diff --git a/assets/kimi.png b/assets/kimi.png new file mode 100644 index 0000000..2839282 Binary files /dev/null and b/assets/kimi.png differ diff --git a/cli/proxima-cli.cjs b/cli/proxima-cli.cjs index 1934cf0..9e673e4 100644 --- a/cli/proxima-cli.cjs +++ b/cli/proxima-cli.cjs @@ -119,7 +119,7 @@ function startSpinner(text) { const FALLBACK_INFO = { mode: 'session', - providers: ['chatgpt', 'claude', 'gemini', 'perplexity'], + providers: ['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi'], models: {}, firstProvider: 'claude' }; diff --git a/electron/api/byok/keys.cjs b/electron/api/byok/keys.cjs index 3e61b3f..50071e7 100644 --- a/electron/api/byok/keys.cjs +++ b/electron/api/byok/keys.cjs @@ -11,7 +11,7 @@ const ENC_VERSION = 1; let _storeCache = null; const KNOWN_PROVIDERS = Object.freeze([ - 'chatgpt', 'claude', 'gemini', 'perplexity', + 'chatgpt', 'claude', 'gemini', 'perplexity', 'kimi', 'deepseek', 'groq', 'xai', 'openrouter', 'together', 'fireworks', 'mistral', 'nvidia', ]); diff --git a/electron/api/byok/model-fetcher.cjs b/electron/api/byok/model-fetcher.cjs index e28e93f..09244ed 100644 --- a/electron/api/byok/model-fetcher.cjs +++ b/electron/api/byok/model-fetcher.cjs @@ -25,6 +25,10 @@ const MODELS_CONFIG = Object.freeze({ 'sonar-deep-research', 'r1-1776', ], }, + kimi: { + url: 'https://api.moonshot.cn/v1/models', + auth: 'bearer', + }, deepseek: { url: 'https://api.deepseek.com/v1/models', auth: 'bearer', diff --git a/electron/api/byok/models.cjs b/electron/api/byok/models.cjs index 73461b2..3e4f481 100644 --- a/electron/api/byok/models.cjs +++ b/electron/api/byok/models.cjs @@ -8,6 +8,7 @@ const FALLBACK_MODELS = Object.freeze({ claude: ['claude-sonnet-5', 'claude-3-5-sonnet-latest', 'claude-3-haiku-20240307'], gemini: ['gemini-3.5-flash', 'gemini-3.1-pro', 'gemini-2.0-flash-lite'], perplexity: ['sonar-pro', 'sonar'], + kimi: ['moonshot-v1-32k', 'moonshot-v1-8k'], deepseek: ['deepseek-chat'], groq: ['llama-3.3-70b-versatile', 'llama-3.1-8b-instant'], xai: ['grok-2-1212', 'grok-beta'], @@ -23,6 +24,7 @@ const DEFAULT_MODELS = Object.freeze({ claude: FALLBACK_MODELS.claude[0], gemini: FALLBACK_MODELS.gemini[0], perplexity: FALLBACK_MODELS.perplexity[0], + kimi: FALLBACK_MODELS.kimi[0], deepseek: FALLBACK_MODELS.deepseek[0], groq: FALLBACK_MODELS.groq[0], xai: FALLBACK_MODELS.xai[0], @@ -44,6 +46,7 @@ const API_ENDPOINTS = Object.freeze({ claude: 'https://api.anthropic.com/v1/messages', google: 'https://generativelanguage.googleapis.com/v1beta/models', perplexity: 'https://api.perplexity.ai/chat/completions', + kimi: 'https://api.moonshot.cn/v1/chat/completions', deepseek: 'https://api.deepseek.com/v1/chat/completions', groq: 'https://api.groq.com/openai/v1/chat/completions', xai: 'https://api.x.ai/v1/chat/completions', @@ -61,6 +64,7 @@ const MAX_TOKENS = Object.freeze({ claude: 4096, gemini: 8192, perplexity: 4096, + kimi: 8192, deepseek: 4096, groq: 4096, xai: 4096, diff --git a/electron/api/byok/router.cjs b/electron/api/byok/router.cjs index 2883a35..4655d1d 100644 --- a/electron/api/byok/router.cjs +++ b/electron/api/byok/router.cjs @@ -16,6 +16,7 @@ const CALLERS = Object.freeze({ claude: anthropic, gemini: google, perplexity: perplexity, + kimi: compatible, deepseek: compatible, groq: compatible, xai: compatible, diff --git a/electron/api/pages/widget.cjs b/electron/api/pages/widget.cjs index e44a472..23d6f9d 100644 --- a/electron/api/pages/widget.cjs +++ b/electron/api/pages/widget.cjs @@ -74,7 +74,7 @@ function getChatJS() { } catch (e) { console.warn('[Widget] Could not fetch models, using defaults:', e.message); _gatewayMode = 'session'; - _enabledProviders = ['chatgpt', 'claude', 'gemini', 'perplexity']; + _enabledProviders = ['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi']; } populateDropdown(); populateBattle(); diff --git a/electron/api/rest-api.cjs b/electron/api/rest-api.cjs index 94c7b8d..3b429d6 100644 --- a/electron/api/rest-api.cjs +++ b/electron/api/rest-api.cjs @@ -29,6 +29,7 @@ const MODEL_ALIASES = { 'gemini': 'gemini', 'gemini-pro': 'gemini', 'gemini-2': 'gemini', 'gemini-2.5': 'gemini', 'google': 'gemini', 'bard': 'gemini', 'perplexity': 'perplexity', 'pplx': 'perplexity', 'sonar': 'perplexity', + 'kimi': 'kimi', 'moonshot': 'kimi', 'moonshotai': 'kimi', 'auto': 'auto', 'all': 'all' }; @@ -342,7 +343,7 @@ function pickBestProvider(preferred) { return enabled.includes(base) ? base : null; } - const priorityList = ['claude', 'chatgpt', 'gemini', 'perplexity', 'deepseek', 'groq', 'xai', 'openrouter', 'together', 'fireworks', 'mistral', 'nvidia']; + const priorityList = ['claude', 'chatgpt', 'gemini', 'perplexity', 'kimi', 'deepseek', 'groq', 'xai', 'openrouter', 'together', 'fireworks', 'mistral', 'nvidia']; const priority = priorityList.filter(p => enabled.includes(p)); if (priority.length === 0) { return enabled.length > 0 ? enabled[0] : null; @@ -566,7 +567,7 @@ function startRestAPI() { } const url = new URL(req.url, `http://localhost:${REST_PORT}`); - const publicPaths = ['/', '/docs', '/cli', '/ws', '/websocket', '/openapi.json', '/v1/openapi.json', '/api-key', '/v1/byok/models', '/v1/models', '/v1/stats', '/v1/functions']; + const publicPaths = ['/', '/docs', '/cli', '/ws', '/websocket', '/openapi.json', '/v1/openapi.json', '/api-key', '/v1/byok/models', '/v1/models', '/v1/stats', '/v1/functions', '/v1/kimi/session']; if (!publicPaths.includes(url.pathname) && !validateApiKey(req)) { return sendError(res, 401, 'Invalid or missing API key', 'authentication_error'); } @@ -574,7 +575,7 @@ function startRestAPI() { const body = req.method === 'POST' ? await parseBody(req) : {}; const providerKey = (req.headers['x-provider-key'] || '').trim(); if (providerKey) body._byokKey = providerKey; - await handleRoute(req.method, url.pathname, body, res); + await handleRoute(req.method, url.pathname, body, res, req.headers, url.searchParams); } catch (err) { console.error('[API] Error:', err.message); if (!res.headersSent) sendError(res, err.statusCode || 500, err.message); } }); diff --git a/electron/api/routes.cjs b/electron/api/routes.cjs index 5fe239c..674f297 100644 --- a/electron/api/routes.cjs +++ b/electron/api/routes.cjs @@ -15,9 +15,11 @@ function createRouteHandler(deps) { byok, } = deps; - return async function handleRoute(method, pathname, body, res) { + return async function handleRoute(method, pathname, body, res, headers, searchParams) { const handleMCPRequest = getHandler(); - const conversationId = (body && (body.conversationId || body.conversation_id || body.sessionId || body.session_id)) || null; + const conversationId = (body && (body.conversationId || body.conversation_id || body.sessionId || body.session_id)) + || (headers && (headers['x-proxima-conversation'] || headers['X-Proxima-Conversation'])) + || null; if (method === 'POST' && pathname === `${API_PREFIX}/chat/completions`) { const fn = (body.function || '').toLowerCase().trim(); @@ -253,10 +255,17 @@ function createRouteHandler(deps) { else sendJSON(res, 200, formatChatResponse(result, provider)); } } else { - const toolPrompt = buildToolCallingPrompt(body); - if (!toolPrompt) return sendError(res, 400, 'No message provided'); + // Kimi's engine needs the raw messages array (it reads the + // system prompt to detect a NEW chat and only forwards the + // latest user turn on follow-ups). Flattening to a string + // via buildToolCallingPrompt would lose the system prompt + // and make the engine reuse a stale/too-long chat_id. + const inputForEngine = provider === 'kimi' + ? (Array.isArray(body.messages) ? body.messages : message) + : buildToolCallingPrompt(body); + if (!inputForEngine) return sendError(res, 400, 'No message provided'); if (TOOL_DEBUG) console.log('[TOOL-CALL/Legacy] Provider:', provider, '| Tools:', body.tools.length); - const result = await queryProvider(provider, toolPrompt, filePath, gemini, null, conversationId, body._byokKey, null, resolved.byokModelId); + const result = await queryProvider(provider, inputForEngine, filePath, gemini, null, conversationId, body._byokKey, null, resolved.byokModelId); if (TOOL_DEBUG) console.log('[TOOL-CALL/Legacy] Response text (first 300):', JSON.stringify(result.text?.slice(0, 300))); const parsed = parseToolCallResponse(result.text); if (TOOL_DEBUG) console.log('[TOOL-CALL/Legacy] Parsed:', parsed.isToolCall, '| toolCalls:', parsed.toolCalls?.length || 0); @@ -283,7 +292,12 @@ function createRouteHandler(deps) { if (resolved.mode === 'single') { const provider = resolved.providers[0]; const isByokActive = !!(body._byokKey || (byok.keys.isEnabled() ? byok.keys.getKey(provider) : null)); - const inputMessage = isByokActive ? (body.messages || message) : message; + // Kimi's engine needs the full messages array so it can send the + // system prompt + context once at chat start and only the new + // user turn afterwards (parent_id chain carries the rest). + const inputMessage = isByokActive + ? (body.messages || message) + : (provider === 'kimi' && Array.isArray(body.messages) ? body.messages : message); let onChunk = null; let streamedAny = false; @@ -373,7 +387,7 @@ function createRouteHandler(deps) { aliases: Object.entries(MODEL_ALIASES).filter(([_, v]) => v === p).map(([k]) => k).filter(k => k !