@massa-ai/core 1.2.1

package/dist/services/metrics/token-metrics.js

@@ -0,0 +1,272 @@
+import { logger, estimateTokens } from "@massa-ai/shared";
+// ── Constants ────────────────────────────────────────────────────
+/** models.dev API endpoint */
+const MODELS_DEV_API = "https://models.dev/api.json";
+/** Cache duration for models.dev pricing data (1 hour) */
+const PRICING_CACHE_TTL_MS = 60 * 60 * 1000;
+/** Fallback pricing for common models when API is unavailable (per 1M tokens in USD) */
+const FALLBACK_PRICING = {
+    "gpt-4": { input: 30, output: 60 },
+    "gpt-4-turbo": { input: 10, output: 30 },
+    "gpt-4o": { input: 2.5, output: 10 },
+    "gpt-4o-mini": { input: 0.15, output: 0.6 },
+    "gpt-3.5-turbo": { input: 0.5, output: 1.5 },
+    "claude-3-opus": { input: 15, output: 75 },
+    "claude-3-opus-20240229": { input: 15, output: 75 },
+    "claude-opus-4-20250514": { input: 15, output: 75 },
+    "claude-3-sonnet": { input: 3, output: 15 },
+    "claude-3-haiku": { input: 0.25, output: 1.25 },
+    "claude-3-5-sonnet": { input: 3, output: 15 },
+    "claude-sonnet-4.5": { input: 3, output: 15 },
+    "claude-3-5-haiku": { input: 0.8, output: 4 },
+    "gemini-1.5-pro": { input: 1.25, output: 5 },
+    "gemini-1.5-flash": { input: 0.075, output: 0.3 },
+    "gemini-2.0-flash-exp": { input: 0, output: 0 },
+};
+/** Maximum time-series history entries to keep */
+const MAX_HISTORY_ENTRIES = 1000;
+// ── Service ──────────────────────────────────────────────────────
+/**
+ * Singleton token metrics collector with models.dev integration.
+ */
+export class TokenMetrics {
+    static instance = null;
+    counters = {
+        sessionCache: 0,
+        redundancyFilter: 0,
+        compression: 0,
+    };
+    totalTokensProcessed = 0;
+    totalTokensDelivered = 0;
+    requestsServed = 0;
+    /** Time-series history for tracking trends */
+    history = [];
+    /** Pricing cache: model ID -> pricing data */
+    pricingCache = new Map();
+    constructor() { }
+    static getInstance() {
+        if (!TokenMetrics.instance) {
+            TokenMetrics.instance = new TokenMetrics();
+        }
+        return TokenMetrics.instance;
+    }
+    // ── Pricing API (models.dev integration) ───────────────────────
+    /**
+     * Get model pricing from models.dev API with caching and fallback.
+     *
+     * @param modelId - Model identifier (e.g., "gpt-4", "claude-3-5-sonnet")
+     * @returns Pricing data or fallback if unavailable
+     */
+    async getModelPricing(modelId) {
+        // Check cache first
+        const cached = this.pricingCache.get(modelId);
+        if (cached && Date.now() - cached.timestamp < PRICING_CACHE_TTL_MS) {
+            return cached.pricing;
+        }
+        try {
+            // Fetch from models.dev API
+            const response = await fetch(MODELS_DEV_API, {
+                headers: { "Accept": "application/json" },
+                signal: AbortSignal.timeout(5000), // 5s timeout
+            });
+            if (!response.ok) {
+                throw new Error(`HTTP ${response.status}`);
+            }
+            const data = await response.json();
+            // Search for model in all providers
+            for (const provider of Object.values(data)) {
+                if (!provider.models)
+                    continue;
+                for (const [id, model] of Object.entries(provider.models)) {
+                    // Match by exact ID or simplified name
+                    const matches = id === modelId ||
+                        id.endsWith(`/${modelId}`) ||
+                        model.name?.toLowerCase().includes(modelId.toLowerCase());
+                    if (matches && model.cost) {
+                        const pricing = {
+                            input: model.cost.input || 0,
+                            output: model.cost.output || 0,
+                        };
+                        // Cache the result
+                        this.pricingCache.set(modelId, { pricing, timestamp: Date.now() });
+                        logger.debug("Fetched pricing from models.dev", { modelId, pricing });
+                        return pricing;
+                    }
+                }
+            }
+            throw new Error("Model not found in models.dev");
+        }
+        catch (error) {
+            logger.warn("Failed to fetch pricing from models.dev, using fallback", {
+                modelId,
+                error: error instanceof Error ? error.message : String(error),
+            });
+            // Use fallback pricing
+            const fallback = FALLBACK_PRICING[modelId] || FALLBACK_PRICING["gpt-4"];
+            // Cache fallback with shorter TTL
+            this.pricingCache.set(modelId, {
+                pricing: fallback,
+                timestamp: Date.now() - (PRICING_CACHE_TTL_MS * 0.9) // Expire sooner
+            });
+            return fallback;
+        }
+    }
+    /**
+     * Calculate cost savings for a given model.
