@nxuss/lemma 0.5.1 → 0.5.3

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

@@ -24,7 +24,7 @@ function getVersion() {
             catch { }
         }
     }
-    return '0.4.6';
+    return '0.5.3';
 }
 const VERSION = getVersion();
 import ComplexityRouter from '../proxy/ComplexityRouter';
@@ -32,6 +32,7 @@ import SemanticScrubber from '../security/SemanticScrubber';
 import CloudSyncClient from '../cloud/CloudSyncClient';
 import { pruneHistoryMessages, squeezePrompt } from '../utils/ContextSqueezer';
 import { AgentMultiplexer } from '../proxy/AgentMultiplexer';
+import { savingsLedger } from '../utils/SavingsLedger';
 const chroma = new ChromaClient({ host: 'localhost', port: 8000 });
 const dummyEmbeddingFunction = { generate: (texts) => Promise.resolve([]) };
 const complexityRouter = new ComplexityRouter();
@@ -53,6 +54,13 @@ function logEvent(event) {
     EVENT_LOG.push(ev);
     if (EVENT_LOG.length > MAX_EVENTS)
         EVENT_LOG.shift();
+    // Print a terminal savings summary every 10 events
+    if (EVENT_LOG.length > 0 && EVENT_LOG.length % 10 === 0) {
+        const snap = savingsLedger.getSnapshot();
+        if (snap.total.tokensSaved > 0) {
+            console.log('\n' + savingsLedger.getTerminalSummary() + '\n');
+        }
+    }
 }
 const FREE_LIMIT = 500;
 const WARN_PCT = 0.8;
@@ -70,13 +78,20 @@ async function readJson(file, fallback) {
     catch { }
     return fallback;
 }
+const writeQueues = new Map();
 async function writeJson(file, data) {
-    try {
-        const tempFile = file + '.tmp';
-        await fs.promises.writeFile(tempFile, JSON.stringify(data, null, 2));
-        await fs.promises.rename(tempFile, file);
-    }
-    catch { }
+    const chain = writeQueues.get(file) || Promise.resolve();
+    const writeOperation = async () => {
+        try {
+            const tempFile = `${file}.${crypto.randomUUID ? crypto.randomUUID() : Math.random().toString(36).substring(2, 15)}.tmp`;
+            await fs.promises.writeFile(tempFile, JSON.stringify(data, null, 2));
+            await fs.promises.rename(tempFile, file);
+        }
+        catch { }
+    };
+    const nextChain = chain.then(writeOperation).catch(() => { });
+    writeQueues.set(file, nextChain);
+    return nextChain;
 }
 function projectHash(name) {
     return crypto.createHash('sha1').update(name).digest('hex').slice(0, 12);
@@ -193,13 +208,198 @@ function ensureGitIgnore() {
     catch { }
     return added;
 }
+const CLIPBOARD_PID_FILE = path.join(CACHE_DIR, 'clipboard.pid');
+export function autoConfigureAll(projectName) {
+    const project = projectName || detectProject();
+    const HOME = process.env.HOME || process.env.USERPROFILE || '~';
+    // 1. Claude Desktop Config
+    try {
+        const isWin = process.platform === 'win32';
+        const claudePath = isWin
+            ? path.join(process.env.APPDATA || '', 'Claude/claude_desktop_config.json')
+            : path.join(HOME, 'Library/Application Support/Claude/claude_desktop_config.json');
+        if (fs.existsSync(path.dirname(claudePath))) {
+            let config = { mcpServers: {} };
+            if (fs.existsSync(claudePath)) {
+                try {
+                    config = JSON.parse(fs.readFileSync(claudePath, 'utf8'));
+                }
+                catch { }
+            }
+            if (!config.mcpServers)
+                config.mcpServers = {};
+            config.mcpServers.lemma = {
+                command: 'npx',
+                args: ['-y', '@nxuss/lemma', 'mcp'],
+                env: { LEMMA_PROJECT: project }
+            };
+            fs.writeFileSync(claudePath, JSON.stringify(config, null, 2));
+            console.log(`✅ [Claude Desktop] Configured automatically! (Restart Claude to activate)`);
+        }
+    }
+    catch (e) {
+        console.log(`⚠️  Could not auto-configure Claude Desktop: ${e.message}`);
+    }
+    // 2. Shell Profiles Environment Override
+    try {
+        const profiles = [
+            path.join(HOME, '.zshrc'),
+            path.join(HOME, '.bashrc'),