== p), }); } - ['chatgpt', 'claude', 'gemini', 'perplexity'].filter(p => !enabled.includes(p)).forEach(p => + ['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi'].filter(p => !enabled.includes(p)).forEach(p => models.push({ id: p, object: 'model', owned_by: 'proxima', status: 'disabled' }) ); } @@ -395,7 +409,7 @@ function createRouteHandler(deps) { function: null, name: 'chat', description: 'Default chat completion. Omit the "function" field to chat with a model.', - fields: { model: 'auto|claude|chatgpt|gemini|perplexity|all (or array of models)', message: 'string (or use messages[])', stream: 'boolean (optional)', filePath: 'local file path for multimodal/Gemini routing (optional)' } + fields: { model: 'auto|claude|chatgpt|gemini|perplexity|kimi|all (or array of models)', message: 'string (or use messages[])', stream: 'boolean (optional)', filePath: 'local file path for multimodal/Gemini routing (optional)' } }, { function: 'search', @@ -472,7 +486,7 @@ function createRouteHandler(deps) { if (method === 'POST' && pathname === `${API_PREFIX}/conversations/new`) { const requested = (body && (body.provider || body.model)) || null; if (!requested) { - return sendError(res, 400, 'A "provider" (or "model") is required: chatgpt, claude, gemini, or perplexity.'); + return sendError(res, 400, 'A "provider" (or "model") is required: chatgpt, claude, gemini, perplexity, or kimi.'); } const target = pickBestProvider(requested); if (!target) { @@ -482,6 +496,20 @@ function createRouteHandler(deps) { catch (e) { return sendError(res, 500, e.message); } } + if (method === 'GET' && pathname === `${API_PREFIX}/kimi/session`) { + // Session-status endpoint for the opencode plugin. The plugin polls this + // before each request; when resetCount changes, it knows Kimi rolled to + // a NEW chat_id and must re-send the system prompt + handoff summary. + const cid = (searchParams && (searchParams.get('conversationId') || searchParams.get('conversation_id') || searchParams.get('sessionId'))) + || (headers && headers['x-proxima-conversation']) + || null; + try { + const r = await handleMCPRequest({ action: 'getKimiSessionStatus', provider: 'kimi', data: { conversationId: cid } }); + if (!r || r.success === false) return sendJSON(res, 200, { success: false, resetCount: 0, chatId: null, parentId: null, error: (r && r.error) || 'unavailable' }); + return sendJSON(res, 200, { success: true, resetCount: r.resetCount || 0, chatId: r.chatId || null, parentId: r.parentId || null }); + } catch (e) { return sendJSON(res, 200, { success: false, resetCount: 0, chatId: null, parentId: null, error: e.message }); } + } + if (method === 'GET' && (pathname === `${API_PREFIX}/openapi.json` || pathname === '/openapi.json')) { try { return sendJSON(res, 200, JSON.parse(require('fs').readFileSync(require('path').join(__dirname, '..', '..', 'docs', 'openapi.json'), 'utf8'))); } catch (e) { return sendError(res, 500, 'OpenAPI spec not found'); } diff --git a/electron/api/ws-server.cjs b/electron/api/ws-server.cjs index cf0156e..4b0fb9a 100644 --- a/electron/api/ws-server.cjs +++ b/electron/api/ws-server.cjs @@ -154,7 +154,7 @@ function closeWebSocket() { } } -async function queryProvider(provider, message, filePath = null, engine = null, conversationId = null, modelId = null) { +async function queryProvider(provider, message, filePath = null, engine = null, conversationId = null, modelId = null, think = true) { const byokKey = byok.keys.isEnabled() ? byok.keys.getKey(provider) : null; if (byokKey) { const messages = [{ role: 'user', content: message }]; @@ -167,7 +167,7 @@ async function queryProvider(provider, message, filePath = null, engine = null, const sendResult = await handleMCPRequest({ action: 'sendMessage', provider, - data: { message, filePath, engine, conversationId } + data: { message, filePath, engine, conversationId, think } }); if (!sendResult.success) { @@ -199,7 +199,7 @@ function pickBestProvider(preferred) { if (found) return found; return null; } - const priorityList = ['claude', 'chatgpt', 'gemini', 'perplexity', 'deepseek', 'groq', 'xai', 'openrouter', 'together', 'fireworks', 'mistral', 'nvidia']; + const priorityList = ['claude', 'chatgpt', 'gemini', 'perplexity', 'kimi', 'deepseek', 'groq', 'xai', 'openrouter', 'together', 'fireworks', 'mistral', 'nvidia']; const priority = priorityList.find(p => enabled.includes(p)); if (priority) return priority; return enabled.length > 0 ? enabled[0] : null; @@ -265,9 +265,10 @@ async function handleWSMessage(ws, clientId, msg) { sendJSON(ws, { type: 'status', id: requestId, status: 'processing', model: provider, timestamp: new Date().toISOString() }); const conversationId = msg.conversationId || msg.conversation_id || msg.sessionId || msg.session_id || null; + const think = msg.think !== false; const startTime = Date.now(); try { - const content = await queryProvider(provider, message, filePath, engine, conversationId, byokModelId); + const content = await queryProvider(provider, message, filePath, engine, conversationId, byokModelId, think); const responseTimeMs = Date.now() - startTime; sendJSON(ws, { @@ -546,7 +547,7 @@ async function handleWSMessage(ws, clientId, msg) { case 'reset': { const requested = msg.model || null; if (!requested) { - sendJSON(ws, { type: 'error', id: requestId, error: 'A "model" (provider) is required to reset: chatgpt, claude, gemini, or perplexity.', timestamp: new Date().toISOString() }); + sendJSON(ws, { type: 'error', id: requestId, error: 'A "model" (provider) is required to reset: chatgpt, claude, gemini, perplexity, or kimi.', timestamp: new Date().toISOString() }); break; } const target = pickBestProvider(requested); diff --git a/electron/browser-manager.cjs b/electron/browser-manager.cjs index ec228c6..4ff59c9 100644 --- a/electron/browser-manager.cjs +++ b/electron/browser-manager.cjs @@ -32,6 +32,11 @@ class BrowserManager { url: 'https://gemini.google.com/app', partition: 'persist:gemini', color: '#4285f4' + }, + kimi: { + url: 'https://www.kimi.com/', + partition: 'persist:kimi', + color: '#6b4fbb' } }; @@ -398,7 +403,7 @@ class BrowserManager { }; view.webContents.on('will-navigate', (event, url) => { - if (isAuthUrl(url)) { + if (isAuthUrl(url) && provider !== 'kimi') { console.log(`[${provider}] Intercepting navigation to auth URL:`, url.substring(0, 80)); event.preventDefault(); this.openAuthPopup(provider, url); @@ -406,7 +411,7 @@ class BrowserManager { }); view.webContents.on('will-redirect', (event, url) => { - if (isAuthUrl(url)) { + if (isAuthUrl(url) && provider !== 'kimi') { console.log(`[${provider}] Intercepting redirect to auth URL:`, url.substring(0, 80)); event.preventDefault(); this.openAuthPopup(provider, url); @@ -434,6 +439,26 @@ class BrowserManager { view.webContents.setWindowOpenHandler(({ url, frameName, features }) => { console.log(`[${provider}] Popup requested:`, url.substring(0, 80)); + // Kimi's login is an in-page modal; its Google OAuth must stay in the + // same Chromium session (not be routed through the Firefox-UA auth + // window), otherwise the OAuth callback breaks. Allow all popups for + // Kimi with the normal Chromium UA/session. + if (provider === 'kimi') { + return { + action: 'allow', + overrideBrowserWindowOptions: { + width: 600, + height: 700, + webPreferences: { + session: ses, + sandbox: true, + contextIsolation: true, + nodeIntegration: false, + } + } + }; + } + const lowerUrl = url.toLowerCase(); const isAuthPopup = lowerUrl.includes('accounts.google.com') || @@ -608,7 +633,8 @@ class BrowserManager { perplexity: 'perplexity.ai', chatgpt: 'chatgpt.com', claude: 'claude.ai', - gemini: 'gemini.google.com' + gemini: 'gemini.google.com', + kimi: 'kimi.com' }; const domain = providerDomains[provider]; @@ -801,6 +827,38 @@ class BrowserManager { return hasInput && !hasSignIn; })() `); + case 'kimi': + // kimi.com renders a chat editor even when logged out, so DOM + // checks are unreliable. Kimi's web session is backed by a JWT + // in localStorage plus an HttpOnly cookie. Login = a valid + // access token is present in storage AND the subscription + // endpoint answers 2xx when called with that token. + return await webContents.executeJavaScript(` + (async function() { + function isJwt(v) { + return typeof v === 'string' && /^eyJ[A-Za-z0-9_-]+\\.[A-Za-z0-9_-]+\\.[A-Za-z0-9_-]+$/.test(v.trim()); + } + var token = null; + try { + var at = localStorage.getItem('access_token'); + if (isJwt(at)) token = at.trim(); + } catch (e) { } + try { + var deviceId = localStorage.getItem('proxima_kimi_device_id') || '7' + String(Math.floor(Math.random() * 9e17)).padStart(18, '0'); + var sessionId = localStorage.getItem('proxima_kimi_session_id') || '17' + String(Math.floor(Math.random() * 9e17)).padStart(17, '0'); + var headers = { 'Content-Type': 'application/json', 'Connect-Protocol-Version': '1', 'Origin': 'https://www.kimi.com', 'X-Msh-Platform': 'web', 'X-Msh-Device-Id': deviceId, 'X-Msh-Session-Id': sessionId }; + if (token) headers['Authorization'] = 'Bearer ' + token; + var res = await fetch('/apiv2/kimi.gateway.order.v1.SubscriptionService/GetSubscription', { + method: 'POST', credentials: 'include', headers: headers, body: '{}' + }); + if (res.status === 401 || res.status === 403) return false; + if (res.ok) return true; + return false; + } catch (e) { + return !!token; + } + })() + `); default: return false; } diff --git a/electron/index-v2.html b/electron/index-v2.html index afc2123..c18724a 100644 --- a/electron/index-v2.html +++ b/electron/index-v2.html @@ -154,6 +154,10 @@ background: linear-gradient(135deg, #4285f4, #8ab4f8); } + .tab-icon.kimi { + background: linear-gradient(135deg, #6b4fbb, #a78bfa); + } + .tab-icon.settings { background: transparent; border: none; @@ -1328,6 +1332,11 @@ Gemini
+
+ Kimi + Kimi +
+
Active Providers
-
0 / 4
+
0 / 5
@@ -1668,6 +1677,28 @@

Account: Session + + +
+
+
+ + Kimi +
+
+
+
+
Status: Checking...