+     *
+     * @param tokensSaved - Number of tokens saved
+     * @param modelId - Model identifier for pricing lookup
+     * @returns Cost savings in USD
+     */
+    async calculateCostSavings(tokensSaved, modelId = "gpt-4") {
+        const pricing = await this.getModelPricing(modelId);
+        // Use input pricing for savings (conservative estimate)
+        return (tokensSaved / 1_000_000) * pricing.input;
+    }
+    // ── Recording API ──────────────────────────────────────────────
+    /**
+     * Record tokens saved by session file cache.
+     * Called from SessionFileCache.check() when returning unchanged/changed chunks.
+     */
+    recordSessionCacheSavings(tokensSaved) {
+        this.counters.sessionCache += tokensSaved;
+        logger.debug("Token metrics: session cache savings", { tokensSaved });
+    }
+    /**
+     * Record tokens saved by redundancy filter.
+     * Called from RedundancyFilter.merge() after deduplicating memories.
+     *
+     * @param removedContent - Content of the removed duplicate memory
+     */
+    recordRedundancyFilterSavings(removedContent) {
+        const tokensSaved = estimateTokens(removedContent, "text");
+        this.counters.redundancyFilter += tokensSaved;
+        logger.debug("Token metrics: redundancy filter savings", { tokensSaved });
+    }
+    /**
+     * Record tokens saved by context compression.
+     * Called from ContextController after compression.
+     */
+    recordCompressionSavings(originalTokens, compressedTokens) {
+        const tokensSaved = Math.max(0, originalTokens - compressedTokens);
+        this.counters.compression += tokensSaved;
+        logger.debug("Token metrics: compression savings", { tokensSaved });
+    }
+    /**
+     * Record a complete context request with all metrics.
+     * Called from ContextController.getOptimizedContext().
+     */
+    recordContextRequest(tokensProcessed, tokensDelivered, sessionCacheSavings = 0, compressionSavings = 0) {
+        this.totalTokensProcessed += tokensProcessed;
+        this.totalTokensDelivered += tokensDelivered;
+        this.requestsServed++;
+        // Update individual counters
+        if (sessionCacheSavings > 0) {
+            this.counters.sessionCache += sessionCacheSavings;
+        }
+        if (compressionSavings > 0) {
+            this.counters.compression += compressionSavings;
+        }
+        // Add to time-series history
+        const totalSavings = this.counters.sessionCache +
+            this.counters.redundancyFilter +
+            this.counters.compression;
+        const compressionRatio = this.totalTokensProcessed > 0
+            ? totalSavings / this.totalTokensProcessed
+            : 0;
+        this.history.push({
+            timestamp: Date.now(),
+            savings: {
+                sessionCache: this.counters.sessionCache,
+                redundancyFilter: this.counters.redundancyFilter,
+                compression: this.counters.compression,
+                total: totalSavings,
+            },
+            compressionRatio,
+        });
+        // Trim history if needed
+        if (this.history.length > MAX_HISTORY_ENTRIES) {
+            this.history.shift();
+        }
+        logger.metric("token_savings_total", totalSavings);
+        logger.metric("compression_ratio", compressionRatio);
+    }
+    // ── Query API ──────────────────────────────────────────────────
+    /**
+     * Get current token metrics statistics.
+     *
+     * @param modelId - Model identifier for cost estimation (defaults to "gpt-4")
+     */
+    async getStats(modelId = "gpt-4") {
+        const totalSavings = this.counters.sessionCache +
+            this.counters.redundancyFilter +
+            this.counters.compression;
+        const overallCompressionRatio = this.totalTokensProcessed > 0
+            ? totalSavings / this.totalTokensProcessed
+            : 0;
+        const estimatedCostSavings = await this.calculateCostSavings(totalSavings, modelId);
+        return {
+            savings: {
+                sessionCache: this.counters.sessionCache,
+                redundancyFilter: this.counters.redundancyFilter,
+                compression: this.counters.compression,
+                total: totalSavings,
+            },
+            totalTokensProcessed: this.totalTokensProcessed,
+            totalTokensDelivered: this.totalTokensDelivered,
+            overallCompressionRatio,
+            requestsServed: this.requestsServed,
+            estimatedCostSavings,
+            model: modelId,
+        };
+    }
+    /**
+     * Get time-series history of token savings.