+            path.join(HOME, '.bash_profile'),
+            path.join(HOME, '.profile')
+        ];
+        const lines = [
+            '',
+            '# Lemma AI Gateway Overrides',
+            'export OPENAI_BASE_URL="http://localhost:8081/v1"',
+            'export ANTHROPIC_BASE_URL="http://localhost:8081"',
+            'export LEMMA_PROJECT="' + project + '"',
+            ''
+        ].join('\n');
+        for (const profile of profiles) {
+            if (fs.existsSync(profile)) {
+                try {
+                    const content = fs.readFileSync(profile, 'utf8');
+                    if (!content.includes('OPENAI_BASE_URL') && !content.includes('LEMMA_PROJECT')) {
+                        fs.appendFileSync(profile, lines);
+                        console.log(`✅ [Shell Profile] Configured ${path.basename(profile)} with local redirect variables!`);
+                    }
+                }
+                catch { }
+            }
+        }
+    }
+    catch (e) {
+        console.log(`⚠️  Could not auto-configure shell profiles: ${e.message}`);
+    }
+    // 3. Cursor & Windsurf settings.json updates
+    try {
+        const isWin = process.platform === 'win32';
+        const isMac = process.platform === 'darwin';
+        const candidates = [];
+        if (isMac) {
+            candidates.push({
+                name: 'Cursor',
+                path: path.join(HOME, 'Library/Application Support/Cursor/User/settings.json')
+            });
+            candidates.push({
+                name: 'Windsurf',
+                path: path.join(HOME, 'Library/Application Support/Windsurf/User/settings.json')
+            });
+        }
+        else if (isWin) {
+            const appData = process.env.APPDATA || '';
+            candidates.push({
+                name: 'Cursor',
+                path: path.join(appData, 'Cursor/User/settings.json')
+            });
+            candidates.push({
+                name: 'Windsurf',
+                path: path.join(appData, 'Windsurf/User/settings.json')
+            });
+        }
+        else {
+            candidates.push({
+                name: 'Cursor',
+                path: path.join(HOME, '.config/Cursor/User/settings.json')
+            });
+            candidates.push({
+                name: 'Windsurf',
+                path: path.join(HOME, '.config/Windsurf/User/settings.json')
+            });
+        }
+        for (const cand of candidates) {
+            const dir = path.dirname(cand.path);
+            if (fs.existsSync(dir)) {
+                let settings = {};
+                if (fs.existsSync(cand.path)) {
+                    try {
+                        settings = JSON.parse(fs.readFileSync(cand.path, 'utf8'));
+                    }
+                    catch { }
+                }
+                let updated = false;
+                // Settings configuration to route custom openai/anthropic models through the proxy
+                const updates = {
+                    "openai.baseURL": "http://localhost:8081/v1",
+                    "openai.apiKey": "dummy-key-for-lemma",
+                    "anthropic.baseURL": "http://localhost:8081",
+                    "anthropic.apiKey": "dummy-key-for-lemma"
+                };
+                for (const [key, value] of Object.entries(updates)) {
+                    if (settings[key] !== value) {
+                        settings[key] = value;
+                        updated = true;
+                    }
+                }
+                if (updated) {
+                    fs.writeFileSync(cand.path, JSON.stringify(settings, null, 2));
+                    console.log(`✅ [${cand.name}] Configured editor settings to route via local gateway!`);
+                }
+            }
+        }
+    }
+    catch (e) {
+        console.log(`⚠️  Could not auto-configure editor settings: ${e.message}`);
+    }
+}
+export function startBackgroundClipboardWatcher() {
+    try {
+        if (fs.existsSync(CLIPBOARD_PID_FILE)) {
+            const oldPid = fs.readFileSync(CLIPBOARD_PID_FILE, 'utf8');
+            try {