+
Latency: --ms
+
Model: Auto
+
Account: Session
+
+
@@ -2456,7 +2487,7 @@

Platform Stats:
- 4 Providers • MCP Enabled • REST Server + 5 Providers • MCP Enabled • REST Server
@@ -2570,7 +2601,8 @@

Object.values(settings.providers || {}).forEach(p => { if (p && p.enabled) activeCount++; }); - document.getElementById('dash-active-count').textContent = `${activeCount} / 4`; + const totalProviders = Object.keys(settings.providers || {}).length; + document.getElementById('dash-active-count').textContent = `${activeCount} / ${totalProviders}`; } // Update Sidebar Footer Status Card top-row dynamically @@ -2582,7 +2614,8 @@

if (isByok) { sbText.textContent = `${activeCount} Active`; } else { - sbText.textContent = `${activeCount} / 4`; + const totalProviders = Object.keys(settings.providers || {}).length; + sbText.textContent = `${activeCount} / ${totalProviders}`; } if (activeCount > 0) { @@ -3311,13 +3344,15 @@

perplexity: 'perplexity.ai', chatgpt: 'chatgpt.com', claude: 'claude.ai', - gemini: 'gemini.google.com' + gemini: 'gemini.google.com', + kimi: 'kimi.com' }; const brandColors = { perplexity: '#20b2aa', chatgpt: '#10a37f', claude: '#cc785c', - gemini: '#4285f4' + gemini: '#4285f4', + kimi: '#6b4fbb' }; const linkElem = document.getElementById('cookie-modal-link'); if (linkElem) { @@ -3358,7 +3393,8 @@

perplexity: 'https://www.perplexity.ai/', chatgpt: 'https://chatgpt.com/', claude: 'https://claude.ai/', - gemini: 'https://gemini.google.com/' + gemini: 'https://gemini.google.com/', + kimi: 'https://www.kimi.com/' }; const targetUrl = urls[cookieModalProvider] || 'https://google.com/'; const name = cookieModalProvider.charAt(0).toUpperCase() + cookieModalProvider.slice(1); diff --git a/electron/ipc/core.cjs b/electron/ipc/core.cjs index 027347b..48141d2 100644 --- a/electron/ipc/core.cjs +++ b/electron/ipc/core.cjs @@ -152,7 +152,8 @@ ipcMain.handle('open-in-system-browser', (event, provider) => { perplexity: 'https://www.perplexity.ai/', chatgpt: 'https://chat.openai.com/', claude: 'https://claude.ai/', - gemini: 'https://gemini.google.com/' + gemini: 'https://gemini.google.com/', + kimi: 'https://www.kimi.com/' }; if (urls[provider]) { shell.openExternal(urls[provider]); diff --git a/electron/ipc/settings.cjs b/electron/ipc/settings.cjs index 1b3a5d5..106c3f0 100644 --- a/electron/ipc/settings.cjs +++ b/electron/ipc/settings.cjs @@ -141,7 +141,8 @@ function registerSettingsHandlers(deps) { perplexity: 'perplexity.ai', chatgpt: 'openai.com', claude: 'claude.ai', - gemini: 'google.com' + gemini: 'google.com', + kimi: 'kimi.com' }; const domain = providerDomains[provider]; diff --git a/electron/main-v2.cjs b/electron/main-v2.cjs index 047f07c..337a076 100644 --- a/electron/main-v2.cjs +++ b/electron/main-v2.cjs @@ -49,6 +49,7 @@ const PROVIDER_PARTITIONS = [ 'persist:chatgpt', 'persist:claude', 'persist:gemini', + 'persist:kimi', ]; async function clearProviderCachesOnStartup() { @@ -99,7 +100,8 @@ const defaultSettings = { perplexity: { enabled: true, loggedIn: false }, chatgpt: { enabled: true, loggedIn: false }, claude: { enabled: true, loggedIn: false }, - gemini: { enabled: true, loggedIn: false } + gemini: { enabled: true, loggedIn: false }, + kimi: { enabled: true, loggedIn: false } }, ipcPort: 19222, theme: 'dark', @@ -521,11 +523,28 @@ async function handleMCPRequest(request) { return { success: true, provider, loggedIn }; } + case 'getKimiSessionStatus': { + const conversationId = data && (data.conversationId || data.conversation_id || data.sessionId || data.session_id); + const wc = browserManager.getWebContents('kimi'); + if (!wc) return { success: false, error: 'Kimi not initialized' }; + try { + const status = await wc.executeJavaScript( + `(window.__proximaKimi && window.__proximaKimi.getSessionStatus) + ? window.__proximaKimi.getSessionStatus(${JSON.stringify(conversationId || null)}) + : { resetCount: 0, chatId: null, parentId: null }` + ); + return { success: true, ...status }; + } catch (err) { + return { success: false, error: err.message }; + } + } + case 'sendMessage': { const baseProviderName = provider.split(':')[0]; const engineOverride = data.gemini || data.engine || null; const targetProvider = engineOverride ? `${provider}:${engineOverride}` : provider; const conversationId = data.conversationId || data.conversation_id || data.sessionId || data.session_id || null; + const thinkMode = data.think !== false; const byokKey = byok.keys.isEnabled() ? byok.keys.getKey(baseProviderName) : null; if (byokKey) { @@ -565,7 +584,8 @@ async function handleMCPRequest(request) { fileSize: uploadResult.fileSize } : null, data.onChunk, - conversationId + conversationId, + thinkMode ); if (!result.response && result.error) { return { success: false, provider, error: result.error, fileUploaded: uploadResult }; @@ -576,14 +596,14 @@ async function handleMCPRequest(request) { return { success: true, provider, response: result.response, fileUploaded: uploadResult }; } catch (fileErr) { console.error('[MCP] File upload failed:', fileErr.message); - const result = await sendMessageToProvider(targetProvider, data.message, null, data.onChunk, conversationId); + const result = await sendMessageToProvider(targetProvider, data.message, null, data.onChunk, conversationId, thinkMode); if (!result.response || result.response.length === 0) { return { success: false, provider, error: result.error || `${provider} returned empty response`, fileError: fileErr.message }; } return { success: true, provider, response: result.response, fileError: fileErr.message }; } } else { - const result = await sendMessageToProvider(targetProvider, data.message, null, data.onChunk, conversationId); + const result = await sendMessageToProvider(targetProvider, data.message, null, data.onChunk, conversationId, thinkMode); if (!result.response && result.error) { return { success: false, provider, error: result.error }; } @@ -617,6 +637,7 @@ async function handleMCPRequest(request) { const targetProviderWithFile = data.engine ? `${provider}:${data.engine}` : provider; const conversationIdWithFile = data.conversationId || data.conversation_id || data.sessionId || data.session_id || null; + const thinkModeWithFile = data.think !== false; const msgResult = await sendMessageToProvider( targetProviderWithFile, data.message, @@ -627,7 +648,8 @@ async function handleMCPRequest(request) { fileSize: fileResult.fileSize } : null, null, - conversationIdWithFile + conversationIdWithFile, + thinkModeWithFile ); const finalResponse = (msgResult && msgResult.response) || ''; return { @@ -924,7 +946,29 @@ async function uploadFileToProvider(provider, filePath) { } } - throw new Error(`File upload is not supported for ${provider} (API mode). Only Gemini, ChatGPT, Perplexity, and Claude support file attachments.`); + if (provider.startsWith('kimi')) { + console.log(`[Kimi API] Direct upload of ${fileName} via Web API...`); + try { + await providerAPI.ensureAPI(provider, webContents); + const token = await webContents.executeJavaScript( + `window.__proximaKimi.uploadFileToKimi(${JSON.stringify(fileBase64)}, ${JSON.stringify(fileName)}, ${JSON.stringify(fileMimeType)})` + ); + return { + success: true, + fileName, + mimeType: fileMimeType, + imageToken: token, + fileAttached: true, + fileSize: fileStats.size, + method: 'kimi-api' + }; + } catch (apiUploadErr) { + console.error(`[Kimi API] Direct upload failed:`, apiUploadErr.message); + throw apiUploadErr; + } + } + + throw new Error(`File upload is not supported for ${provider} (API mode). Only Gemini, ChatGPT, Perplexity, Claude, and Kimi support file attachments.`); } function getMimeType(filePath) { @@ -995,7 +1039,7 @@ app.on('activate', () => { }); app.on('certificate-error', (event, webContents, url, error, certificate, callback) => { - const trustedDomains = ['perplexity.ai', 'openai.com', 'chatgpt.com', 'claude.ai', 'anthropic.com', 'gemini.google.com', 'accounts.google.com']; + const trustedDomains = ['perplexity.ai', 'openai.com', 'chatgpt.com', 'claude.ai', 'anthropic.com', 'gemini.google.com', 'accounts.google.com', 'kimi.com', 'moonshot.cn']; let host; try { host = new URL(url).hostname.toLowerCase(); diff --git a/electron/providers/api.cjs b/electron/providers/api.cjs index 1309db9..e7a6541 100644 --- a/electron/providers/api.cjs +++ b/electron/providers/api.cjs @@ -15,7 +15,8 @@ function _loadScript(provider) { chatgpt: 'chatgpt-engine.js', claude: 'claude-engine.js', gemini: 'gemini-engine.js', - perplexity: 'perplexity-engine.js' + perplexity: 'perplexity-engine.js', + kimi: 'kimi-engine.js' }; const filename = scriptMap[provider]; @@ -66,7 +67,8 @@ async function isAPIReady(provider, webContents) { chatgpt: 'typeof window.__proximaChatGPT !== "undefined"', claude: 'typeof window.__proximaClaude !== "undefined"', gemini: 'typeof window.__proximaGemini !== "undefined"', - perplexity: 'typeof window.__proximaPerplexity !== "undefined"' + perplexity: 'typeof window.__proximaPerplexity !== "undefined"', + kimi: 'typeof window.__proximaKimi !== "undefined"' }; const check = checkMap[provider]; @@ -122,7 +124,7 @@ async function ensureAPI(provider, webContents) { return await injectAPI(provider, webContents); } -async function sendViaAPI(provider, webContents, message, attachments = null, onChunk = null, conversationId = null) { +async function sendViaAPI(provider, webContents, message, attachments = null, onChunk = null, conversationId = null, think = true) { let baseProvider = provider; let engine = 'auto'; if (provider.indexOf(':') !== -1) { @@ -141,7 +143,8 @@ async function sendViaAPI(provider, webContents, message, attachments = null, on chatgpt: '__proximaChatGPT', claude: '__proximaClaude', gemini: '__proximaGemini', - perplexity: '__proximaPerplexity' + perplexity: '__proximaPerplexity', + kimi: '__proximaKimi' }; const apiObj = sendMap[baseProvider]; @@ -154,7 +157,10 @@ async function sendViaAPI(provider, webContents, message, attachments = null, on const startTime = Date.now(); let executeStr; - if (baseProvider === 'gemini' || baseProvider === 'chatgpt' || baseProvider === 'perplexity' || baseProvider === 'claude') { + if (baseProvider === 'chatgpt') { + const escapedAttachments = attachments ? JSON.stringify(attachments) : 'null'; + executeStr = `window.${apiObj}.send(${escapedMessage}, ${JSON.stringify(engine)}, ${escapedAttachments}, ${JSON.stringify(conversationId)}, ${JSON.stringify({ think: think !== false })})`; + } else if (baseProvider === 'gemini' || baseProvider === 'perplexity' || baseProvider === 'claude' || baseProvider === 'kimi') { const escapedAttachments = attachments ? JSON.stringify(attachments) : 'null'; executeStr = `window.