@nxuss/lemma 0.8.0 → 0.8.2

package/dist/esm/cli/lemma-proxy.js

@@ -38,25 +38,22 @@ import SemanticScrubber from '../security/SemanticScrubber';
 import CloudSyncClient from '../cloud/CloudSyncClient';
 import { pruneHistoryMessages, squeezePrompt } from '../utils/ContextSqueezer';
 import { AgentMultiplexer } from '../proxy/AgentMultiplexer';
+import { SseRelay } from '../proxy/SseRelay';
+import { KeyManager } from '../proxy/KeyManager';
+import { Gatekeeper } from '../proxy/Gatekeeper';
 import { ProjectRegistry } from '../proxy/ProjectRegistry';
 import { shouldInject, buildMinimalContext, buildDeepContext, sniffProjectFromMessages } from '../proxy/ContextInjector';
 import { savingsLedger } from '../utils/SavingsLedger';
+import { CACHE_DIR, PID_FILE, PORT_FILE, STATS_FILE, USAGE_FILE, LICENSE_FILE, CLIPBOARD_PID_FILE, FREE_TIER_LIMIT } from '../proxy/ProjectStore';
 const chroma = new ChromaClient({ host: 'localhost', port: 8000 });
 const dummyEmbeddingFunction = { generate: (texts) => Promise.resolve([]) };
 const complexityRouter = new ComplexityRouter();
 const semanticScrubber = new SemanticScrubber();
 const cloudSync = new CloudSyncClient();
 const agentMultiplexer = new AgentMultiplexer();
+const keyManager = new KeyManager();
+const gatekeeper = new Gatekeeper(keyManager);
 // ── Paths ──────────────────────────────────────────────────────────────────────
-const HOME = os.homedir();
-const CACHE_DIR = path.join(HOME, '.lemma-cache');
-const PID_FILE = path.join(CACHE_DIR, 'proxy.pid');
-const PORT_FILE = path.join(CACHE_DIR, 'proxy.port');
-const STATS_FILE = path.join(CACHE_DIR, 'stats.json');
-const USAGE_FILE = path.join(CACHE_DIR, 'usage.json');
-const LICENSE_FILE = path.join(CACHE_DIR, 'license.json');
-const CLIPBOARD_PID_FILE = path.join(CACHE_DIR, 'clipboard.pid');
-const FREE_LIMIT = 300;
 const VALIDATE_URL = 'https://lemma.nxus.studio/api/validate-license';
 async function checkDependency(url) {
     try {
@@ -504,21 +501,8 @@ export function startBackgroundClipboardWatcher(cliOpts) {
     }
 }
 // ── License helpers ────────────────────────────────────────────────────────────
-let isProCached = null;
-let lastLicenseCheck = 0;
-const LICENSE_CACHE_TTL = 10000; // 10 seconds cache
-async function loadLicense() {
-    return readJson(LICENSE_FILE, { isPro: false });
-}
-async function isPro() {
-    const now = Date.now();
-    if (isProCached !== null && (now - lastLicenseCheck) < LICENSE_CACHE_TTL) {
-        return isProCached;
-    }
-    const l = await loadLicense();
-    isProCached = !!(l && l.isPro);
-    lastLicenseCheck = now;
-    return isProCached;
+function isPro() {
+    return gatekeeper.getMode() === 'pro';
 }
 async function validateKeyRemote(key) {
     return new Promise((resolve, reject) => {
@@ -642,7 +626,7 @@ async function getEmbedding(text) {
     }
 }
 async function semanticGet(provider, prompt, projectName = 'global') {
-    if (!await isPro())
+    if (!isPro())
         return null;
     const emb = await getEmbedding(prompt);
     if (!emb)
@@ -735,7 +719,7 @@ async function semanticGet(provider, prompt, projectName = 'global') {
     return null;
 }
 async function semanticSet(provider, prompt, data, projectName = 'global') {
-    if (!await isPro())
+    if (!isPro())
         return;
     const emb = await getEmbedding(prompt);
     if (!emb)
@@ -838,53 +822,6 @@ function setSseHeaders(res, fromCache, similarity, tier) {
     res.setHeader('X-Lemma-Similarity', (similarity || 1.0).toFixed(3));
     res.setHeader('X-Lemma-Tier', tier);
 }
-function buildOaiChunk(id, model, text, stop) {
-    return { id, object: 'chat.completion.chunk', created: Math.floor(Date.now() / 1000), model,
-        choices: [{ index: 0, delta: stop ? {} : { content: text }, finish_reason: stop ? 'stop' : null }] };
-}
-function buildAntChunk(text, stop) {
-    if (stop)
-        return { type: 'message_stop' };
-    return { type: 'content_block_delta', index: 0, delta: { type: 'text_delta', text } };
-}
-function buildGemChunk(text, stop) {
-    return { candidates: [{ content: { parts: [{ text }] }, finishReason: stop ? 'STOP' : undefined }] };
-}
-async function simulateHitStream(res, cachedData, provider, model, tier, similarity, cacheHeader = 'HIT') {