+                process.kill(parseInt(oldPid, 10), 0); // Check if process is still alive
+                console.log(`✅ [Clipboard] Watcher is already running in background (PID: ${oldPid})`);
+                return;
+            }
+            catch {
+                // Process is dead, clean up pid file
+                try {
+                    fs.unlinkSync(CLIPBOARD_PID_FILE);
+                }
+                catch { }
+            }
+        }
+        console.log(`🔄 Starting Clipboard Watcher in background...`);
+        const { spawn } = require('child_process');
+        // Resolve path of lemma command or script
+        const scriptPath = fs.existsSync(path.join(__dirname, 'dist', 'cjs', 'cli', 'lemma-proxy.js'))
+            ? path.join(__dirname, 'dist', 'cjs', 'cli', 'lemma-proxy.js')
+            : (fs.existsSync(path.join(__dirname, 'lemma-proxy.ts')) ? path.join(__dirname, 'lemma-proxy.ts') : __filename);
+        const isTs = scriptPath.endsWith('.ts');
+        const cmd = isTs ? 'npx' : 'node';
+        const args = isTs ? ['tsx', scriptPath, 'clipboard'] : [scriptPath, 'clipboard'];
+        const cp = spawn(cmd, args, {
+            stdio: 'ignore',
+            detached: true,
+            env: { ...process.env }
+        });
+        cp.unref();
+        if (cp.pid) {
+            fs.writeFileSync(CLIPBOARD_PID_FILE, String(cp.pid));
+            console.log(`✅ [Clipboard] Watcher has been started (PID: ${cp.pid})`);
+        }
+    }
+    catch (err) {
+        console.error(`❌ Failed to automatically start Clipboard Watcher in background: ${err.message}`);
+    }
+}
 // ── 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();
-    return !!(l && l.isPro);
+    isProCached = !!(l && l.isPro);
+    lastLicenseCheck = now;
+    return isProCached;
 }
 async function validateKeyRemote(key) {
     return new Promise((resolve, reject) => {
@@ -292,7 +492,15 @@ function cacheGet(provider, prompt) {
     return e;
 }
 function cacheSet(provider, prompt, data) {
-    CACHE.set(cacheKey(provider, prompt), { input: prompt, data, createdAt: Date.now(), ttl: TTL_MS });
+    const MAX_CACHE_SIZE = 5000;
+    const k = cacheKey(provider, prompt);
+    if (!CACHE.has(k) && CACHE.size >= MAX_CACHE_SIZE) {
+        // Evict oldest cache entry (FIFO eviction)
+        const oldestKey = CACHE.keys().next().value;
+        if (oldestKey)
+            CACHE.delete(oldestKey);
+    }
+    CACHE.set(k, { input: prompt, data, createdAt: Date.now(), ttl: TTL_MS, provider });
 }
 // ── Semantic Search (Pro) ──────────────────────────────────────────────────────
 async function getEmbedding(text) {
@@ -337,7 +545,7 @@ async function semanticGet(provider, prompt) {
                 distance = distances[0][i];
             }
             const similarity = Math.max(0, 1 - distance);
-            const THRESHOLD = 0.7;
+            const THRESHOLD = 0.82; // raised from 0.7 to 0.82 to prevent false positives
             if (similarity >= THRESHOLD) {
                 return {
                     data: JSON.parse(metadata.response),
@@ -564,6 +772,16 @@ async function relayMissStream(res, body, provider, tier) {
         req.end();
     });
 }
+class WriteQueue {
+    constructor() {
+        this.queue = Promise.resolve();
+    }
+    async enqueue(fn) {
+        this.queue = this.queue.then(fn).catch(() => { });
+        return this.queue;
+    }
+}
+const usageWriteQueue = new WriteQueue();
 // ── Proxy Server class ─────────────────────────────────────────────────────────
 class LemmaServer {
     constructor(port, projectName, openBrowser = true) {
@@ -693,23 +911,76 @@ class LemmaServer {
                     res.status(500).json({ error: e.message });
                 }
             });
-            this.app.post('/api/cache/clear', (req, res) => {
-                const count = CACHE.size;
-                CACHE.clear();
-                EVENT_LOG.length = 0; // Clear live event log
-                // Reset and persist statistics for the current project
-                if (this.stats[this.projectName]) {