${apiObj}.send(${escapedMessage}, ${JSON.stringify(engine)}, ${escapedAttachments}, ${JSON.stringify(conversationId)})`; } else { @@ -236,10 +242,11 @@ async function resetConversation(provider, webContentsGetter) { chatgpt: '__proximaChatGPT', claude: '__proximaClaude', gemini: '__proximaGemini', - perplexity: '__proximaPerplexity' + perplexity: '__proximaPerplexity', + kimi: '__proximaKimi' }; - const providers = provider ? [provider] : ['chatgpt', 'claude', 'gemini', 'perplexity']; + const providers = provider ? [provider] : ['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi']; for (const p of providers) { const apiObj = resetMap[p]; diff --git a/electron/providers/engines/chatgpt-engine.js b/electron/providers/engines/chatgpt-engine.js index c593fd7..8596173 100644 --- a/electron/providers/engines/chatgpt-engine.js +++ b/electron/providers/engines/chatgpt-engine.js @@ -197,6 +197,139 @@ var _cachedScripts = null; var _cachedDpl = null; + var _cachedModels = null; + var _cachedModelsAt = 0; + + // Mirrors the web app's Think-toggle logic: thinking_effort is only sent for models + // that report configurableThinkingEffort, and only with a value present in their + // thinkingEfforts list. Otherwise the field is omitted entirely. + async function _getModelConfig(modelSlug) { + try { + if (!_cachedModels || Date.now() - _cachedModelsAt > 300000) { + var mres = await fetch('/backend-api/models', { credentials: 'include' }); + if (mres.ok) { + _cachedModels = (await mres.json()).models || []; + _cachedModelsAt = Date.now(); + } + } + if (!_cachedModels) return null; + for (var i = 0; i < _cachedModels.length; i++) { + if (_cachedModels[i].slug === modelSlug) return _cachedModels[i]; + } + return null; + } catch (e) { + return null; + } + } + + function _resolveThinkingEffort(modelConfig, think) { + if (!think || !modelConfig || !modelConfig.configurableThinkingEffort) return undefined; + var efforts = (modelConfig.thinkingEfforts || []).map(function (e) { return e.thinking_effort; }).filter(Boolean); + if (!efforts.length) return undefined; + if (efforts.indexOf('xhigh') !== -1) return 'xhigh'; + if (efforts.indexOf('standard') !== -1) return 'standard'; + return efforts[efforts.length - 1]; + } + + // Drives the actual composer "Think"/"Reason" toggle button in the page (the same button + // a user clicks on chatgpt.com), so the web app's own UI logic produces the backend payload. + // We only ever turn it ON; if it is already ON we never click (so it can never toggle OFF). + function _findThinkToggle() { + var selectors = [ + '[data-testid="composer-intelligence-button"]', + '[data-testid="composer-intelligence-pro-thinking-effort-trigger"]', + 'button[aria-label*="Think" i]', + 'button[aria-label*="Reason" i]', + 'button[data-testid*="think" i]', + 'button[data-testid*="reason" i]' + ]; + for (var i = 0; i < selectors.length; i++) { + var el = document.querySelector(selectors[i]); + if (el) return el; + } + var pills = document.querySelectorAll('button.__composer-pill'); + for (var k = 0; k < pills.length; k++) { + var ptext = (pills[k].innerText || '').trim(); + if (/^(Think|Reason|Deep Think)$/i.test(ptext)) return pills[k]; + } + var buttons = document.querySelectorAll('button'); + for (var j = 0; j < buttons.length; j++) { + var txt = (buttons[j].innerText || '').trim(); + if (/^(Think|Reason|Deep Think)$/i.test(txt)) return buttons[j]; + } + return null; + } + + function _isToggleOn(btn) { + if (!btn) return false; + var ap = btn.getAttribute('aria-pressed'); + if (ap === 'true') return true; + if (ap === 'false') return false; + var ds = btn.getAttribute('data-state'); + if (ds === 'on') return true; + if (ds === 'off') return false; + var ariaChecked = btn.getAttribute('aria-checked'); + if (ariaChecked === 'true') return true; + if (ariaChecked === 'false') return false; + var cls = (btn.className || '').toLowerCase(); + if (cls.indexOf('active') !== -1 || cls.indexOf('selected') !== -1 || cls.indexOf('on') !== -1) return true; + return false; + } + + function _clickButton(btn) { + var rect = btn.getBoundingClientRect(); + var opts = { + bubbles: true, + cancelable: true, + view: window, + clientX: rect.left + (rect.width / 2), + clientY: rect.top + (rect.height / 2) + }; + try { btn.dispatchEvent(new MouseEvent('pointerdown', Object.assign({}, opts, { pointerId: 1, isPrimary: true, button: 0 }))); } catch (e) { } + try { btn.dispatchEvent(new MouseEvent('mousedown', opts)); } catch (e) { } + try { btn.dispatchEvent(new MouseEvent('mouseup', opts)); } catch (e) { } + try { btn.dispatchEvent(new MouseEvent('click', opts)); } catch (e) { } + try { btn.click(); } catch (e) { } + } + + function _ensureThinkToggleOn() { + return new Promise(function (resolve) { + try { + var btn = _findThinkToggle(); + if (!btn) { + console.log('[Proxima ChatGPT] Think toggle button not found in DOM'); + resolve({ found: false, on: false }); + return; + } + if (_isToggleOn(btn)) { + console.log('[Proxima ChatGPT] Think toggle already ON — leaving it alone (never toggling OFF)'); + resolve({ found: true, on: true }); + return; + } + _clickButton(btn); + console.log('[Proxima ChatGPT] Think toggle click dispatched (was off)'); + var attempts = 0; + var iv = setInterval(function () { + attempts++; + var current = _findThinkToggle(); + if (current && _isToggleOn(current)) { + clearInterval(iv); + console.log('[Proxima ChatGPT] Think toggle confirmed ON'); + resolve({ found: true, on: true }); + return; + } + if (attempts >= 12) { + clearInterval(iv); + console.log('[Proxima ChatGPT] Think toggle did NOT turn ON after click (still off)'); + resolve({ found: true, on: false }); + } + }, 200); + } catch (e) { + console.error('[Proxima ChatGPT] _ensureThinkToggleOn error:', e.message); + resolve({ found: false, on: false, error: e.message }); + } + }); + } async function _getScriptsAndDpl() { if (_cachedScripts) return { scripts: _cachedScripts, dpl: _cachedDpl }; @@ -371,9 +504,11 @@ return fileId; } - async function send(message, engine, attachments, sessionId) { + async function send(message, engine, attachments, sessionId, options) { activateSession(sessionId); + var think = !(options && options.think === false); + var token = await _getToken(); var deviceId = ''; @@ -461,6 +596,31 @@ websocket_request_id: crypto.randomUUID() }; + var resolvedModel = (engine && engine !== 'auto') ? engine : 'auto'; + payload.model = resolvedModel; + + // "Think" mode = the composer's Think toggle (chatgpt.com). We drive the real UI + // button (never toggling it OFF) so the web app's own state reflects think mode, + // then mirror its backend behavior in our payload: thinking_effort is only sent + // when the selected model has configurableThinkingEffort and the value is in its + // thinkingEfforts list; otherwise the field is omitted (free accounts report + // configurableThinkingEffort:false, so nothing is sent — real think mode there + // means selecting a reasoning-variant model such as 'gpt-5-6-t-mini'). + if (think) { + var toggle = await _ensureThinkToggleOn(); + console.log('[Proxima ChatGPT] Think toggle ensure: found=' + toggle.found + ' on=' + toggle.on + (toggle.error ? ' error=' + toggle.error : '')); + } + var modelConfig = await _getModelConfig(resolvedModel); + var effort = _resolveThinkingEffort(modelConfig, think); + if (effort) { + payload.thinking_effort = effort; + console.log('[Proxima ChatGPT] Thinking mode ON (thinking_effort=' + effort + ')'); + } else if (think && modelConfig && !modelConfig.configurableThinkingEffort) { + console.log('[Proxima ChatGPT] Thinking mode ON but not configurable for model ' + resolvedModel + ' — field omitted'); + } else { + console.log('[Proxima ChatGPT] Thinking mode OFF'); + } + if (_conversationId) { payload.conversation_id = _conversationId; console.log('[Proxima ChatGPT] Continuing conversation:', _conversationId); diff --git a/electron/providers/engines/kimi-engine.js b/electron/providers/engines/kimi-engine.js new file mode 100644 index 0000000..acaa9f5 --- /dev/null +++ b/electron/providers/engines/kimi-engine.js @@ -0,0 +1,643 @@ +/** + * Proxima — Kimi Engine + * Runs inside kimi.com BrowserView context. Sends queries via the internal + * Connect (CONNECT/gRPC-Web) protocol used by the Kimi web app: + * + * POST /apiv2/kimi.gateway.chat.v1.ChatService/Chat + * + * The request body is a Connect frame (1 flag byte + 4-byte big-endian length + * + JSON payload). The response is a stream of the same 5-byte-framed JSON + * events. Answer deltas live in event.block.text.content; reasoning lives in + * event.block.think.content. + * + * ⚠️ Uses the same non-official protocol the kimi.com web app itself uses. + * If Kimi changes their protocol this engine may need updating. + */ +(function () { + if (window.__proximaKimi) return; + + var TIMEOUT = 360000; + var CHAT_PATH = '/apiv2/kimi.gateway.chat.v1.ChatService/Chat'; + var REFRESH_PATH = '/api/auth/token/refresh'; + var DEFAULT_SCENARIO = 'SCENARIO_K2D5'; + + var _accessToken = null; + var _refreshToken = null; + var _currentSessionId = null; + var _sessions = {}; + try { + var saved = localStorage.getItem('proxima_kimi_sessions'); + if (saved) _sessions = JSON.parse(saved); + } catch (e) { } + + var MAX_SESSIONS = 200; + function _pruneSessions() { + var keys = Object.keys(_sessions); + for (var i = 0; i < keys.length && Object.keys(_sessions).length > MAX_SESSIONS; i++) { + if (keys[i] !== _currentSessionId) delete _sessions[keys[i]]; + } + } + + // Session change signal for the opencode plugin. The engine bumps this + // counter every time it rolls a Kimi chat over to a NEW chat_id (context + // limit recovery). The plugin polls Proxima's session-status endpoint and, + // on change, re-sends the system prompt + a comprehensive handoff summary. + function getSessionStatus(sessionId) { + var s = _sessions[sessionId || 'default']; + if (!s) return { resetCount: 0, chatId: null, parentId: null }; + return { + resetCount: (s.resetCount || 0), + chatId: s.chatId || null, + parentId: s.parentId || null + }; + } + + function _deviceId() { + try { + var d = localStorage.getItem('proxima_kimi_device_id'); + if (d && /^[0-9]{19}$/.test(d)) return d; + var nd = String(Math.floor(7000000000000000000 + Math.random() * 999999999999999999)); + localStorage.setItem('proxima_kimi_device_id', nd); + return nd; + } catch (e) { + return '7' + String(Math.floor(Math.random() * 9e17)).padStart(18, '0'); + } + } + + function _sessionHeader() { + try { + var s = localStorage.getItem('proxima_kimi_session_id'); + if (s && /^[0-9]{19}$/.test(s)) return s; + var ns = String(Math.floor(1700000000000000000 + Math.random() * 99999999999999999)); + localStorage.setItem('proxima_kimi_session_id', ns); + return ns; + } catch (e) { + return '17' + String(Math.floor(Math.random() * 9e17)).padStart(17, '0'); + } + } + + function _isJwt(v) { + return typeof v === 'string' && /^eyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+$/.test(v.trim()); + } + + // Parse the exp claim from a JWT (no signature verification needed for + // expiry math). Returns seconds-epoch, or 0 if unparsable. + function _jwtExp(jwt) { + try { + var parts = jwt.split('.'); + if (parts.length !== 3) return 0; + var b64 = parts[1]; + b64 += '='.repeat((4 - (b64.length % 4)) % 4); + var payload = JSON.parse(decodeURIComponent(escape(atob(b64)))); + return typeof payload.exp === 'number' ? payload.exp : 0; + } catch (e) { + return 0; + } + } + + // True if the token is missing or expired within 120s (proactive refresh + // window so Kimi never sees a stale token). + function _tokenNeedsRefresh(jwt) { + if (!