@nxuss/lemma 0.5.2 → 0.5.3

package/README.md

@@ -1,4 +1,4 @@
-# 🧠 Lemma v0.5.2
+# 🧠 Lemma v0.5.3
 
 [![npm version](https://img.shields.io/npm/v/@nxuss/lemma.svg?color=orange&style=flat-square)](https://www.npmjs.com/package/@nxuss/lemma)
 [![License](https://img.shields.io/npm/l/@nxuss/lemma.svg?color=blue&style=flat-square)](https://github.com/Nxusbets/lemma/blob/main/LICENSE)

package/dist/cjs/cli/lemma-proxy.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"lemma-proxy.d.ts","sourceRoot":"","sources":["../../../src/cli/lemma-proxy.ts"],"names":[],"mappings":";AAoNA,wBAAgB,gBAAgB,CAAC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CA0I3D;AAED,wBAAgB,+BAA+B,IAAI,IAAI,CAyCtD;AAmsCD,wBAAsB,eAAe,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CA6KhL"}
+{"version":3,"file":"lemma-proxy.d.ts","sourceRoot":"","sources":["../../../src/cli/lemma-proxy.ts"],"names":[],"mappings":";AA8NA,wBAAgB,gBAAgB,CAAC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CA0I3D;AAED,wBAAgB,+BAA+B,IAAI,IAAI,CAyCtD;AAuxCD,wBAAsB,eAAe,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CA6KhL"}

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

@@ -32,7 +32,7 @@ function getVersion() {
             catch { }
         }
     }
-    return '0.5.2';
+    return '0.5.3';
 }
 const VERSION = getVersion();
 const ComplexityRouter_1 = __importDefault(require("../proxy/ComplexityRouter"));
@@ -86,13 +86,20 @@ async function readJson(file, fallback) {
     catch { }
     return fallback;
 }
+const writeQueues = new Map();
 async function writeJson(file, data) {
-    try {
-        const tempFile = file + '.tmp';
-        await fs_1.default.promises.writeFile(tempFile, JSON.stringify(data, null, 2));
-        await fs_1.default.promises.rename(tempFile, file);
-    }
-    catch { }
+    const chain = writeQueues.get(file) || Promise.resolve();
+    const writeOperation = async () => {
+        try {
+            const tempFile = `${file}.${crypto_1.default.randomUUID ? crypto_1.default.randomUUID() : Math.random().toString(36).substring(2, 15)}.tmp`;
+            await fs_1.default.promises.writeFile(tempFile, JSON.stringify(data, null, 2));
+            await fs_1.default.promises.rename(tempFile, file);
+        }
+        catch { }
+    };
+    const nextChain = chain.then(writeOperation).catch(() => { });
+    writeQueues.set(file, nextChain);
+    return nextChain;
 }
 function projectHash(name) {
     return crypto_1.default.createHash('sha1').update(name).digest('hex').slice(0, 12);
@@ -386,12 +393,21 @@ function startBackgroundClipboardWatcher() {
     }
 }
 // ── 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) => {
@@ -484,7 +500,15 @@ 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) {
@@ -529,7 +553,7 @@ 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),
@@ -756,6 +780,16 @@ 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) {
@@ -885,24 +919,57 @@ 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
-                SavingsLedger_1.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(() => { });
+            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_1.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}`);
+                    }
                 }
-                res.json({ success: true, clearedEntries: count });
+                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) => {
@@ -946,89 +1013,36 @@ 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;
-        // Record complexity routing savings when model is downgraded
-        if (routingDecision.complexity === 'low' && model !== originalModel) {
-            // Estimated token savings: cheap model costs ~5x less, so avg prompt of ~800 tokens saves ~640 tokens
-            const estimatedPromptTokens = Math.floor(prompt.length / 4);
-            const routingSavings = Math.floor(estimatedPromptTokens * 0.8); // conservative 80% cost reduction
-            SavingsLedger_1.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_1.default.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 = (0, ContextSqueezer_1.squeezePrompt)(enrichedPrompt, prompt, fileHistory);
-        const squeezedPrompt = squeezeResult.squeezed;
-        const compressionSavings = squeezeResult.originalSize - squeezeResult.squeezedSize;
-        if (compressionSavings > 0) {
-            SavingsLedger_1.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 to save tokens on historical messages!
-                const pruneResult = (0, ContextSqueezer_1.pruneHistoryMessages)(req.body.messages);
-                req.body.messages = pruneResult.messages;
-                historyTokensSaved = pruneResult.tokensSavedEstimate + Math.floor(compressionSavings / 4);
-                if (pruneResult.tokensSavedEstimate > 0) {
-                    SavingsLedger_1.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;
-                }
-            }
-        }
+        // ── 1. Cache-First Hot Path ──
+        // Perform exact cache check immediately (resolves in <1ms)
         const hit = cacheGet(provider, prompt);
         if (hit) {
-            const cacheTokensSaved = 2000 + historyTokensSaved;
+            const cacheTokensSaved = 2000;
             SavingsLedger_1.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') {
@@ -1066,10 +1080,10 @@ 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));
@@ -1086,22 +1100,86 @@ 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_1.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_1.default.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 = (0, ContextSqueezer_1.squeezePrompt)(enrichedPrompt, prompt, fileHistory);
+        const squeezedPrompt = squeezeResult.squeezed;
+        const compressionSavings = squeezeResult.originalSize - squeezeResult.squeezedSize;
+        if (compressionSavings > 0) {
+            SavingsLedger_1.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 = (0, ContextSqueezer_1.pruneHistoryMessages)(req.body.messages);
+                req.body.messages = pruneResult.messages;
+                historyTokensSaved = pruneResult.tokensSavedEstimate + Math.floor(compressionSavings / 4);
+                if (pruneResult.tokensSavedEstimate > 0) {
+                    SavingsLedger_1.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);
         }
@@ -1113,9 +1191,11 @@ 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));
@@ -1152,6 +1232,15 @@ 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_1.default.createServer(this.app);
         server.on('error', (e) => {
             if (e.code === 'EADDRINUSE') {