-                    this.stats[this.projectName] = {
-                        total: 0,
-                        hits: 0,
-                        misses: 0,
-                        totalLatency: 0,
-                        totalTokensSaved: 0,
-                        providers: {}
-                    };
-                    writeJson(STATS_FILE, this.stats).catch(() => { });
+            this.app.post('/api/cache/clear', async (req, res) => {
+                const { provider } = req.body || {};
+                let clearedCount = 0;
+                if (!provider) {
+                    // Global clear
+                    clearedCount = CACHE.size;
+                    CACHE.clear();
+                    EVENT_LOG.length = 0; // Clear live event log
+                    savingsLedger.reset(); // Reset savings counters
+                    // Reset and persist statistics for the current project
+                    if (this.stats[this.projectName]) {
+                        this.stats[this.projectName] = {
+                            total: 0,
+                            hits: 0,
+                            misses: 0,
+                            totalLatency: 0,
+                            totalTokensSaved: 0,
+                            providers: {}
+                        };
+                        writeJson(STATS_FILE, this.stats).catch(() => { });
+                    }
+                    // Clear Chroma DB collection
+                    try {
+                        await chroma.deleteCollection({ name: 'lemma-cache' }).catch(() => { });
+                    }
+                    catch (e) {
+                        console.error(`[CacheClear] Failed to clear Chroma collection: ${e.message}`);
+                    }
+                }
+                else {
+                    // Granular clear
+                    for (const [k, e] of CACHE.entries()) {
+                        if (e.provider === provider) {
+                            CACHE.delete(k);
+                            clearedCount++;
+                        }
+                    }
+                    // Selective delete from Chroma DB collection using provider metadata filter
+                    try {
+                        const collection = await chroma.getOrCreateCollection({
+                            name: 'lemma-cache',
+                            embeddingFunction: dummyEmbeddingFunction,
+                            metadata: { "hnsw:space": "cosine" }
+                        });
+                        await collection.delete({ where: { provider } });
+                    }
+                    catch (e) {
+                        console.error(`[CacheClear] Failed to selectively delete provider ${provider} from Chroma: ${e.message}`);
+                    }
+                }
+                res.json({ success: true, clearedEntries: clearedCount });
+            });
+            // Savings breakdown by source for the dashboard
+            this.app.get('/api/savings-breakdown', (req, res) => {
+                res.json(savingsLedger.getSnapshot());
+            });
+            // Cross-process savings reporter — MCP and other processes POST here
+            this.app.post('/api/savings-record', (req, res) => {
+                const { source, charsBefore, charsAfter, tokens } = req.body || {};
+                const validSources = ['clipboard', 'contextSqueeze', 'historyPrune', 'complexityRouting', 'cache'];
+                if (!source || !validSources.includes(source)) {
+                    return res.status(400).json({ error: 'Invalid source' });
+                }
+                if (typeof tokens === 'number' && tokens > 0) {
+                    savingsLedger.recordTokens(source, tokens);
                 }
-                res.json({ success: true, clearedEntries: count });
+                else if (typeof charsBefore === 'number' && typeof charsAfter === 'number') {
+                    savingsLedger.record(source, charsBefore, charsAfter);
+                }