_isJwt(jwt)) return true; + var exp = _jwtExp(jwt); + if (!exp) return false; // can't verify — assume it's fine + return (Date.now() / 1000) + 120 >= exp; + } + + function _scanStorageToken() { + try { + var storages = [localStorage, sessionStorage]; + for (var si = 0; si < storages.length; si++) { + var store = storages[si]; + var access = null; + var refresh = null; + for (var i = 0; i < store.length; i++) { + var key = store.key(i); + var val = ''; + try { val = store.getItem(key) || ''; } catch (e) { continue; } + var candidate = val.trim(); + if (!_isJwt(candidate)) { + try { + var parsed = JSON.parse(val); + if (parsed && typeof parsed === 'object') { + var deep = _deepFindToken(parsed); + if (deep) candidate = deep; + } + } catch (e2) { } + } + if (!_isJwt(candidate)) continue; + var lower = key.toLowerCase(); + if (lower.includes('refresh')) { + if (!refresh) refresh = candidate; + } else if (lower.includes('access') || lower.includes('token') || lower.includes('auth')) { + if (!access) access = candidate; + } else if (!access) { + access = candidate; + } + } + if (access) { _refreshToken = refresh || _refreshToken; return access; } + if (refresh) { _refreshToken = refresh; } + } + } catch (e) { } + + // Fallback: check cookies for token-ish values. + try { + var cookies = document.cookie.split(';'); + for (var ci = 0; ci < cookies.length; ci++) { + var cookie = cookies[ci].trim(); + var eq = cookie.indexOf('='); + if (eq === -1) continue; + var cName = cookie.substring(0, eq).trim(); + var cVal = decodeURIComponent(cookie.substring(eq + 1).trim()); + if (_isJwt(cVal)) { + var lName = cName.toLowerCase(); + if (lName.includes('refresh')) { if (!_refreshToken) _refreshToken = cVal; } + else if (lName.includes('access') || lName.includes('token')) { return cVal; } + else if (!_accessToken) { return cVal; } + } + } + } catch (e) { } + + return null; + } + + function _deepFindToken(obj, depth) { + if (!depth) depth = 0; + if (depth > 6 || !obj || typeof obj !== 'object') return null; + if (typeof obj === 'string' && _isJwt(obj)) return obj; + for (var key of Object.keys(obj)) { + var val = obj[key]; + if (typeof val === 'string' && _isJwt(val)) return val; + if (val && typeof val === 'object') { + var found = _deepFindToken(val, depth + 1); + if (found) return found; + } + } + return null; + } + + async function _refreshAccessToken() { + // Always rediscover the refresh token from storage first — Kimi rotates + // refresh tokens too, so a stale in-memory one will be rejected. + _scanStorageToken(); + if (!_refreshToken) return null; + try { + var res = await fetch(REFRESH_PATH, { + method: 'GET', + credentials: 'include', + headers: { + 'Authorization': 'Bearer ' + _refreshToken, + 'Accept': 'application/json', + 'X-Msh-Platform': 'web', + 'X-Msh-Device-Id': _deviceId(), + 'X-Msh-Session-Id': _sessionHeader(), + 'R-Timezone': Intl.DateTimeFormat().resolvedOptions().timeZone || 'UTC', + 'Origin': 'https://www.kimi.com' + } + }); + if (res.ok) { + var data = await res.json(); + var token = data.access_token || data.token; + if (token && _isJwt(token)) { + _accessToken = token; + // Persist the fresh token so later engine loads reuse it + // instead of re-reading an expired one from storage. + try { + localStorage.setItem('access_token', token); + } catch (e) { } + return token; + } + } + } catch (e) { } + return null; + } + + // Kimi revokes access tokens server-side (rotation) while the JWT exp claim + // stays ~30 days out, so expiry math alone can't detect a stale token. Always + // prefer the token the site currently has in storage over our cached copy. + function _getAccessToken(force) { + var storageToken = _scanStorageToken(); + if (storageToken) { + _accessToken = storageToken; + return Promise.resolve(storageToken); + } + if (_accessToken && !force) { + return Promise.resolve(_accessToken); + } + return _refreshAccessToken(); + } + + function _encodeConnect(payload) { + var body = new TextEncoder().encode(JSON.stringify(payload)); + var header = new Uint8Array(5); + header[0] = 0x00; + var len = body.length; + header[1] = (len >>> 24) & 0xff; + header[2] = (len >>> 16) & 0xff; + header[3] = (len >>> 8) & 0xff; + header[4] = len & 0xff; + var out = new Uint8Array(5 + body.length); + out.set(header, 0); + out.set(body, 5); + return out; + } + + function _isAuthErrorMsg(msg) { + if (!msg) return false; + var lower = String(msg).toLowerCase(); + return lower.indexOf('invalid user token') !== -1 || + lower.indexOf('invalid token') !== -1 || + lower.indexOf('unauthorized') !== -1 || + lower.indexOf('unauthenticated') !== -1 || + lower.indexOf('session expired') !== -1 || + lower.indexOf('not logged in') !== -1 || + lower.indexOf('authentication') !== -1 || + lower.indexOf('expired') !== -1; + } + + // Kimi's server-enforced conversation-context limit. No bypass exists + // (web K2.6 ≈128K tokens; K3 1M only on top-tier membership). The only + // recovery is a NEW chat_id. We detect it, then restart with the stored + // system prompt + the current user message (research-backed). + function _isTooLongMsg(msg) { + if (!msg) return false; + var lower = String(msg).toLowerCase(); + return lower.indexOf('too long') !== -1 || + lower.indexOf('start a new session') !== -1 || + lower.indexOf('start a new chat') !== -1 || + lower.indexOf('conversation') !== -1 && lower.indexOf('long') !== -1 || + lower.indexOf('context') !== -1 && lower.indexOf('limit') !== -1; + } + + function _AuthError(msg) { + var e = new Error(msg); + e.__kimiAuth = true; + return e; + } + + function _TooLongError(msg) { + var e = new Error(msg); + e.__kimiTooLong = true; + return e; + } + + // The engine may receive either a plain string or a JSON-encoded messages + // array (opencode / REST clients send [{role,content}, ...]). Normalize to + // an array and flatten content parts to plain text. + function _parseIncomingMessage(message) { + if (Array.isArray(message)) return message; + if (typeof message === 'string') { + var trimmed = message.trim(); + if (trimmed.charAt(0) === '[') { + try { + var parsed = JSON.parse(trimmed); + if (Array.isArray(parsed)) return parsed; + } catch (e) { } + } + return [{ role: 'user', content: message }]; + } + return [{ role: 'user', content: String(message || '') }]; + } + + function _contentText(content) { + if (content === null || content === undefined) return ''; + if (typeof content === 'string') return content; + if (Array.isArray(content)) { + var parts = []; + for (var i = 0; i < content.length; i++) { + var item = content[i]; + if (typeof item === 'string') parts.push(item); + else if (item && typeof item === 'object' && typeof item.text === 'string') parts.push(item.text); + } + return parts.join('\n'); + } + return String(content); + } + + async function _parseStream(response, session) { + var reader = response.body.getReader(); + var buffer = new Uint8Array(0); + var decoder = new TextDecoder(); + var answer = ''; + var reasoning = ''; + var done = false; + + try { + while (!done) { + var chunk = await reader.read(); + if (chunk.done) break; + + var merged = new Uint8Array(buffer.length + chunk.value.length); + merged.set(buffer, 0); + merged.set(chunk.value, buffer.length); + buffer = merged; + + var offset = 0; + while (offset + 5 <= buffer.length) { + var flag = buffer[offset]; + var length = ((buffer[offset + 1] << 24) | (buffer[offset + 2] << 16) | + (buffer[offset + 3] << 8) | buffer[offset + 4]) >>> 0; + var frameEnd = offset + 5 + length; + if (frameEnd > buffer.length) break; + + var payload = buffer.slice(offset + 5, frameEnd); + offset = frameEnd; + + // Trailers (flag & 0x80) carry headers only — skip. + if (flag & 0x80) continue; + + var text = decoder.decode(payload).trim(); + if (!text) continue; + + var event; + try { event = JSON.parse(text); } catch (e) { continue; } + + if (event.error) { + var err = event.error; + var msg = err.message || JSON.stringify(err); + // Conversation hit the server-side context limit — recover + // by starting a NEW chat (no bypass exists). + if (_isTooLongMsg(msg)) throw _TooLongError(msg); + // Kimi reports expired/invalid sessions as a stream error + // inside an HTTP 200 body — flag it so send() can refresh + // the token and retry instead of surfacing a raw failure. + if (_isAuthErrorMsg(msg)) throw _AuthError(msg); + throw new Error(msg); + } + if (event.chat && event.chat.id) { + session.chatId = event.chat.id; + } + if (event.message && event.message.role === 'assistant' && event.message.id) { + session.parentId = event.message.id; + } + if (event.block) { + if (event.block.think && event.block.think.content) { + reasoning += event.block.think.content; + } + if (event.block.text && event.block.text.content) { + answer += event.block.text.content; + } + } + if ('done' in event) { + done = true; + break; + } + } + + if (offset > 0) { + buffer = buffer.slice(offset); + } + } + } finally { + try { reader.releaseLock(); } catch (e) { } + } + + return { answer: answer.trim(), reasoning: reasoning.trim() }; + } + + function activateSession(sessionId) { + if (!sessionId) sessionId = 'default'; + if (sessionId !== _currentSessionId) { + _currentSessionId = sessionId; + if (!_sessions[sessionId]) { + _sessions[sessionId] = { chatId: null, parentId: null, systemPrompt: null, lastUser: null, resetCount: 0 }; + } + } + return _sessions[_currentSessionId]; + } + + function saveSession() { + if (!_currentSessionId) return; + try { + _pruneSessions(); + localStorage.setItem('proxima_kimi_sessions', JSON.stringify(_sessions)); + } catch (e) { } + } + + async function send(message, engine, attachments, sessionId) { + var session = activateSession(sessionId); + + // Normalize the incoming payload. opencode / REST clients pass either a + // plain string or a JSON-encoded messages array. Extract the system + // prompt (sent ONCE at chat start) and the latest user turn (the only + // thing follow-ups need — the parent_id chain carries the rest). + var msgs = _parseIncomingMessage(message); + var systemParts = []; + var userParts = []; + for (var i = 0; i < msgs.length; i++) { + var m = msgs[i]; + if (!m || typeof m !== 'object') continue; + if (m.role === 'system') systemParts.push(_contentText(m.content)); + else if (m.role === 'user') userParts.push(_contentText(m.content)); + } + var systemPrompt = systemParts.filter(Boolean).join('\n'); + + // Decide whether this request starts a NEW Kimi chat or continues the + // existing one. The incoming system prompt is the discriminator: + // * FIRST turn — no stored systemPrompt yet → new chat + // * ROLLOVER handoff — system prompt changed (summary appended by the + // plugin) → new chat (the old chat hit Kimi's context limit) + // * Follow-up — system prompt identical to stored (opencode can + // be restarted between turns; the plugin state that strips the prompt + // lives per-process, so the same prompt may legitimately reappear) → + // REUSE the existing chat_id so the conversation actually continues. + // A stale chat_id that is already too long self-heals: Kimi answers with + // the too-long error, the engine detects it below and rolls the session. + var isSameSystemPrompt = systemPrompt && session.systemPrompt === systemPrompt; + if (systemPrompt && !isSameSystemPrompt) { + if (session.chatId || session.parentId) { + console.log('[Proxima Kimi] System prompt changed on this request — starting a fresh chat (was chatId=' + (session.chatId || 'none') + ').'); + } + session.chatId = null; + session.parentId = null; + } + // Only the LATEST user turn is the new message — everything earlier is + // already on Kimi's server via chat_id/parent_id (follow-up) or will be + // included as first-turn context (new chat). + var lastUser = userParts.filter(Boolean).pop() || (typeof message === 'string' ? message : ''); + + // Remember the system prompt for this conversation so a context-limit + // reset can re-establish it in the fresh chat. + if (systemPrompt) session.systemPrompt = systemPrompt; + session.lastUser = lastUser; + + // First turn of a Kimi chat → send system prompt (which the opencode + // plugin augments with a comprehensive handoff summary after a rollover) + // + the current user turn. Do NOT re-send raw history here: after a + // context-limit rollover opencode re-sends the whole array and re-injecting + // it would instantly slam the new chat's limit again. Real context is + // carried by the summary inside the system prompt, not by replaying + // past turns. + var isNewChat = !session.chatId; + var content; + if (isNewChat) { + content = (systemPrompt ? 