kiku-agent 1.2.0 → 1.3.0

package/server/web-chat/core/connection.js

@@ -1,194 +1,218 @@
-import fs from 'node:fs';
-import { createHash } from 'node:crypto';
-import { spawn } from 'node:child_process';
-import { LLMClient, LMStudioClient } from '../../agent/llmClient.js';
-import { KikuAgent } from '../../agent/agentCore.js';
-import { AgentMemory } from '../../agent/memory.js';
-import { ModelManager } from '../../agent/modelManager.js';
-import { config } from '../../agent/config.js';
-import { PUBLIC_REMOTE_TOOL_NAMES } from '../../agent/publicTools.js';
-
-export const webAgentCache = new Map();
-export let sharedWebMemory = new AgentMemory({ persistence: false });
-export let currentAgent = null;
-const MAX_WEB_AGENT_SESSIONS = 24;
-
-export function setSharedWebMemory(mem) {
-  if (mem) sharedWebMemory = mem;
-}
-
-export function setCurrentAgent(agent) {
-  if (agent) {
-    currentAgent = agent;
-    if (agent.memory) sharedWebMemory = agent.memory;
-  }
-}
-
-export function normalizeWebSessionId(value) {
-  const normalized = String(value || '').trim().replace(/[^a-zA-Z0-9_-]+/g, '_').slice(0, 100);
-  return normalized || null;
-}
-
-export function credentialFingerprint(value) {
-  if (!value) return 'no-key';
-  return createHash('sha256').update(String(value)).digest('hex').slice(0, 12);
-}
-
-export function computeSessionHash(messages) {
-  if (!Array.isArray(messages) || messages.length === 0) return '';
-  const payload = messages.map(m => (m.role || '') + ':' + (m.parts?.[0]?.content || m.text || '')).join('|');
-  return createHash('sha256').update(payload).digest('hex').slice(0, 16);
-}
-
-export function getWebAgentCacheKey({ sessionId, provider, model, apiKey }) {
-  if (!sessionId) return null;
-  return [sessionId, provider || 'lmstudio', model || 'default', credentialFingerprint(apiKey)].join(':');
-}
-
-export async function prepareWebAgent(agent, options = {}) {
-  await agent.ready();
-  if (Array.isArray(options.history) && options.history.length > 0) {
-    const agentMsgCount = agent.session?.messages?.length || 0;
-    const frontendMsgCount = options.history.length;
-    await agent.hydrateConversationHistory(options.history, {
-      sessionId: options.sessionId,
-      maxMessages: 24,
-      force: frontendMsgCount > agentMsgCount,
-    });
-  }
-  return agent;
-}
-
-export function cacheWebAgentConnection(cacheKey, connection) {
-  if (!cacheKey) return;
-  webAgentCache.delete(cacheKey);
-  webAgentCache.set(cacheKey, connection);
-  while (webAgentCache.size > MAX_WEB_AGENT_SESSIONS) {
-    const oldestKey = webAgentCache.keys().next().value;
-    const oldest = webAgentCache.get(oldestKey);
-    oldest?.agent?.saveSession?.().catch(() => {});
-    const msgCount = oldest?.agent?.session?.messages?.length || 0;
-    if (msgCount === 0) {
-      const storageFile = oldest?.agent?.session?.storageFile;
-      if (storageFile) {
-        fs.promises.unlink(storageFile).catch(() => {});
-      }
-    }
-    webAgentCache.delete(oldestKey);
-  }
-}
-
-export async function waitForLMStudio(baseUrl, timeoutMs = 8000) {
-  const client = new LMStudioClient({ baseUrl });
-  const start = Date.now();
-  while (Date.now() - start < timeoutMs) {
-    try {
-      const health = await client.checkConnection();
-      if (health.connected) return health;
-    } catch {}
-    await new Promise(r => setTimeout(r, 600));
-  }
-  return null;
-}
-
-export async function attemptAutoStartLMStudio(baseUrl, model) {
-  const isWin = process.platform === 'win32';
-  const isMac = process.platform === 'darwin';
-
-  const candidates = isWin
-    ? [
-      `${process.env.LOCALAPPDATA}\\LM Studio\\LM Studio.exe`,
-      `${process.env.LOCALAPPDATA}\\Programs\\LM Studio\\LM Studio.exe`,
-      'C:\\Program Files\\LM Studio\\LM Studio.exe'
-    ]
-    : isMac
-      ? ['/Applications/LM Studio.app/Contents/MacOS/LM Studio']