+                res.json({ ok: true, snapshot: savingsLedger.getSnapshot().total });
             });
             this.app.get('/dashboard/*', (req, res) => res.sendFile(path.join(this.dashboardPath, 'index.html')));
         }
@@ -734,76 +1005,36 @@ class LemmaServer {
             logEvent({ type: 'privacy:mask', count: Object.keys(tokenMap).length, project: this.projectName });
         }
         const originalModel = req.body.model || 'gpt-4o';
-        const routingDecision = complexityRouter.evaluate(prompt, originalModel);
-        const model = routingDecision.model;
-        if (req.body.model)
-            req.body.model = model;
-        // Apply the Agent Multiplexer to local-verify the prompt in parallel
-        const enrichedPrompt = await agentMultiplexer.multiplexPrompt(prompt, model);
-        // Resolve conversation session key and retrieve/create the fileHistory map
-        let fileHistory;
-        if (req.body.messages && req.body.messages.length > 0) {
-            try {
-                const firstMsg = req.body.messages[0];
-                const content = typeof firstMsg.content === 'string' ? firstMsg.content : JSON.stringify(firstMsg.content);
-                const sessionKey = crypto.createHash('sha1').update(content).digest('hex');
-                if (!this.sessions.has(sessionKey)) {
-                    this.sessions.set(sessionKey, { fileHistory: new Map() });
-                }
-                fileHistory = this.sessions.get(sessionKey).fileHistory;
-            }
-            catch { }
-        }
-        // Compress long TS/JS codebase file blocks in the enriched prompt using the Context Squeezer AST Tree-Shaker!
-        // The query context is the user's primary prompt to ensure we keep relevant methods!
-        const squeezeResult = squeezePrompt(enrichedPrompt, prompt, fileHistory);
-        const squeezedPrompt = squeezeResult.squeezed;
-        const compressionSavings = squeezeResult.originalSize - squeezeResult.squeezedSize;
-        if (compressionSavings > 0) {
-            console.log(`⚡ [Squeezer] Compressed prompt from ${squeezeResult.originalSize} to ${squeezeResult.squeezedSize} chars (Saved ${(squeezeResult.compressionRatio * 100).toFixed(1)}% context tokens!)`);
-        }
-        if (provider === 'openai' || provider === 'anthropic') {
-            if (req.body.messages && req.body.messages.length > 0) {
-                injectSqueezedContent(req.body.messages[req.body.messages.length - 1], squeezedPrompt);
-                // Apply conversational history pruning to save tokens on historical messages!
-                const pruneResult = pruneHistoryMessages(req.body.messages);
-                req.body.messages = pruneResult.messages;
-                historyTokensSaved = pruneResult.tokensSavedEstimate + Math.floor(compressionSavings / 4);
-            }
-        }
-        else if (provider === 'gemini') {
-            if (req.body.contents && req.body.contents.length > 0) {
-                const lastContent = req.body.contents[req.body.contents.length - 1];
-                if (lastContent.parts && lastContent.parts.length > 0) {
-                    lastContent.parts[lastContent.parts.length - 1].text = squeezedPrompt;
-                }
-            }
-        }
+        // ── 1. Cache-First Hot Path ──
+        // Perform exact cache check immediately (resolves in <1ms)
         const hit = cacheGet(provider, prompt);
         if (hit) {
-            await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000 + historyTokensSaved);
+            const cacheTokensSaved = 2000;
+            savingsLedger.recordTokens('cache', cacheTokensSaved);
+            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, model, tier, 1.0);
+                return simulateHitStream(res, unmaskedData, provider, originalModel, tier, 1.0);
             res.setHeader('X-Lemma-Cache', 'HIT');
             res.setHeader('X-Lemma-Similarity', '1.000');
             res.setHeader('X-Lemma-Tier', tier);
             return res.json(unmaskedData);
         }
+        // Perform semantic cache check (resolves in ~15ms)