'system:' + systemPrompt + '\n' : '') + 'user:' + lastUser; + } else { + content = 'user:' + lastUser; + } + + var token = await _getAccessToken(false); + + var scenario = DEFAULT_SCENARIO; + if (engine && engine !== 'auto') { + var e = String(engine).toUpperCase(); + if (e.indexOf('SCENARIO_') === 0) scenario = e; + } + + var headers = { + 'Accept': '*/*', + 'Content-Type': 'application/connect+json', + 'Connect-Protocol-Version': '1', + 'Origin': 'https://www.kimi.com', + 'X-Msh-Platform': 'web', + 'X-Msh-Device-Id': _deviceId(), + 'X-Msh-Session-Id': _sessionHeader(), + 'R-Timezone': Intl.DateTimeFormat().resolvedOptions().timeZone || 'UTC', + 'Sec-Fetch-Dest': 'empty', + 'Sec-Fetch-Mode': 'cors', + 'Sec-Fetch-Site': 'same-origin' + }; + if (token) headers['Authorization'] = 'Bearer ' + token; + + var controller = new AbortController(); + var timeoutId = setTimeout(function () { controller.abort(); }, TIMEOUT); + + // Build the Connect payload for the current session state. + var buildPayload = function () { + var payload = { + scenario: scenario, + tools: [], + message: { + role: 'user', + blocks: [{ + message_id: crypto.randomUUID ? crypto.randomUUID() : '', + text: { content: content } + }], + scenario: scenario + }, + options: { thinking: false } + }; + if (session.chatId) payload.chat_id = session.chatId; + if (session.parentId) payload.message.parent_id = session.parentId; + return payload; + }; + + try { + var res = await fetch(CHAT_PATH, { + method: 'POST', + credentials: 'include', + headers: headers, + body: _encodeConnect(buildPayload()), + signal: controller.signal + }); + + var needAuthRetry = (res.status === 401 || res.status === 403); + + if (res.status === 429) throw new Error('Kimi rate limited'); + if (!res.ok && !needAuthRetry) { + var errBody = await res.text().catch(function () { return ''; }); + throw new Error('Kimi API error (' + res.status + '): ' + errBody.substring(0, 300)); + } + + var result; + var tooLong = false; + try { + result = needAuthRetry ? null : await _parseStream(res, session); + } catch (parseErr) { + if (parseErr && parseErr.__kimiAuth && !needAuthRetry) { + needAuthRetry = true; + } else if (parseErr && parseErr.__kimiTooLong) { + tooLong = true; + } else { + throw parseErr; + } + } + + if (needAuthRetry) { + // Kimi revokes access tokens server-side even when the JWT exp is + // far in the future, so the stored token can be stale/rejected. + // Force a refresh-token EXCHANGE (not a storage re-read — that + // would return the same revoked token). + if (token) _accessToken = null; + var fresh = await _refreshAccessToken(); + if (!fresh) { + fresh = await _getAccessToken(true); + } + if (fresh && fresh !== token) { + headers['Authorization'] = 'Bearer ' + fresh; + res = await fetch(CHAT_PATH, { + method: 'POST', + credentials: 'include', + headers: headers, + body: _encodeConnect(buildPayload()), + signal: controller.signal + }); + if (res.status === 429) throw new Error('Kimi rate limited'); + if (!res.ok) { + var errBody2 = await res.text().catch(function () { return ''; }); + throw new Error('Kimi API error (' + res.status + '): ' + errBody2.substring(0, 300)); + } + result = await _parseStream(res, session); + } else { + throw _AuthError('Not logged in to Kimi (no valid token found)'); + } + } + + if (tooLong) { + // No bypass exists for Kimi's context limit — the research-backed + // recovery is a NEW chat_id. The opencode plugin owns the context + // restoration: it polls Proxima's session status (resetCount), + // detects the rollover, and re-sends the system prompt + a + // comprehensive handoff summary on the next message. So we do NOT + // auto-recover here with the bare current question (the user + // explicitly wants a summary of recent work, not the question only). + console.log('[Proxima Kimi] Conversation too long — rolling session (resetCount++). Plugin will restore context.'); + session.chatId = null; + session.parentId = null; + session.resetCount = (session.resetCount || 0) + 1; + saveSession(); + result = { + answer: '**Conversation too long** — Kimi\'s context limit was reached, so I started a new session for you. Your next message will continue with a summary of our recent work (system prompt + context restored automatically).', + reasoning: '' + }; + } + + saveSession(); + + if (!result.answer) { + throw new Error('Kimi returned an empty response' + + (result.reasoning ? ' (thinking only)' : '')); + } + return result.answer; + } finally { + clearTimeout(timeoutId); + } + } + + function newConversation(sessionId) { + if (sessionId) { + delete _sessions[sessionId]; + if (_currentSessionId === sessionId) { + _currentSessionId = null; + } + } else if (_currentSessionId) { + delete _sessions[_currentSessionId]; + _currentSessionId = null; + } else { + // Nothing active — reset the loose globals as a safety net. + _accessToken = null; + } + try { + localStorage.setItem('proxima_kimi_sessions', JSON.stringify(_sessions)); + } catch (e) { } + console.log('[Proxima Kimi] Conversation reset:', sessionId || _currentSessionId || 'current'); + } + // Kimi's web protocol does not expose a public upload endpoint from the + // page context; text files are inlined into the message upstream. + async function uploadFileToKimi() { + throw new Error('Kimi web API does not support direct file upload yet. Text files are inlined into the message instead.'); + } + + window.__proximaKimi = { send: send, newConversation: newConversation, uploadFileToKimi: uploadFileToKimi, getSessionStatus: getSessionStatus }; + console.log('[Proxima] Kimi engine loaded'); +})(); diff --git a/electron/providers/sender.cjs b/electron/providers/sender.cjs index cff251e..f879377 100644 --- a/electron/providers/sender.cjs +++ b/electron/providers/sender.cjs @@ -8,16 +8,16 @@ function sleep(ms) { return new Promise(resolve => setTimeout(resolve, ms)); } // Chain sends to the same provider sequentially to prevent BrowserView execution collisions. const _sendQueues = {}; -function sendMessageToProvider(provider, message, attachments = null, onChunk = null, conversationId = null) { +function sendMessageToProvider(provider, message, attachments = null, onChunk = null, conversationId = null, think = true) { const baseProvider = (provider || '').split(':')[0]; - const run = () => _sendMessageToProviderImpl(provider, message, attachments, onChunk, conversationId); + const run = () => _sendMessageToProviderImpl(provider, message, attachments, onChunk, conversationId, think); const prior = _sendQueues[baseProvider] || Promise.resolve(); const next = prior.then(run, run); _sendQueues[baseProvider] = next.then(() => { }, () => { }); return next; } -async function _sendMessageToProviderImpl(provider, message, attachments = null, onChunk = null, conversationId = null) { +async function _sendMessageToProviderImpl(provider, message, attachments = null, onChunk = null, conversationId = null, think = true) { const baseProvider = provider.split(':')[0]; let webContents = browserManager.getWebContents(baseProvider); if (!webContents) { @@ -42,7 +42,7 @@ async function _sendMessageToProviderImpl(provider, message, attachments = null, for (let attempt = 1; attempt <= MAX_TRANSIENT_RETRIES; attempt++) { try { console.log(`[${provider}] API attempt ${attempt}/${MAX_TRANSIENT_RETRIES}...`); - const apiResponse = await providerAPI.sendViaAPI(provider, webContents, message, attachments, onChunk, conversationId); + const apiResponse = await providerAPI.sendViaAPI(provider, webContents, message, attachments, onChunk, conversationId, think); if (apiResponse && apiResponse.length > 0) { console.log(`[${provider}] [OK] API response captured (${apiResponse.length} chars) on attempt ${attempt}`); diff --git a/src/agentic/smart-router.js b/src/agentic/smart-router.js index eacea8c..6f6535f 100644 --- a/src/agentic/smart-router.js +++ b/src/agentic/smart-router.js @@ -25,6 +25,12 @@ const PROVIDER_PROFILES = { weaknesses: ['code-gen', 'creative'], speedTier: 3, qualityTier: 2, + }, + kimi: { + strengths: ['long-context', 'code-gen', 'reasoning', 'general', 'writing', 'translation'], + weaknesses: ['web-search', 'current-events'], + speedTier: 2, + qualityTier: 2, } }; diff --git a/src/agentic/task-orchestrator.js b/src/agentic/task-orchestrator.js index 1e612e5..ef20b72 100644 --- a/src/agentic/task-orchestrator.js +++ b/src/agentic/task-orchestrator.js @@ -28,6 +28,11 @@ const PROVIDER_ROLES = { strengths: ['analysis', 'multimodal', 'research', 'code-gen', 'data'], weight: 7, }, + kimi: { + role: 'Scribe', + strengths: ['long-context', 'writing', 'translation', 'general', 'synthesis'], + weight: 6, + }, }; const TASK_PATTERNS = [ @@ -137,6 +142,10 @@ class TaskOrchestrator { assignments.push({ provider: 'gemini', role: 'analyst', assignment: 'analyze' }); } + if (available.includes('kimi') && !assignments.find(a => a.provider === 'kimi')) { + assignments.push({ provider: 'kimi', role: 'scribe', assignment: 'synthesize' }); + } + if (assignments.length === 0) { for (const p of available) { diff --git a/src/config/defaults.js b/src/config/defaults.js index 1780142..7dec8ba 100644 --- a/src/config/defaults.js +++ b/src/config/defaults.js @@ -22,7 +22,7 @@ export const DEFAULTS = { RPM_WINDOW_MS: 60000, - PROVIDER_ORDER: ['chatgpt', 'claude', 'perplexity', 'gemini'], + PROVIDER_ORDER: ['chatgpt', 'claude', 'perplexity', 'gemini', 'kimi'], PROVIDER_HEALTH_CHECK_INTERVAL_MS: 60000, ROUTER_SCORE_THRESHOLD: 15, @@ -45,6 +45,7 @@ export const DEFAULTS = { claude: 800000, gemini: 1500000, perplexity: 1000000, + kimi: 900000, }, @@ -58,4 +59,5 @@ export const PROVIDER_INFO = { claude: { name: 'Claude', url: 'https://claude.ai/' }, gemini: { name: 'Gemini', url: 'https://gemini.google.com/app' }, perplexity: { name: 'Perplexity', url: 'https://www.perplexity.ai/' }, + kimi: { name: 'Kimi', url: 'https://www.kimi.com/' }, }; diff --git a/src/cost/token-tracker.js b/src/cost/token-tracker.js index 9c6d80d..ee29971 100644 --- a/src/cost/token-tracker.js +++ b/src/cost/token-tracker.js @@ -14,6 +14,7 @@ const