const dataDir = options.dataDir || KIKU_DATA_DIR;
   const [dataWritable, speaker, ffmpegStatic, fluentFfmpeg, ytDlp, lm, gpu] = await Promise.all([
-    checkDataDirectory(),
+    checkDataDirectory(dataDir),
     optionalImport('speaker'),
     optionalImport('ffmpeg-static'),
     optionalImport('fluent-ffmpeg'),
@@ -54,7 +55,7 @@ export async function runDoctor(options = {}) {
   const nodeMajor = Number(process.versions.node.split('.')[0]);
   const checks = [
     { id: 'node', required: true, ok: nodeMajor >= 20, detail: `Node ${process.versions.node} (requires >=20)` },
-    { id: 'data', required: true, ok: dataWritable, detail: KIKU_DATA_DIR },
+    { id: 'data', required: true, ok: dataWritable, detail: dataDir },
     { id: 'gpu', required: false, ok: Boolean(gpu.available), detail: gpu.available && gpu.primary ? `${gpu.primary.name} (${gpu.primary.totalGb} GB VRAM${gpu.primary.driverVersion !== 'N/A' ? `, Driver ${gpu.primary.driverVersion}` : ''})` : 'CPU fallback / no dedicated acceleration detected' },
     { id: 'lmstudio', required: false, ok: Boolean(lm.connected), detail: lm.connected ? `${lm.models?.length || 0} loaded model(s)` : 'offline; catalog-only mode is available' },
     { id: 'ffmpeg', required: false, ok: ffmpegStatic && fluentFfmpeg, detail: 'audio add-on only; intentionally not installed with the core CLI' },
@@ -64,6 +65,6 @@ export async function runDoctor(options = {}) {
   return {
     ok: checks.every(check => !check.required || check.ok),
     checks,
-    dataDir: KIKU_DATA_DIR,
+    dataDir,
   };
 }

package/server/agent/llmClient.js

@@ -1,9 +1,12 @@
-import { config, CLOUD_PROVIDERS } from './config.js';
+import { config, CLOUD_PROVIDERS, normalizeProviderModel } from './config.js';
 
 export class LLMClient {
   constructor(options = {}) {
-    const providerId = (options.provider || config.activeProvider || 'lmstudio').toLowerCase();
-    const providerDef = CLOUD_PROVIDERS[providerId] || CLOUD_PROVIDERS.lmstudio;
+    const providerId = (options.provider || config.activeProvider || 'lmstudio').toLowerCase();
+    const providerDef = CLOUD_PROVIDERS[providerId];
+    if (!providerDef) {
+      throw new Error(`Unknown provider "${providerId}". Supported: ${Object.keys(CLOUD_PROVIDERS).join(', ')}`);
+    }
 
     this.provider = providerId;
     this.providerName = providerDef.name || providerId;
@@ -11,7 +14,10 @@ export class LLMClient {
     this.apiKey = options.apiKey !== undefined
       ? options.apiKey
       : (providerDef.apiKey !== undefined ? providerDef.apiKey : (providerDef.isLocal ? (config.lmStudio.apiKey || 'lm-studio') : ''));
-    this.model = options.model || providerDef.defaultModel || config.lmStudio.model || 'google/gemma-4-e4b';
+    this.model = normalizeProviderModel(
+      providerId,
+      options.model || providerDef.defaultModel || config.lmStudio.model || 'google/gemma-4-e4b'
+    );
     this.timeout = options.timeout || config.lmStudio.requestTimeoutMs || 180000;
     this.extraHeaders = { ...(providerDef.extraHeaders || {}), ...(options.extraHeaders || {}) };
   }
@@ -27,7 +33,7 @@ export class LLMClient {
     this.providerName = providerDef.name || pId;
     this.baseUrl = (options.baseUrl || providerDef.baseUrl).replace(/\/+$/, '');
     this.apiKey = options.apiKey !== undefined ? options.apiKey : (providerDef.apiKey || '');
-    this.model = options.model || providerDef.defaultModel;
+    this.model = normalizeProviderModel(pId, options.model || providerDef.defaultModel);
     this.extraHeaders = { ...(providerDef.extraHeaders || {}), ...(options.extraHeaders || {}) };
     return this.getProviderInfo();
   }
@@ -40,7 +46,7 @@ export class LLMClient {
   setModel(model) {
     const trimmed = String(model || '').trim();
     if (!trimmed) throw new Error('A model ID is required.');
-    this.model = trimmed;
+    this.model = normalizeProviderModel(this.provider, trimmed);
     return this.model;
   }
 
@@ -187,8 +193,9 @@ export class LLMClient {
     onToken = null,
     onThought = null
   }) {
-    const isReasoningModel = /^(o1|o3|deepseek-reasoner)/i.test(this.model) || this.model.includes('reasoner');
-    const isO1OrO3 = /^(o1|o3)/i.test(this.model);
+    const isLegacyReasoningModel = /^(o1|o3|deepseek-reasoner)/i.test(this.model) || this.model.includes('reasoner');
+    const usesCompletionTokenLimit = /^(o1|o3|gpt-5(?:\.|-|$))/i.test(this.model);
+    const rejectsTemperature = /^(o1|o3|gpt-5(?:\.|-|$))/i.test(this.model);
 
     const buildPayload = (includeTools = true, includeTemp = true) => {
       const p = {
@@ -197,17 +204,17 @@ export class LLMClient {
         stream
       };
 
-      if (!isO1OrO3 && includeTemp && temperature !== undefined && temperature !== null) {
-        p.temperature = temperature;
-      }
-
-      if (isO1OrO3) {
-        p.max_completion_tokens = maxTokens;
-      } else {
-        p.max_tokens = maxTokens;
-      }
-
-      if (includeTools && tools && Array.isArray(tools) && tools.length > 0 && !isReasoningModel) {
+      if (!rejectsTemperature && includeTemp && temperature !== undefined && temperature !== null) {
+        p.temperature = temperature;
+      }
+
+      if (usesCompletionTokenLimit) {
+        p.max_completion_tokens = maxTokens;
+      } else {
+        p.max_tokens = maxTokens;
+      }
+
+      if (includeTools && tools && Array.isArray(tools) && tools.length > 0 && !isLegacyReasoningModel) {
         p.tools = tools;
         p.tool_choice = "auto";
       }