         if (pro) {
             const semHit = await semanticGet(provider, prompt);
             if (semHit) {
                 if (semHit.similarity >= 0.90) {
-                    await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000 + historyTokensSaved);
+                    await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000);
                     const unmaskedData = semanticScrubber.unmask(semHit.data, tokenMap);
                     if (isStream)
-                        return simulateHitStream(res, unmaskedData, provider, model, tier, semHit.similarity);
+                        return simulateHitStream(res, unmaskedData, provider, originalModel, tier, semHit.similarity);
                     res.setHeader('X-Lemma-Cache', 'HIT');
                     res.setHeader('X-Lemma-Similarity', semHit.similarity.toFixed(3));
                     res.setHeader('X-Lemma-Tier', tier);
                     return res.json(unmaskedData);
                 }
-                else if (semHit.similarity >= 0.70) {
+                else if (semHit.similarity >= 0.82) { // raised from 0.70 to 0.82 to prevent false positives
                     console.log(`💡 [CARS] Near Cache Hit (Similarity: ${(semHit.similarity * 100).toFixed(1)}%). Synthesizing answer locally via Ollama...`);
                     let cachedContent = '';
                     if (provider === 'openai') {
@@ -841,10 +1072,10 @@ Adjusted Answer:`;
                                 else if (provider === 'anthropic') {
                                     synthesizedData.content[0].text = adjustedText;
                                 }
-                                await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000 + historyTokensSaved);
+                                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, model, tier, semHit.similarity);
+                                    return simulateHitStream(res, unmaskedData, provider, originalModel, tier, semHit.similarity);
                                 }
                                 res.setHeader('X-Lemma-Cache', 'HIT-CARS');
                                 res.setHeader('X-Lemma-Similarity', semHit.similarity.toFixed(3));
@@ -861,22 +1092,86 @@ Adjusted Answer:`;
             const cloudHit = await cloudSync.get(prompt);
             if (cloudHit) {
                 cacheSet(provider, prompt, cloudHit.data);
-                await recordStat(this.stats, this.projectName, true, Date.now() - t0, provider, 2000 + historyTokensSaved);
+                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, model, tier, cloudHit.similarity || 0.95);
+                    return simulateHitStream(res, unmaskedData, provider, originalModel, tier, 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);
                 return res.json(unmaskedData);
             }
         }
+        // ── 2. Cache Miss Path ──
+        const routingDecision = complexityRouter.evaluate(prompt, originalModel);
+        const model = routingDecision.model;
+        if (req.body.model)
+            req.body.model = model;
+        // Record complexity routing savings when model is downgraded
+        if (routingDecision.complexity === 'low' && model !== originalModel) {
+            const estimatedPromptTokens = Math.floor(prompt.length / 4);
+            const routingSavings = Math.floor(estimatedPromptTokens * 0.8);
+            savingsLedger.recordTokens('complexityRouting', routingSavings);
+        }
+        // Apply the Agent Multiplexer to local-verify the prompt in parallel
+        const enrichedPrompt = await agentMultiplexer.multiplexPrompt(prompt, model);
+        // Resolve conversation session key and retrieve/create the fileHistory map
+        let fileHistory;
+        if (req.body.messages && req.body.messages.length > 0) {
+            try {
+                const firstMsg = req.body.messages[0];
+                const content = typeof firstMsg.content === 'string' ? firstMsg.content : JSON.stringify(firstMsg.content);
+                const sessionKey = crypto.createHash('sha1').update(content).digest('hex');