-    setSseHeaders(res, cacheHeader, similarity, tier);
-    const id = `lemma-${Date.now()}`;
-    let text = '';
-    try {
-        if (provider === 'openai')
-            text = cachedData?.choices?.[0]?.message?.content || '';
-        else if (provider === 'anthropic') {
-            const c = cachedData?.content;
-            text = Array.isArray(c) ? c.map((x) => x.text || '').join('') : String(c || '');
-        }
-        else if (provider === 'gemini') {
-            text = cachedData?.candidates?.[0]?.content?.parts?.[0]?.text || '';
-        }
-    }
-    catch { }
-    const chunkSize = 30;
-    for (let i = 0; i < text.length; i += chunkSize) {
-        const chunk = text.slice(i, i + chunkSize);
-        let evt;
-        if (provider === 'openai')
-            evt = buildOaiChunk(id, model, chunk, false);
-        else if (provider === 'anthropic')
-            evt = buildAntChunk(chunk, false);
-        else if (provider === 'gemini')
-            evt = buildGemChunk(chunk, false);
-        res.write(provider === 'gemini' ? `${JSON.stringify(evt)}\n` : `data: ${JSON.stringify(evt)}\n\n`);
-        await new Promise(r => setTimeout(r, 15));
-    }
-    const stopEvt = provider === 'openai' ? buildOaiChunk(id, model, '', true) : (provider === 'anthropic' ? buildAntChunk('', true) : buildGemChunk('', true));
-    res.write(provider === 'gemini' ? `${JSON.stringify(stopEvt)}\n` : `data: ${JSON.stringify(stopEvt)}\n\n`);
-    if (provider !== 'gemini')
-        res.write('data: [DONE]\n\n');
-    res.end();
-}
 async function relayMissStream(res, body, provider, tier) {
     setSseHeaders(res, false, 1.0, tier);
     const apiKey = provider === 'openai' ? process.env.OPENAI_API_KEY : process.env.ANTHROPIC_API_KEY;
@@ -994,8 +931,8 @@ class LemmaServer {
     }
     setupRoutes() {
         this.app.get('/health', async (req, res) => {
-            const pro = await isPro();
-            res.json({ status: 'ok', project: this.projectName, mode: pro ? 'pro' : 'standard', tier: pro ? 'pro' : 'free', port: this.port });
+            const gs = gatekeeper.getStatus();
+            res.json({ status: 'ok', project: this.projectName, mode: gs.mode, tier: gs.isPro ? 'pro' : 'free', port: this.port });
         });
         // MCP Status endpoint
         this.app.get('/api/mcp-status', (req, res) => {
@@ -1094,7 +1031,7 @@ class LemmaServer {
                     activeAgents,
                     totalAgents,
                     uptime: process.uptime(),
-                    tier: (await isPro()) ? 'pro' : 'free'
+                    tier: (isPro()) ? 'pro' : 'free'
                 });
             });
             this.app.get('/api/agents', async (req, res) => {
@@ -1681,14 +1618,14 @@ class LemmaServer {
     }
     async handleCompletion(req, res, provider) {
         const t0 = Date.now();
-        const pro = await isPro();
+        const pro = isPro();
         const tier = pro ? 'pro' : 'standard';
         const isStream = !!req.body.stream;
         let historyTokensSaved = 0;
         if (!pro) {
             const usage = await loadUsage();
-            if (usage.count >= FREE_LIMIT) {
-                return res.status(429).json({ error: { message: `Free tier limit (${FREE_LIMIT}/mo). Upgrade: https://lemma.nxus.studio/upgrade` } });
+            if (usage.count >= FREE_TIER_LIMIT) {
+                return res.status(429).json({ error: { message: `Free tier limit (${FREE_TIER_LIMIT}/mo). Upgrade: https://lemma.nxus.studio/upgrade` } });
             }
         }
         const rawPrompt = extractPrompt(req.body, provider);
@@ -1707,7 +1644,7 @@ class LemmaServer {
             await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, cacheTokensSaved);
             const unmaskedData = semanticScrubber.unmask(hit.data, tokenMap);
             if (isStream)
-                return simulateHitStream(res, unmaskedData, provider, originalModel, tier, 1.0);
+                return SseRelay.simulateHit(res, unmaskedData, { provider: provider, tier, similarity: 1.0 });
             res.setHeader('X-Lemma-Cache', 'HIT');
             res.setHeader('X-Lemma-Similarity', '1.000');
             res.setHeader('X-Lemma-Tier', tier);