+     * Useful for charting trends over time.
+     */
+    getHistory() {
+        return [...this.history];
+    }
+    /**
+     * Get the most recent N history entries.
+     */
+    getRecentHistory(limit = 100) {
+        return this.history.slice(-limit);
+    }
+    /**
+     * Reset all counters (useful for testing).
+     */
+    reset() {
+        this.counters.sessionCache = 0;
+        this.counters.redundancyFilter = 0;
+        this.counters.compression = 0;
+        this.totalTokensProcessed = 0;
+        this.totalTokensDelivered = 0;
+        this.requestsServed = 0;
+        this.history.length = 0;
+        logger.info("Token metrics reset");
+    }
+    /**
+     * Get a human-readable summary of token savings.
+     *
+     * @param modelId - Model identifier for cost estimation (defaults to "gpt-4")
+     */
+    async getSummary(modelId = "gpt-4") {
+        const stats = await this.getStats(modelId);
+        const pct = (n, d) => d === 0 ? 0 : (n / d) * 100;
+        const lines = [
+            "Token Savings Summary",
+            "━".repeat(50),
+            `Total Saved: ${stats.savings.total.toLocaleString()} tokens`,
+            `  └─ Session Cache: ${stats.savings.sessionCache.toLocaleString()} (${pct(stats.savings.sessionCache, stats.savings.total).toFixed(1)}%)`,
+            `  └─ Redundancy Filter: ${stats.savings.redundancyFilter.toLocaleString()} (${pct(stats.savings.redundancyFilter, stats.savings.total).toFixed(1)}%)`,
+            `  └─ Compression: ${stats.savings.compression.toLocaleString()} (${pct(stats.savings.compression, stats.savings.total).toFixed(1)}%)`,
+            "",
+            `Requests Served: ${stats.requestsServed}`,
+            `Overall Compression: ${(stats.overallCompressionRatio * 100).toFixed(1)}%`,
+            `Estimated Cost Savings: $${stats.estimatedCostSavings.toFixed(2)} (${stats.model} pricing)`,
+        ];
+        return lines.join("\n");
+    }
+}
+//# sourceMappingURL=token-metrics.js.map

package/dist/services/metrics/token-metrics.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"token-metrics.js","sourceRoot":"","sources":["../../../src/services/metrics/token-metrics.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AA2C1D,oEAAoE;AAEpE,8BAA8B;AAC9B,MAAM,cAAc,GAAG,6BAA6B,CAAC;AAErD,0DAA0D;AAC1D,MAAM,oBAAoB,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAE5C,wFAAwF;AACxF,MAAM,gBAAgB,GAAiC;IACrD,OAAO,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IAClC,aAAa,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IACxC,QAAQ,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,EAAE,EAAE;IACpC,aAAa,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE;IAC3C,eAAe,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IAC5C,eAAe,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IAC1C,wBAAwB,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IACnD,wBAAwB,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;IACnD,iBAAiB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;IAC3C,gBAAgB,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC/C,mBAAmB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;IAC7C,mBAAmB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