+                if (!this.sessions.has(sessionKey)) {
+                    // Enforce 500 active sessions cap to prevent memory leak (FIFO eviction)
+                    const MAX_SESSIONS = 500;
+                    if (this.sessions.size >= MAX_SESSIONS) {
+                        const oldestKey = this.sessions.keys().next().value;
+                        if (oldestKey)
+                            this.sessions.delete(oldestKey);
+                    }
+                    this.sessions.set(sessionKey, { fileHistory: new Map() });
+                }
+                fileHistory = this.sessions.get(sessionKey).fileHistory;
+            }
+            catch { }
+        }
+        // Compress long TS/JS codebase file blocks in the enriched prompt
+        const squeezeResult = squeezePrompt(enrichedPrompt, prompt, fileHistory);
+        const squeezedPrompt = squeezeResult.squeezed;
+        const compressionSavings = squeezeResult.originalSize - squeezeResult.squeezedSize;
+        if (compressionSavings > 0) {
+            savingsLedger.record('contextSqueeze', squeezeResult.originalSize, squeezeResult.squeezedSize);
+            console.log(`⚡ [Squeezer] Compressed prompt from ${squeezeResult.originalSize} to ${squeezeResult.squeezedSize} chars (Saved ${(squeezeResult.compressionRatio * 100).toFixed(1)}% context tokens!)`);
+        }
+        if (provider === 'openai' || provider === 'anthropic') {
+            if (req.body.messages && req.body.messages.length > 0) {
+                injectSqueezedContent(req.body.messages[req.body.messages.length - 1], squeezedPrompt);
+                // Apply conversational history pruning
+                const pruneResult = pruneHistoryMessages(req.body.messages);
+                req.body.messages = pruneResult.messages;
+                historyTokensSaved = pruneResult.tokensSavedEstimate + Math.floor(compressionSavings / 4);
+                if (pruneResult.tokensSavedEstimate > 0) {
+                    savingsLedger.recordTokens('historyPrune', pruneResult.tokensSavedEstimate);
+                }
+            }
+        }
+        else if (provider === 'gemini') {
+            if (req.body.contents && req.body.contents.length > 0) {
+                const lastContent = req.body.contents[req.body.contents.length - 1];
+                if (lastContent.parts && lastContent.parts.length > 0) {
+                    lastContent.parts[lastContent.parts.length - 1].text = squeezedPrompt;
+                }
+            }
+        }
         if (isStream) {
             await recordStat(this.stats, this.projectName, false, Date.now() - t0, provider, historyTokensSaved);
             if (!pro) {
-                const u = await loadUsage();
-                u.count++;
-                await writeJson(USAGE_FILE, u);
+                await usageWriteQueue.enqueue(async () => {
+                    const u = await loadUsage();
+                    u.count++;
+                    await writeJson(USAGE_FILE, u);
+                });
             }
             return relayMissStream(res, req.body, provider, tier);
         }
@@ -888,9 +1183,11 @@ Adjusted Answer:`;
                 await cloudSync.set(prompt, data);
             }
             if (!pro) {
-                const u = await loadUsage();
-                u.count++;
-                await writeJson(USAGE_FILE, u);
+                await usageWriteQueue.enqueue(async () => {
+                    const u = await loadUsage();
+                    u.count++;
+                    await writeJson(USAGE_FILE, u);
+                });
             }
             await recordStat(this.stats, this.projectName, false, Date.now() - t0, provider, historyTokensSaved);
             return res.json(semanticScrubber.unmask(data, tokenMap));
@@ -927,6 +1224,15 @@ Adjusted Answer:`;
     }
     async start() {
         this.stats = await loadStats();
+        // Start active cache eviction sweep every hour
+        setInterval(() => {
+            const now = Date.now();
+            for (const [k, e] of CACHE.entries()) {