@@ -1722,7 +1659,7 @@ class LemmaServer {
                     const unmaskedData = semanticScrubber.unmask(semHit.data, tokenMap);
                     const cacheHeader = semHit.hiveMind ? 'HIT-HIVE-MIND' : 'HIT';
                     if (isStream)
-                        return simulateHitStream(res, unmaskedData, provider, originalModel, tier, semHit.similarity, cacheHeader);
+                        return SseRelay.simulateHit(res, unmaskedData, { provider: provider, tier, similarity: semHit.similarity, cacheHeader });
                     res.setHeader('X-Lemma-Cache', cacheHeader);
                     res.setHeader('X-Lemma-Similarity', semHit.similarity.toFixed(3));
                     res.setHeader('X-Lemma-Tier', tier);
@@ -1772,7 +1709,7 @@ Adjusted Answer:`;
                                 await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000);
                                 const unmaskedData = semanticScrubber.unmask(synthesizedData, tokenMap);
                                 if (isStream) {
-                                    return simulateHitStream(res, unmaskedData, provider, originalModel, tier, semHit.similarity);
+                                    return SseRelay.simulateHit(res, unmaskedData, { provider: provider, tier, similarity: semHit.similarity });
                                 }
                                 res.setHeader('X-Lemma-Cache', 'HIT-CARS');
                                 res.setHeader('X-Lemma-Similarity', semHit.similarity.toFixed(3));
@@ -1792,7 +1729,7 @@ Adjusted Answer:`;
                 await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000);
                 const unmaskedData = semanticScrubber.unmask(cloudHit.data, tokenMap);
                 if (isStream)
-                    return simulateHitStream(res, unmaskedData, provider, originalModel, tier, cloudHit.similarity || 0.95);
+                    return SseRelay.simulateHit(res, unmaskedData, { provider: provider, tier, similarity: cloudHit.similarity || 0.95 });
                 res.setHeader('X-Lemma-Cache', 'HIT');
                 res.setHeader('X-Lemma-Similarity', (cloudHit.similarity || 0.95).toFixed(3));
                 res.setHeader('X-Lemma-Tier', tier);
@@ -1999,14 +1936,23 @@ Adjusted Answer:`;
             }
             throw e;
         });
-        server.listen(this.port, async () => {
+        const BIND_HOST = process.env.LEMMA_BIND_HOST || '127.0.0.1';
+        server.listen(this.port, BIND_HOST, async () => {
             fs.writeFileSync(PID_FILE, String(process.pid));
             fs.writeFileSync(PORT_FILE, String(this.port));
-            const pro = await isPro();
+            // Security warning if binding to all interfaces without auth
+            if (BIND_HOST !== '127.0.0.1' && BIND_HOST !== 'localhost') {
+                const authEnabled = process.env.AUTH_ENABLED === 'true';
+                if (!authEnabled) {
+                    console.warn(`\n\x1b[33m⚠️  [SECURITY] Lemma proxy is bound to ${BIND_HOST} (all interfaces) with AUTH_ENABLED=false.`);
+                    console.warn(`   Set AUTH_ENABLED=true and AUTH_REQUIRE_API_KEY=true before exposing to any network.\x1b[0m\n`);
+                }
+            }
+            const pro = isPro();
             console.log(`\x1b[1m\x1b[35m┌──────────────────────────────────────────────────────────┐\x1b[0m`);
             console.log(`\x1b[1m\x1b[35m│   🚀 LEMMA PROXY v${VERSION} — THE UNIVERSAL AI GATEWAY      │\x1b[0m`);
             console.log(`\x1b[1m\x1b[35m└──────────────────────────────────────────────────────────┘\x1b[0m`);
-            console.log(`\n📁 Project : ${this.projectName}\n🔌 Port    : ${this.port}`);
+            console.log(`\n📁 Project : ${this.projectName}\n🔌 Port    : ${this.port}  (bound to ${BIND_HOST})`);
             if (this.dashboardPath && fs.existsSync(this.dashboardPath)) {
                 console.log(`📊 Dashboard UI          : \x1b[36mhttp://localhost:${this.port}/dashboard/\x1b[0m`);
                 if (this.openBrowser) {
@@ -2045,12 +1991,19 @@ Adjusted Answer:`;
             console.log(`\n   \x1b[1m\x1b[33mlemma mcp\x1b[0m\n`);
             console.log('────────────────────────────────────────────────\n');