-      : ['lmstudio', '/usr/local/bin/lmstudio', '/opt/lmstudio/bin/lmstudio'];
-
-  const exe = candidates.find(p => {
-    if (p.includes('/') || p.includes('\\')) {
-      try { return fs.existsSync(p); } catch { return false; }
-    }
-    return true;
-  });
-
-  if (!exe) return null;
-
-  try {
-    let port = '1234';
-    try { const parsedUrl = new URL(baseUrl); if (parsedUrl.port) port = parsedUrl.port; } catch {}
-
-    const child = spawn(exe, isWin
-      ? ['--model', model || 'google/gemma-4-e4b', '--serve', '--port', port]
-      : ['serve', '--model', model || 'google/gemma-4-e4b', '--port', port],
-      { detached: true, stdio: 'ignore', windowsHide: true }
-    );
-    child.unref();
-
-    return await waitForLMStudio(baseUrl, 10000);
-  } catch {
-    return null;
-  }
-}
-
-export async function resolveAgentConnection(customOptions = {}) {
-  const provider = customOptions.provider || 'lmstudio';
-  const requestedModel = customOptions.model || config.providers?.[provider]?.model || 'google/gemma-4-e4b';
-  const baseUrl = customOptions.baseUrl || config.providers?.[provider]?.baseUrl || 'http://127.0.0.1:1234/v1';
-  const sessionId = normalizeWebSessionId(customOptions.sessionId);
-  // A remotely reachable endpoint must never inherit provider credentials from
-  // the server process. Only an explicitly supplied per-request key is used.
-  const apiKey = customOptions.apiKey || '';
-
-  const client = new LLMClient({
-    provider,
-    baseUrl,
-    model: requestedModel,
-    apiKey
-  });
-
-  const modelManager = new ModelManager(client, { requestedModel });
-  let resolution = await modelManager.resolve({ requestedModel });
-
-  if (provider === 'lmstudio' && resolution.state === 'server_offline' && process.env.NODE_ENV !== 'test' && !process.execArgv.includes('--test')) {
-    const started = await attemptAutoStartLMStudio(baseUrl, requestedModel);
-    if (started) {
-      resolution = await modelManager.resolve({ requestedModel });
-    }
-  }
-
-  if (resolution.connected && resolution.selectedModel) {
-    const agent = new KikuAgent({
-      client,
-      thinkingMode: customOptions.thinking || 'show',
-      memory: new AgentMemory({ persistence: false }),
-      sessionId,
-      sessionPersistence: false,
-      allowedToolNames: PUBLIC_REMOTE_TOOL_NAMES,
-    });
-    await prepareWebAgent(agent, { history: customOptions.history, sessionId });
-    const connection = {
-      ready: true,
-      agent,
-      client,
-      provider,
-      model: resolution.selectedModel,
-      models: resolution.models || [],
-      usedFallback: resolution.usedFallback || false,
-      resolution,
-      contentHash: computeSessionHash(agent.session?.messages || [])
-    };
-    return connection;
-  }
-
-  return {
-    ready: false,
-    client,
-    provider,
-    model: requestedModel,
-    models: resolution.models || [],
-    state: resolution.state,
-    error: resolution.error || 'Server offline',
-    resolution
-  };
-}
+import fs from 'node:fs';
+import { createHash } from 'node:crypto';
+import { spawn } from 'node:child_process';
+import { LLMClient } from '../../agent/llmClient.js';
+import { KikuAgent } from '../../agent/agentCore.js';
+import { AgentMemory } from '../../agent/memory.js';
+import { ModelManager } from '../../agent/modelManager.js';
+import { config, getProviderBaseUrl, getPersistedActiveProvider } from '../../agent/config.js';
+import { PUBLIC_REMOTE_TOOL_NAMES } from '../../agent/publicTools.js';
+
+export const webAgentCache = new Map();
+export let sharedWebMemory = new AgentMemory({ persistence: false });
+export let currentAgent = null;
+const MAX_WEB_AGENT_SESSIONS = 24;
+
+export function setSharedWebMemory(mem) {
+  if (mem) sharedWebMemory = mem;
+}
+
+export function setCurrentAgent(agent) {
+  if (agent) {
+    currentAgent = agent;
+    if (agent.memory) sharedWebMemory = agent.memory;
+  }
+}
+
+export function normalizeWebSessionId(value) {