+                if (e.ttl > 0 && now - e.createdAt > e.ttl) {
+                    CACHE.delete(k);
+                }
+            }
+        }, 3600000).unref();
         const server = http.createServer(this.app);
         server.on('error', (e) => {
             if (e.code === 'EADDRINUSE') {
@@ -1046,10 +1352,17 @@ program.command('start')
         if (resp.data && resp.data.status === 'ok') {
             console.log(`\n⚠️  Lemma Proxy is already running on port ${port} (Project: ${resp.data.project})`);
             console.log(`💡 Use 'lemma status' for details or 'lemma stop' to restart.\n`);
+            // Keep configurations up to date even if server was already started
+            const runningProject = resp.data.project || detectProject();
+            autoConfigureAll(runningProject);
+            startBackgroundClipboardWatcher();
             process.exit(0);
         }
     }
     catch { }
+    const startProject = opts.project || detectProject();
+    autoConfigureAll(startProject);
+    startBackgroundClipboardWatcher();
     // Automatically enable autopilot if stack is enabled and user is Pro
     const runAutopilot = opts.autopilot || opts.stack;
     if (runAutopilot) {
@@ -1148,6 +1461,19 @@ program.command('stop')
     }
     if (fs.existsSync(PID_FILE))
         fs.unlinkSync(PID_FILE);
+    // Stop background Clipboard Watcher if active
+    try {
+        if (fs.existsSync(CLIPBOARD_PID_FILE)) {
+            const cbPid = fs.readFileSync(CLIPBOARD_PID_FILE, 'utf8');
+            try {
+                process.kill(parseInt(cbPid, 10), 'SIGTERM');
+                console.log(`✅ Stopped background Clipboard Watcher (PID: ${cbPid})`);
+            }
+            catch { }
+            fs.unlinkSync(CLIPBOARD_PID_FILE);
+        }
+    }
+    catch { }
 });
 program.command('status')
     .description('Check proxy status')
@@ -1241,41 +1567,8 @@ program.command('init')
         fs.writeFileSync(envFile, lemmaConfig);
         console.log('✅ Created .env with Lemma configuration');
     }
-    console.log('\n🧠 \x1b[35mAuto-Setup MCP for AI IDEs?\x1b[0m');
-    console.log('   Lemma can automatically configure Claude Desktop and open settings for Cursor/Windsurf.');
-    const { exec } = require('child_process');
-    const isWin = process.platform === 'win32';
-    const claudePath = isWin
-        ? path.join(process.env.APPDATA || '', 'Claude/claude_desktop_config.json')
-        : path.join(process.env.HOME || '~', 'Library/Application Support/Claude/claude_desktop_config.json');
-    if (fs.existsSync(path.dirname(claudePath))) {
-        try {
-            let config = { mcpServers: {} };
-            if (fs.existsSync(claudePath)) {
-                config = JSON.parse(fs.readFileSync(claudePath, 'utf8'));
-            }
-            if (!config.mcpServers)
-                config.mcpServers = {};
-            config.mcpServers.lemma = {
-                command: 'npx',
-                args: ['-y', '@nxuss/lemma', 'mcp'],
-                env: { LEMMA_PROJECT: project }
-            };
-            fs.writeFileSync(claudePath, JSON.stringify(config, null, 2));
-            console.log(`✅ \x1b[32mClaude Desktop configured automatically!\x1b[0m (Restart Claude to activate)`);
-        }
-        catch (e) {
-            console.log('⚠️  Could not auto-configure Claude Desktop. No worries, you can do it manually.');
-        }
-    }
-    console.log('🚀 Opening IDE configuration pages...');
-    const urls = [
-        'https://docs.anthropic.com/en/docs/agents-and-tools/mcp#using-mcp-in-claude-desktop',
-        'https://cursor.com',
-        'https://codeium.com/windsurf'
-    ];
-    const opener = process.platform === 'darwin' ? 'open' : (process.platform === 'win32' ? 'start' : 'xdg-open');
-    urls.forEach(url => exec(`${opener} ${url}`));
+    console.log('\n🧠 \x1b[35mAuto-Setup MCP and redirection overrides for AI IDEs...\x1b[0m');
+    autoConfigureAll(project);
     console.log('\n✨ Project initialized for the Agentic Era!');
     console.log('🚀 Run "lemma start" to begin.\n');
 });