             // Auto-start the Local Clipboard Optimizer Watcher in the background
-            try {
-                const { startClipboardWatcher } = require('../utils/ClipboardWatcher');
-                startClipboardWatcher(1000);
+            const clipboardDisabled = process.env.LEMMA_DISABLE_CLIPBOARD === 'true';
+            if (clipboardDisabled) {
+                console.log(`📋 Clipboard Watcher     : \x1b[90m[DISABLED via LEMMA_DISABLE_CLIPBOARD]\x1b[0m`);
             }
-            catch (err) {
-                console.warn(`⚠️ Could not auto-initialize Clipboard Watcher: ${err.message}`);
+            else {
+                console.log(`📋 Clipboard Watcher     : \x1b[33m[ACTIVE]\x1b[0m — polling clipboard every 1s (disable: LEMMA_DISABLE_CLIPBOARD=true)`);
+                try {
+                    const { startClipboardWatcher } = require('../utils/ClipboardWatcher');
+                    startClipboardWatcher(1000);
+                }
+                catch (err) {
+                    console.warn(`⚠️ Could not auto-initialize Clipboard Watcher: ${err.message}`);
+                }
             }
         });
         process.on('SIGTERM', () => server.close());
@@ -2089,7 +2042,7 @@ program
     .version(VERSION)
     .addHelpText('after', `
 \x1b[1mPlan Comparison\x1b[0m
-  \x1b[32mFree\x1b[0m    ${FREE_LIMIT} req/mo · Exact cache · Privacy firewall · MCP tools
+  \x1b[32mFree\x1b[0m    ${FREE_TIER_LIMIT} req/mo · Exact cache · Privacy firewall · MCP tools
   \x1b[35mPro\x1b[0m     Unlimited  · Semantic cache · Cloud sync · Autopilot · Heal
          \x1b[36mhttps://lemma.nxus.studio/upgrade\x1b[0m
 
@@ -2165,7 +2118,7 @@ program.command('start')
     const runAutopilot = opts.autopilot || opts.stack;
     if (runAutopilot) {
         try {
-            const pro = await isPro();
+            const pro = isPro();
             if (!pro) {
                 if (opts.autopilot) {
                     console.log(`\n🛸 \x1b[35m[Autopilot]\x1b[0m Active background auto-healing watcher is a Lemma Pro feature.`);
@@ -2309,9 +2262,9 @@ program.command('stats')
     console.log(`Cache Hits     : ${s.hits} (${((s.hits / s.total || 0) * 100).toFixed(1)}%)`);
     console.log(`Tokens Saved   : ${s.totalTokensSaved.toLocaleString()}`);
     console.log(`Avg Latency    : ${(s.totalLatency / (s.total || 1)).toFixed(2)}ms`);
-    console.log(`Tier           : ${(await isPro()) ? 'PRO' : 'FREE'}`);
-    if (!await isPro())
-        console.log(`Free Usage     : ${usage.count}/${FREE_LIMIT} requests`);
+    console.log(`Tier           : ${(isPro()) ? 'PRO' : 'FREE'}`);
+    if (!isPro())
+        console.log(`Free Usage     : ${usage.count}/${FREE_TIER_LIMIT} requests`);
     console.log(`──────────────────────────────────\n`);
 });
 program.command('activate <key>')
@@ -2327,6 +2280,7 @@ program.command('activate <key>')
         const res = await validateKeyRemote(key.trim());
         if (res.valid) {
             await writeJson(LICENSE_FILE, { key, isPro: true, activatedAt: new Date().toISOString(), validatedRemotely: true });
+            gatekeeper.reload();
             console.log('✅ License activated! Enjoy Pro features (Server-verified).');
         }
         else {
@@ -2567,7 +2521,7 @@ program.command('heal')
     .option('--apply', 'Apply the auto-generated code fix automatically', false)
     .action(async (opts) => {
     try {
-        const pro = await isPro();
+        const pro = isPro();
         if (!pro) {
             console.log(`\n🩺 \x1b[35m[Heal]\x1b[0m Guided code diagnostics and auto-healing is a Lemma Pro feature.`);
             console.log(`💡 Upgrade to Pro in your Dashboard or visit: \x1b[4mhttps://lemma.nxus.studio/upgrade\x1b[0m\n`);
@@ -2596,7 +2550,7 @@ program.command('autopilot')
     .option('--dir <directory>', 'Directory to monitor for changes', 'src')
     .action(async (opts) => {
     try {
-        const pro = await isPro();
+        const pro = isPro();
         if (!pro) {
             console.log(`\n🛸 \x1b[35m[Autopilot]\x1b[0m Active background auto-healing watcher is a Lemma Pro feature.`);
             console.log(`💡 Upgrade to Pro in your Dashboard or visit: \x1b[4mhttps://lemma.nxus.studio/upgrade\x1b[0m\n`);