MODEL_PRICING = { 'claude': { input: 4.00, output: 20.00 }, 'gemini': { input: 2.50, output: 10.00 }, 'perplexity': { input: 5.00, output: 20.00 }, + 'kimi': { input: 3.00, output: 16.00 }, 'default': { input: 5.00, output: 20.00 }, }; @@ -26,6 +27,7 @@ function _pricingKey(model) { if (m.includes('claude') || m.includes('sonnet') || m.includes('opus') || m.includes('haiku')) return 'claude'; if (m.includes('gemini') || m.includes('bison') || m.includes('palm')) return 'gemini'; if (m.includes('sonar') || m.includes('perplexity') || m.includes('pplx')) return 'perplexity'; + if (m.includes('kimi') || m.includes('moonshot')) return 'kimi'; return 'default'; } diff --git a/src/mcp/helpers.js b/src/mcp/helpers.js index 55eca2e..0d2a585 100644 --- a/src/mcp/helpers.js +++ b/src/mcp/helpers.js @@ -46,7 +46,7 @@ export function getEnabledProviders(dirname) { console.error('[MCP] Error reading enabled providers:', e); } - return new Set(['chatgpt', 'claude', 'gemini', 'perplexity']); + return new Set(['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi']); } diff --git a/src/mcp/index.js b/src/mcp/index.js index c3b3667..f872f11 100644 --- a/src/mcp/index.js +++ b/src/mcp/index.js @@ -57,8 +57,9 @@ const perplexity = new AIProvider('perplexity', ipcClient, isEnabled); const chatgpt = new AIProvider('chatgpt', ipcClient, isEnabled); const claude = new AIProvider('claude', ipcClient, isEnabled); const gemini = new AIProvider('gemini', ipcClient, isEnabled); +const kimi = new AIProvider('kimi', ipcClient, isEnabled); -const allProviders = { perplexity, chatgpt, claude, gemini }; +const allProviders = { perplexity, chatgpt, claude, gemini, kimi }; (function _initByokProviders() { @@ -165,10 +166,10 @@ function pickBestProvider(taskType) { } const priorities = { - coding: ['claude', 'chatgpt', 'gemini', 'perplexity'], - research: ['perplexity', 'gemini', 'chatgpt', 'claude'], - general: ['claude', 'chatgpt', 'gemini', 'perplexity'], - review: ['claude', 'chatgpt', 'gemini', 'perplexity'], + coding: ['claude', 'chatgpt', 'gemini', 'perplexity', 'kimi'], + research: ['perplexity', 'gemini', 'chatgpt', 'claude', 'kimi'], + general: ['claude', 'chatgpt', 'gemini', 'perplexity', 'kimi'], + review: ['claude', 'chatgpt', 'gemini', 'perplexity', 'kimi'], }; const order = priorities[taskType] || priorities.general; for (const name of order) { @@ -265,8 +266,8 @@ server.resource( } ); -// The 4 session tool names that map 1:1 to a BYOK provider. -const SESSION_TOOL_MAP = { chatgpt: 'ask_chatgpt', claude: 'ask_claude', gemini: 'ask_gemini', perplexity: 'ask_perplexity' }; +// The 5 session tool names that map 1:1 to a BYOK provider. +const SESSION_TOOL_MAP = { chatgpt: 'ask_chatgpt', claude: 'ask_claude', gemini: 'ask_gemini', perplexity: 'ask_perplexity', kimi: 'ask_kimi' }; server.resource( 'models', 'proxima://models', @@ -302,7 +303,7 @@ server.resource( model: `${name} (web session)`, tool: SESSION_TOOL_MAP[name] || `ask_model('${name}', message)`, })); - hint = 'Session mode. Use ask_chatgpt, ask_claude, ask_gemini, ask_perplexity for browser-based providers.'; + hint = 'Session mode. Use ask_chatgpt, ask_claude, ask_gemini, ask_perplexity, ask_kimi for browser-based providers.'; } return { diff --git a/src/mcp/ipc-bridge.js b/src/mcp/ipc-bridge.js index 92b2001..e551ae0 100644 --- a/src/mcp/ipc-bridge.js +++ b/src/mcp/ipc-bridge.js @@ -283,13 +283,14 @@ export class AIProvider { } - async _doChat(message, filePath = null, engine = null) { + async _doChat(message, filePath = null, engine = null, think = true) { await this.ensureInitialized(); console.error(`[${this.name}] Sending message...`); const sendResult = await this.ipc.send('sendMessage', this.name, { message, filePath, engine, conversationId: 'mcp-session', + think, }); if (sendResult.response && sendResult.response.length > 0) { @@ -306,9 +307,9 @@ export class AIProvider { }); } - async chat(message, useCache = true, filePath = null, engine = null) { + async chat(message, useCache = true, filePath = null, engine = null, think = true) { - const cacheKey = `${message}\u0000${filePath || ''}\u0000${engine || ''}`; + const cacheKey = `${message}\u0000${filePath || ''}\u0000${engine || ''}\u0000${think ? 'think' : 'nothink'}`; if (useCache && this.cache.has(cacheKey)) { @@ -327,7 +328,7 @@ export class AIProvider { const responsePromise = this._queue.then(async () => { console.error(`[${this.name}] Processing request (${position} of ${this._queueLength})...`); - const response = await this._doChat(message, filePath, engine); + const response = await this._doChat(message, filePath, engine, think); this.cache.set(cacheKey, { response, time: Date.now() }); this._queueLength--; return response; diff --git a/src/mcp/pipeline.js b/src/mcp/pipeline.js index 1e222ef..4e2730e 100644 --- a/src/mcp/pipeline.js +++ b/src/mcp/pipeline.js @@ -74,7 +74,7 @@ export function createSmartChat(deps) { let response; try { response = await withRetry( - () => providerInstance.chat(processedMessage, true, options.filePath, engineOverride), + () => providerInstance.chat(processedMessage, true, options.filePath, engineOverride, options.think !== false), { maxRetries: 2, baseDelay: 1, label: providerName } ); smartRouterV2.recordSuccess(providerName, Date.now() - span.startedAt); diff --git a/src/mcp/tools-chat.js b/src/mcp/tools-chat.js index 1e99d47..94119d6 100644 --- a/src/mcp/tools-chat.js +++ b/src/mcp/tools-chat.js @@ -46,19 +46,20 @@ export function register(server, deps) { server.registerTool('ask_chatgpt', { title: 'Ask ChatGPT', - description: 'Send a message to ChatGPT specifically. Use ask_model for any other/BYOK provider, smart_query to auto-pick the best provider, or ask_all_ais to query several at once.', + description: 'Send a message to ChatGPT specifically (defaults to Thinking/think mode, like the "Think" toggle on chatgpt.com). Use ask_model for any other/BYOK provider, smart_query to auto-pick the best provider, or ask_all_ais to query several at once.', inputSchema: { message: z.string().describe('Message to send to ChatGPT'), files: z.array(z.string()).optional().describe(FILES_DESC), + think: z.boolean().optional().describe('Enable ChatGPT think mode (deep reasoning). Default: true. Set to false for instant/low-effort mode.'), }, annotations: CHAT, - }, async ({ message, files }) => { + }, async ({ message, files, think }) => { const disabled = checkDisabled('chatgpt'); if (disabled) return disabled; try { const { textFiles, uploadFilePath } = getFilesSetup(files); const fullMessage = buildMessageWithFiles(message, textFiles); - return toolResponse(await smartChat(fullMessage, 'chatgpt', { filePath: uploadFilePath })); + return toolResponse(await smartChat(fullMessage, 'chatgpt', { filePath: uploadFilePath, think: think !== false })); } catch (err) { return toolError(err); } @@ -124,18 +125,39 @@ export function register(server, deps) { } }); - const enabledList = [...getEnabledProviders()].join(', ') || 'gemini, chatgpt, claude, perplexity'; + server.registerTool('ask_kimi', { + title: 'Ask Kimi', + description: 'Send a message to Kimi specifically (Moonshot AI — strong at long-context). Use ask_model for other/BYOK providers, smart_query to auto-pick, or ask_all_ais for several at once.', + inputSchema: { + message: z.string().describe('Message to send to Kimi'), + files: z.array(z.string()).optional().describe(FILES_DESC), + }, + annotations: CHAT, + }, async ({ message, files }) => { + const disabled = checkDisabled('kimi'); + if (disabled) return disabled; + try { + const { textFiles, uploadFilePath } = getFilesSetup(files); + const fullMessage = buildMessageWithFiles(message, textFiles); + return toolResponse(await smartChat(fullMessage, 'kimi', { filePath: uploadFilePath })); + } catch (err) { + return toolError(err); + } + }); + + const enabledList = [...getEnabledProviders()].join(', ') || 'gemini, chatgpt, claude, perplexity, kimi'; server.registerTool('ask_model', { title: 'Ask Any Model', - description: 'Universal chat: send a message to ANY enabled provider by name (the 4 session providers OR any configured BYOK provider). Use this when you need a provider other than the four dedicated ask_* tools.', + description: 'Universal chat: send a message to ANY enabled provider by name (the session providers OR any configured BYOK provider). Use this when you need a provider other than the dedicated ask_* tools.', inputSchema: { provider: z.string().describe(`Provider name. Currently available: ${enabledList}`), message: z.string().describe('Message to send'), model: z.string().optional().describe('Specific model ID override (uses provider default if omitted)'), files: z.array(z.string()).optional().describe('Optional: file paths to include as context. Supports line ranges like "path/file.js:10-50".'), + think: z.boolean().optional().describe('Enable think/deep-reasoning mode (ChatGPT and compatible providers). Default: true.'), }, annotations: CHAT, - }, async ({ provider, message, model, files }) => { + }, async ({ provider, message, model, files, think }) => { const providerName = provider.toLowerCase().trim(); const disabled = checkDisabled(providerName); if (disabled) return disabled; @@ -146,6 +168,7 @@ export function register(server, deps) { return toolResponse(await smartChat(fullMessage, providerName, { filePath: uploadFilePath, engine: model || null, + think: think !== false, })); } catch (err) { return toolError(err); @@ -253,7 +276,7 @@ export function register(server, deps) { title: 'New Conversation (reset)', description: 'Reset conversation memory/context for a provider (or all enabled providers if none named). Use when you want a fresh thread with no prior context.', inputSchema: { - provider: z.string().optional().describe('Which provider to reset: chatgpt, claude, gemini, or perplexity. If omitted, resets all enabled providers.'), + provider: z.string().optional().describe('Which provider to reset: chatgpt, claude, gemini, perplexity, or kimi. If omitted, resets all enabled providers.'), }, annotations: Object.freeze({ readOnlyHint: false, destructiveHint: false, idempotentHint: true, openWorldHint: true }), }, async ({ provider }) => { @@ -268,7 +291,7 @@ export function register(server, deps) { return toolResponse({ success: true, provider: name, message: `Started new ${name} conversation` }); } const reset = []; - for (const p of ['perplexity', 'chatgpt', 'claude', 'gemini']) { + for (const p of ['perplexity', 'chatgpt', 'claude', 'gemini', 'kimi']) { if (enabled.has(p)) { await allProviders[p].newConversation(); reset.push(p); diff --git a/src/mcp/tools-code.js b/src/mcp/tools-code.js index a1c6c98..9a98f02 100644 --- a/src/mcp/tools-code.js +++ b/src/mcp/tools-code.js @@ -20,7 +20,7 @@ export function register(server, deps) { }); - const PROVIDER_DESC = 'AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select best for coding.'; + const PROVIDER_DESC = 'AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select best for coding.'; server.registerTool('verify_code', { title: 'Verify Code', @@ -186,7 +186,7 @@ export function register(server, deps) { inputSchema: { error: z.string().describe('Error message or stack trace to explain'), context: z.string().optional().describe('What