+  const normalized = String(value || '').trim().replace(/[^a-zA-Z0-9_-]+/g, '_').slice(0, 100);
+  return normalized || null;
+}
+
+export function credentialFingerprint(value) {
+  if (!value) return 'no-key';
+  return createHash('sha256').update(String(value)).digest('hex').slice(0, 12);
+}
+
+export function computeSessionHash(messages) {
+  if (!Array.isArray(messages) || messages.length === 0) return '';
+  const payload = messages.map(m => (m.role || '') + ':' + (m.parts?.[0]?.content || m.text || '')).join('|');
+  return createHash('sha256').update(payload).digest('hex').slice(0, 16);
+}
+
+export function getWebAgentCacheKey({ sessionId, provider, model, apiKey }) {
+  if (!sessionId) return null;
+  return [sessionId, provider || 'lmstudio', model || 'default', credentialFingerprint(apiKey)].join(':');
+}
+
+export async function prepareWebAgent(agent, options = {}) {
+  await agent.ready();
+  if (Array.isArray(options.history) && options.history.length > 0) {
+    const agentMsgCount = agent.session?.messages?.length || 0;
+    const frontendMsgCount = options.history.length;
+    await agent.hydrateConversationHistory(options.history, {
+      sessionId: options.sessionId,
+      maxMessages: 24,
+      force: frontendMsgCount > agentMsgCount,
+    });
+  }
+  return agent;
+}
+
+export function cacheWebAgentConnection(cacheKey, connection) {
+  if (!cacheKey) return;
+  webAgentCache.delete(cacheKey);
+  webAgentCache.set(cacheKey, connection);
+  while (webAgentCache.size > MAX_WEB_AGENT_SESSIONS) {
+    const oldestKey = webAgentCache.keys().next().value;
+    const oldest = webAgentCache.get(oldestKey);
+    oldest?.agent?.saveSession?.().catch(() => {});
+    const msgCount = oldest?.agent?.session?.messages?.length || 0;
+    if (msgCount === 0) {
+      const storageFile = oldest?.agent?.session?.storageFile;
+      if (storageFile) {
+        fs.promises.unlink(storageFile).catch(() => {});
+      }
+    }
+    webAgentCache.delete(oldestKey);
+  }
+}
+
+export async function waitForLMStudio(baseUrl, timeoutMs = 8000) {
+  const client = new LLMClient({ provider: 'lmstudio', baseUrl });
+  const start = Date.now();
+  while (Date.now() - start < timeoutMs) {
+    try {
+      const health = await client.checkConnection();
+      if (health.connected) return health;
+    } catch {}
+    await new Promise(r => setTimeout(r, 600));
+  }
+  return null;
+}
+
+export async function attemptAutoStartLMStudio(baseUrl, model) {
+  const isWin = process.platform === 'win32';
+  const isMac = process.platform === 'darwin';
+
+  const candidates = isWin
+    ? [
+      `${process.env.LOCALAPPDATA}\\LM Studio\\LM Studio.exe`,
+      `${process.env.LOCALAPPDATA}\\Programs\\LM Studio\\LM Studio.exe`,
+      'C:\\Program Files\\LM Studio\\LM Studio.exe'
+    ]
+    : isMac
+      ? ['/Applications/LM Studio.app/Contents/MacOS/LM Studio']
+      : ['lmstudio', '/usr/local/bin/lmstudio', '/opt/lmstudio/bin/lmstudio'];
+
+  const exe = candidates.find(p => {
+    if (p.includes('/') || p.includes('\\')) {
+      try { return fs.existsSync(p); } catch { return false; }
+    }
+    return true;
+  });
+
+  if (!exe) return null;
+
+  try {
+    let port = '1234';
+    try { const parsedUrl = new URL(baseUrl); if (parsedUrl.port) port = parsedUrl.port; } catch {}
+
+    const safeModel = (typeof model === 'string' && /^[a-zA-Z0-9_\-\.\/]+$/.test(model) && !model.startsWith('-'))
+      ? model
+      : 'google/gemma-4-e4b';
+    const safePort = (/^\d{1,5}$/.test(port) && Number(port) > 0 && Number(port) <= 65535) ? port : '1234';
+
+    const child = spawn(exe, isWin
+      ? ['--model', safeModel, '--serve', '--port', safePort]
+      : ['serve', '--model', safeModel, '--port', safePort],
+      { detached: true, stdio: 'ignore', windowsHide: true }
+    );
+    child.unref();
+
+    return await waitForLMStudio(baseUrl, 10000);
+  } catch {
+    return null;
+  }