;IAC7C,kBAAkB,EAAE,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE;IAC7C,gBAAgB,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE;IAC5C,kBAAkB,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE;IACjD,sBAAsB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE;CAChD,CAAC;AAEF,kDAAkD;AAClD,MAAM,mBAAmB,GAAG,IAAI,CAAC;AAEjC,oEAAoE;AAEpE;;GAEG;AACH,MAAM,OAAO,YAAY;IACf,MAAM,CAAC,QAAQ,GAAwB,IAAI,CAAC;IAEnC,QAAQ,GAAG;QAC1B,YAAY,EAAE,CAAC;QACf,gBAAgB,EAAE,CAAC;QACnB,WAAW,EAAE,CAAC;KACf,CAAC;IAEM,oBAAoB,GAAG,CAAC,CAAC;IACzB,oBAAoB,GAAG,CAAC,CAAC;IACzB,cAAc,GAAG,CAAC,CAAC;IAE3B,8CAA8C;IAC7B,OAAO,GAA6B,EAAE,CAAC;IAExD,8CAA8C;IAC7B,YAAY,GAAG,IAAI,GAAG,EAAwD,CAAC;IAEhG,gBAAuB,CAAC;IAExB,MAAM,CAAC,WAAW;QAChB,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE,CAAC;YAC3B,YAAY,CAAC,QAAQ,GAAG,IAAI,YAAY,EAAE,CAAC;QAC7C,CAAC;QACD,OAAO,YAAY,CAAC,QAAQ,CAAC;IAC/B,CAAC;IAED,kEAAkE;IAElE;;;;;OAKG;IACH,KAAK,CAAC,eAAe,CAAC,OAAe;QACnC,oBAAoB;QACpB,MAAM,MAAM,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,MAAM,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,MAAM,CAAC,SAAS,GAAG,oBAAoB,EAAE,CAAC;YACnE,OAAO,MAAM,CAAC,OAAO,CAAC;QACxB,CAAC;QAED,IAAI,CAAC;YACH,4BAA4B;YAC5B,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,cAAc,EAAE;gBAC3C,OAAO,EAAE,EAAE,QAAQ,EAAE,kBAAkB,EAAE;gBACzC,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,aAAa;aACjD,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,IAAI,KAAK,CAAC,QAAQ,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;YAC7C,CAAC;YAED,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;YAEnC,oCAAoC;YACpC,KAAK,MAAM,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,IAA2B,CAAC,EAAE,CAAC;gBAClE,IAAI,CAAC,QAAQ,CAAC,MAAM;oBAAE,SAAS;gBAE/B,KAAK,MAAM,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,MAA6B,CAAC,EAAE,CAAC;oBACjF,uCAAuC;oBACvC,MAAM,OAAO,GACX,EAAE,KAAK,OAAO;wBACd,EAAE,CAAC,QAAQ,CAAC,IAAI,OAAO,EAAE,CAAC;wBAC1B,KAAK,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC;oBAE5D,IAAI,OAAO,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;wBAC1B,MAAM,OAAO,GAAiB;4BAC5B,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC;4BAC5B,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC;yBAC/B,CAAC;wBAEF,mBAAmB;wBACnB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,EAAE,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;wBAEnE,MAAM,CAAC,KAAK,CAAC,iCAAiC,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC;wBACtE,OAAO,OAAO,CAAC;oBACjB,CAAC;gBACH,CAAC;YACH,CAAC;YAED,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAC;QACnD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,CAAC,IAAI,CAAC,yDAAyD,EAAE;gBACrE,OAAO;gBACP,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;aAC9D,CAAC,CAAC;YAEH,uBAAuB;YACvB,MAAM,QAAQ,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;YAExE,kCAAkC;YAClC,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,EAAE;gBAC7B,OAAO,EAAE,QAAQ;gBACjB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,oBAAoB,GAAG,GAAG,CAAC,CAAC,gBAAgB;aACtE,CAAC,CAAC;YAEH,OAAO,QAAQ,CAAC;QAClB,CAAC;IACH,CAAC;IAED;;;;;;OAMG;IACH,KAAK,CAAC,oBAAoB,CAAC,WAAmB,EAAE,UAAkB,OAAO;QACvE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC;QACpD,wDAAwD;QACxD,OAAO,CAAC,WAAW,GAAG,SAAS,