you were doing when the error occurred'), - provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select.'), + provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select.'), }, annotations: ANALYSIS, }, async ({ error, context: ctx, provider: pn }) => diff --git a/src/mcp/tools-content.js b/src/mcp/tools-content.js index 6ea0a08..c426ee6 100644 --- a/src/mcp/tools-content.js +++ b/src/mcp/tools-content.js @@ -24,7 +24,7 @@ export function register(server, deps) { input: z.string().describe('Main input — URL, topic, text, or task description'), detail: z.string().optional().describe('Extra context — focus area, writing style, data type to extract, specific question'), body: z.string().optional().describe('Content body — text to improve, or content to analyze/extract from'), - provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select'), + provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select'), }, annotations: GEN, }, async ({ action, input, detail, body, provider: pn }) => { @@ -51,7 +51,7 @@ export function register(server, deps) { item1: z.string().describe('First item to compare'), item2: z.string().describe('Second item to compare'), context: z.string().optional().describe('Context for comparison'), - provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select'), + provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select'), }, annotations: GEN, }, async ({ item1, item2, context, provider: pn }) => { diff --git a/src/mcp/tools-search.js b/src/mcp/tools-search.js index 14b9822..e1f6b07 100644 --- a/src/mcp/tools-search.js +++ b/src/mcp/tools-search.js @@ -37,7 +37,7 @@ export function register(server, deps) { timeframe: z.string().optional().describe('For news: timeframe like "today", "this week", "2024"'), year: z.string().optional().describe('For stats: specific year'), files: z.array(z.string()).optional().describe('Optional: file paths for context. Supports line ranges like "path/file.js:10-50"'), - provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select'), + provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select'), }, annotations: READ_OPEN, }, async ({ query, type, language, timeframe, year, files, provider: pn }) => { @@ -68,7 +68,7 @@ export function register(server, deps) { code: z.string().optional().describe('Optional: existing code to analyze and apply design improvements on'), files: z.array(z.string()).optional().describe('Optional: file paths of existing code to improve with better UI/UX. Supports line ranges like "path/file.js:10-50"'), style: z.string().optional().describe('Design style preference: modern, minimal, glassmorphism, dark, corporate, playful, etc.'), - provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, or any configured BYOK provider. Default: auto-select best for coding'), + provider: z.string().optional().describe('AI provider: chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider. Default: auto-select best for coding'), }, annotations: READ_OPEN, }, async ({ description, code, files, style, provider: pn }) => { diff --git a/src/mcp/tools-utility.js b/src/mcp/tools-utility.js index baef0be..8f5fbb4 100644 --- a/src/mcp/tools-utility.js +++ b/src/mcp/tools-utility.js @@ -33,7 +33,7 @@ export function register(server, deps) { inputSchema: { filePath: z.string().describe('Absolute path to the file or directory to analyze'), question: z.string().optional().describe('Specific question about the file/codebase'), - provider: z.string().optional().describe('Which AI to use (chatgpt, claude, gemini, perplexity, or any configured BYOK provider). Default: claude'), + provider: z.string().optional().describe('Which AI to use (chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider). Default: claude'), grep: z.string().optional().describe('Optional regex pattern to search within the file/codebase before analysis'), symbols: z.string().optional().describe('Comma-separated function/class names to extract from file. E.g. "smartChat,toolResponse,guardrails". Auto-resolves dependencies.'), }, @@ -127,7 +127,7 @@ export function register(server, deps) { inputSchema: { filePath: z.string().describe('Absolute path to the code file to review'), focus: z.string().optional().describe('What to focus on (bugs, performance, security, style)'), - provider: z.string().optional().describe('Which AI to use (chatgpt, claude, gemini, perplexity, or any configured BYOK provider). Default: claude'), + provider: z.string().optional().describe('Which AI to use (chatgpt, claude, gemini, perplexity, kimi, or any configured BYOK provider). Default: claude'), }, annotations: Object.freeze({ readOnlyHint: true, destructiveHint: false, idempotentHint: false, openWorldHint: true }), }, async ({ filePath, focus, provider: pn }) => { diff --git a/tests/config/defaults.test.js b/tests/config/defaults.test.js index 8218d5d..55b73a8 100644 --- a/tests/config/defaults.test.js +++ b/tests/config/defaults.test.js @@ -21,10 +21,10 @@ test('retry delays are internally consistent (base <= max)', () => { assert.ok(DEFAULTS.MAX_RETRIES >= 0); }); -test('PROVIDER_ORDER lists the four session providers', () => { +test('PROVIDER_ORDER lists the session providers', () => { assert.deepEqual( [...DEFAULTS.PROVIDER_ORDER].sort(), - ['chatgpt', 'claude', 'gemini', 'perplexity'], + ['chatgpt', 'claude', 'gemini', 'kimi', 'perplexity'], ); }); diff --git a/tests/electron/providers/sender.test.js b/tests/electron/providers/sender.test.js index b0cc81a..abc46ba 100644 --- a/tests/electron/providers/sender.test.js +++ b/tests/electron/providers/sender.test.js @@ -92,3 +92,14 @@ test('sendMessageToProvider: a failing send does not wedge the queue for the nex const r = await sender.sendMessageToProvider('gemini', 'b'); assert.deepEqual(r, { response: 'recovered' }); }); + +test('sendMessageToProvider: defaults think=true and forwards the think flag to sendViaAPI', async () => { + let captured; + sendImpl = async (...args) => { captured = args; return 'ok'; }; + await sender.sendMessageToProvider('chatgpt', 'hi'); + assert.equal(captured[6], true, 'think must default to true'); + + captured = null; + await sender.sendMessageToProvider('chatgpt', 'hi', null, null, null, false); + assert.equal(captured[6], false, 'explicit think=false must be forwarded'); +}); diff --git a/tests/fixtures/mcp-harness.js b/tests/fixtures/mcp-harness.js index eb9731c..0f55b0a 100644 --- a/tests/fixtures/mcp-harness.js +++ b/tests/fixtures/mcp-harness.js @@ -5,7 +5,7 @@ import { z } from 'zod'; import { toolResponse, toolError } from '../../src/mcp/helpers.js'; export function makeHarness(overrides = {}) { - const enabled = new Set(overrides.enabled || ['chatgpt', 'claude', 'gemini', 'perplexity']); + const enabled = new Set(overrides.enabled || ['chatgpt', 'claude', 'gemini', 'perplexity', 'kimi']); const smartChatCalls = []; const ipcCalls = []; const tools = new Map(); diff --git a/tests/mcp/helpers.test.js b/tests/mcp/helpers.test.js index 309c28f..eb86734 100644 --- a/tests/mcp/helpers.test.js +++ b/tests/mcp/helpers.test.js @@ -47,9 +47,9 @@ function write(name, obj) { } -test('getEnabledProviders: returns the 4 core providers by default when no config exists', () => { +test('getEnabledProviders: returns the 5 core providers by default when no config exists', () => { const set = getEnabledProviders(); - assert.deepEqual([...set].sort(), ['chatgpt', 'claude', 'gemini', 'perplexity']); + assert.deepEqual([...set].sort(), ['chatgpt', 'claude', 'gemini', 'kimi', 'perplexity']); }); test('getEnabledProviders: session mode reads enabled-providers.json', () => { diff --git a/tests/mcp/ipc-bridge.test.js b/tests/mcp/ipc-bridge.test.js index f08c95b..77bb237 100644 --- a/tests/mcp/ipc-bridge.test.js +++ b/tests/mcp/ipc-bridge.test.js @@ -244,6 +244,21 @@ test('AIProvider.chat: cache key includes filePath (same text + different file assert.equal(sendMessageCalls, 2, 'a different attached file must not reuse the cached answer'); }); +test('AIProvider.chat: forwards think=true by default and includes think in the cache key', async () => { + const sends = []; + const ipc = mockIpc(async (action, provider, data) => { + if (action === 'initProvider') return {}; + if (action === 'sendMessage') { sends.push({ ...data }); return { response: 'r' }; } + return {}; + }); + const p = new AIProvider('claude', ipc, () => true); + await p.chat('q'); + await p.chat('q', true, null, null, false); + assert.equal(sends.length, 2, 'toggling think must not reuse the cached answer'); + assert.equal(sends[0].think, true, 'think must default to true'); + assert.equal(sends[1].think, false, 'explicit think=false must be forwarded'); +}); + test('AIProvider.chat: gateway {success:false,error:"rate limit"} throws a RateLimitError', async () => { const ipc = mockIpc(async (action) => (action === 'initProvider' ? {} : { success: false, error: 'rate limit exceeded' })); const p = new AIProvider('claude', ipc, () => true); diff --git a/tests/mcp/tools-chat.test.js b/tests/mcp/tools-chat.test.js index 7141598..6ce578a 100644 --- a/tests/mcp/tools-chat.test.js +++ b/tests/mcp/tools-chat.test.js @@ -6,11 +6,11 @@ import assert from 'node:assert'; import { register } from '../../src/mcp/tools-chat.js'; import { registerModule, textOf } from '../fixtures/mcp-harness.js'; -test('tools-chat: registers exactly the 8 documented chat tools with schemas', () => { +test('tools-chat: registers exactly the 9 documented chat tools with schemas', () => { const { tools } = registerModule(register); assert.deepEqual( [...tools.keys()].sort(), - ['ask_all_ais', 'ask_chatgpt', 'ask_claude', 'ask_gemini', 'ask_model', 'new_conversation', 'smart_query'].sort() + ['ask_all_ais', 'ask_chatgpt', 'ask_claude', 'ask_gemini', 'ask_kimi', 'ask_model', 'new_conversation', 'smart_query'].sort() .concat('ask_perplexity').sort(), ); // Every tool exposes a non-empty description and an inputSchema object. @@ -43,6 +43,24 @@ test('ask_claude: a smartChat failure is returned as an isError toolError', asyn assert.match(textOf(res), /provider down/); }); +test('ask_chatgpt: defaults to think mode (think=true) and honors an explicit think=false', async () => { + const h = registerModule(register); + await h.tools.get('ask_chatgpt').handler({ message: 'hello' }); + assert.equal(h.smartChatCalls[0].provider, 'chatgpt'); + assert.deepEqual(h.smartChatCalls[0].opts, { filePath: null, think: true }); + + await h.tools.get('ask_chatgpt').handler({ message: 'hello', think: false }); + assert.equal(h.smartChatCalls[1].opts.think, false); +}); + +test('ask_model: forwards think default true and honors an explicit think=false', async () => { + const h = registerModule(register); + await h.tools.get('ask_model').handler({ provider: 'chatgpt', message: 'hi' }); + assert.equal(h.smartChatCalls[0].opts.think, true); + await h.tools.get('ask_model').handler({ provider: 'chatgpt', message: 'hi', think: false }); + assert.equal(h.smartChatCalls[1].opts.think, false); +}); + test('ask_model: normalizes provider case and forwards the model override as engine', async () => { const h = registerModule(register); await h.tools.get('ask_model').handler({ provider: 'Claude', message: 'hi', model: 'claude-x' });