+}
+
+export async function resolveAgentConnection(customOptions = {}) {
+  const isAuto = !customOptions.provider || customOptions.provider === 'auto';
+  const serverProvider = currentAgent?.client?.provider || config.activeProvider || 'lmstudio';
+  const provider = isAuto ? serverProvider : customOptions.provider;
+
+  const serverModel = currentAgent?.client?.model || config.providers?.[provider]?.model || 'google/gemma-4-e4b';
+  const requestedModel = (customOptions.model && !isAuto)
+    ? customOptions.model
+    : (provider === serverProvider ? serverModel : (config.providers?.[provider]?.model || 'google/gemma-4-e4b'));
+
+  const isServerAgentMatch = Boolean(currentAgent?.client && provider === currentAgent.client.provider);
+  const baseUrl = (typeof customOptions.baseUrl === 'string' && customOptions.baseUrl.trim()
+    ? customOptions.baseUrl.trim().replace(/\/+$/, '')
+    : null)
+    || (isServerAgentMatch ? currentAgent.client.baseUrl : null)
+    || getProviderBaseUrl(provider);
+
+  const sessionId = normalizeWebSessionId(customOptions.sessionId);
+ 
+  const apiKey = customOptions.apiKey
+    || (isServerAgentMatch ? currentAgent.client.apiKey : '')
+    || '';
+
+  const client = new LLMClient({
+    provider,
+    baseUrl,
+    model: requestedModel,
+    apiKey
+  });
+
+  const modelManager = new ModelManager(client, { requestedModel });
+  let resolution = await modelManager.resolve({ requestedModel });
+
+  if (provider === 'lmstudio' && resolution.state === 'server_offline' && process.env.NODE_ENV !== 'test' && !process.execArgv.includes('--test')) {
+    const started = await attemptAutoStartLMStudio(baseUrl, requestedModel);
+    if (started) {
+      resolution = await modelManager.resolve({ requestedModel });
+    }
+  }
+
+  if (provider === 'vllm' && resolution.state === 'server_offline' && !resolution.error?.includes('WSL')) {
+    const wslHint = 'vLLM runs on WSL/Linux port 8000 (Windows cannot host it natively). Start it with: wsl -d Ubuntu -- vllm serve <model> --port 8000';
+    resolution.error = resolution.error ? `${resolution.error}\n  Notice: ${wslHint}` : wslHint;
+  }
+
+  if (resolution.connected && resolution.selectedModel) {
+    const agent = new KikuAgent({
+      client,
+      thinkingMode: customOptions.thinking || 'show',
+      memory: new AgentMemory({ persistence: false }),
+      sessionId,
+      sessionPersistence: false,
+      allowedToolNames: PUBLIC_REMOTE_TOOL_NAMES,
+    });
+    await prepareWebAgent(agent, { history: customOptions.history, sessionId });
+    const connection = {
+      ready: true,
+      agent,
+      client,
+      provider,
+      model: resolution.selectedModel,
+      models: resolution.models || [],
+      usedFallback: resolution.usedFallback || false,
+      resolution,
+      contentHash: computeSessionHash(agent.session?.messages || [])
+    };
+    return connection;
+  }
+
+  return {
+    ready: false,
+    client,
+    provider,
+    model: requestedModel,
+    models: resolution.models || [],
+    state: resolution.state,
+    error: resolution.error || 'Server offline',
+    resolution
+  };
+}

package/server/web-chat/core/conversationContext.js

@@ -13,7 +13,7 @@ const DOMAIN_PATTERNS = {
   fashion_event: /(?:vogue|gala|red\s*carpet|fashion|แฟชั่น|กาล่า|พรมแดง|งานแฟชั่น|盛典|红毯|时尚)/i,
   drama: /(?:ซีรีส์|ละคร|ภาพยนตร์|หนัง|นักแสดง|บทบาท|ตัวละคร|in\s*blossom|yunxi|ฮวาเจียน|หยุนซี|剧|电视剧|电影|角色)/i,
   music: /(?:เพลง|อัลบั้ม|ดนตรี|ost|single|concert|คอนเสิร์ต|music|song|歌曲|音乐|演唱会)/i,
-  brand: /(?:แบรนด์|พรีเซนเตอร์|ambassador|endorsement|versace|luxury|代言|品牌)/i,
+  brand: /(?:แบรนด์|พรีเซนเตอร์|พรีเซ็นเตอร์|โฆษณา|แคมเปญ|โปรโมท|โปรโมต|การโปรโมท|ambassador|endorsement|commercial|cf|promo|campaign|versace|prsr|pasha|fred|luxury|代言|品牌|商务|推广)/i,
   awards: /(?:รางวัล|เกียรติยศ|award|ceremony|获奖|奖项)/i,
   profile: /(?:ประวัติ|จวีจิ้งอี|ju\s*jingyi|kiku|鞠婧祎|profile|who\s+is)/i
 };