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC;IACnD,CAAC;IAED,kEAAkE;IAElE;;;OAGG;IACH,yBAAyB,CAAC,WAAmB;QAC3C,IAAI,CAAC,QAAQ,CAAC,YAAY,IAAI,WAAW,CAAC;QAC1C,MAAM,CAAC,KAAK,CAAC,sCAAsC,EAAE,EAAE,WAAW,EAAE,CAAC,CAAC;IACxE,CAAC;IAED;;;;;OAKG;IACH,6BAA6B,CAAC,cAAsB;QAClD,MAAM,WAAW,GAAG,cAAc,CAAC,cAAc,EAAE,MAAM,CAAC,CAAC;QAC3D,IAAI,CAAC,QAAQ,CAAC,gBAAgB,IAAI,WAAW,CAAC;QAC9C,MAAM,CAAC,KAAK,CAAC,0CAA0C,EAAE,EAAE,WAAW,EAAE,CAAC,CAAC;IAC5E,CAAC;IAED;;;OAGG;IACH,wBAAwB,CAAC,cAAsB,EAAE,gBAAwB;QACvE,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,cAAc,GAAG,gBAAgB,CAAC,CAAC;QACnE,IAAI,CAAC,QAAQ,CAAC,WAAW,IAAI,WAAW,CAAC;QACzC,MAAM,CAAC,KAAK,CAAC,oCAAoC,EAAE,EAAE,WAAW,EAAE,CAAC,CAAC;IACtE,CAAC;IAED;;;OAGG;IACH,oBAAoB,CAClB,eAAuB,EACvB,eAAuB,EACvB,sBAA8B,CAAC,EAC/B,qBAA6B,CAAC;QAE9B,IAAI,CAAC,oBAAoB,IAAI,eAAe,CAAC;QAC7C,IAAI,CAAC,oBAAoB,IAAI,eAAe,CAAC;QAC7C,IAAI,CAAC,cAAc,EAAE,CAAC;QAEtB,6BAA6B;QAC7B,IAAI,mBAAmB,GAAG,CAAC,EAAE,CAAC;YAC5B,IAAI,CAAC,QAAQ,CAAC,YAAY,IAAI,mBAAmB,CAAC;QACpD,CAAC;QACD,IAAI,kBAAkB,GAAG,CAAC,EAAE,CAAC;YAC3B,IAAI,CAAC,QAAQ,CAAC,WAAW,IAAI,kBAAkB,CAAC;QAClD,CAAC;QAED,6BAA6B;QAC7B,MAAM,YAAY,GAChB,IAAI,CAAC,QAAQ,CAAC,YAAY;YAC1B,IAAI,CAAC,QAAQ,CAAC,gBAAgB;YAC9B,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC;QAE5B,MAAM,gBAAgB,GACpB,IAAI,CAAC,oBAAoB,GAAG,CAAC;YAC3B,CAAC,CAAC,YAAY,GAAG,IAAI,CAAC,oBAAoB;YAC1C,CAAC,CAAC,CAAC,CAAC;QAER,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC;YAChB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,OAAO,EAAE;gBACP,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,YAAY;gBACxC,gBAAgB,EAAE,IAAI,CAAC,QAAQ,CAAC,gBAAgB;gBAChD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,WAAW;gBACtC,KAAK,EAAE,YAAY;aACpB;YACD,gBAAgB;SACjB,CAAC,CAAC;QAEH,yBAAyB;QACzB,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,mBAAmB,EAAE,CAAC;YAC9C,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QACvB,CAAC;QAED,MAAM,CAAC,MAAM,CAAC,qBAAqB,EAAE,YAAY,CAAC,CAAC;QACnD,MAAM,CAAC,MAAM,CAAC,mBAAmB,EAAE,gBAAgB,CAAC,CAAC;IACvD,CAAC;IAED,kEAAkE;IAElE;;;;OAIG;IACH,KAAK,CAAC,QAAQ,CAAC,UAAkB,OAAO;QACtC,MAAM,YAAY,GAChB,IAAI,CAAC,QAAQ,CAAC,YAAY;YAC1B,IAAI,CAAC,QAAQ,CAAC,gBAAgB;YAC9B,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC;QAE5B,MAAM,uBAAuB,GAC3B,IAAI,CAAC,oBAAoB,GAAG,CAAC;YAC3B,CAAC,CAAC,YAAY,GAAG,IAAI,CAAC,oBAAoB;YAC1C,CAAC,CAAC,CAAC,CAAC;QAER,MAAM,oBAAoB,GAAG,MAAM,IAAI,CAAC,oBAAoB,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;QAEpF,OAAO;YACL,OAAO,EAAE;gBACP,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,YAAY;gBACxC,gBAAgB,EAAE,IAAI,CAAC,QAAQ,CAAC,gBAAgB;gBAChD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,WAAW;gBACtC,KAAK,EAAE,YAAY;aACpB;YACD,oBAAoB,EAAE,IAAI,CAAC,oBAAoB;YAC/C,oBAAoB,EAAE,IAAI,CAAC,oBAAoB;YAC/C,uBAAuB;YACvB,cAAc,EAAE,IAAI,CAAC,cAAc;YACnC,oBAAoB;YACpB,KAAK,EAAE,OAAO;SACf,CAAC;IACJ,CAAC;IAED;;;OAGG;IACH,UAAU;QACR,OAAO,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;IAC3B,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,QAAgB,GAAG;QAClC,OAAO,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;IAED;;OAEG;IACH,KAAK;QACH,IAAI,CAAC,QAAQ,CAAC,YAAY,GAAG,CAAC,CAAC;QAC/B,IAAI,CAAC,QAAQ,CAAC,gBAAgB,GAAG,CAAC,CAAC;QACnC,IAAI,CAAC,QAAQ,CAAC,WAAW,GAAG,CAAC,CAAC;QAC9B,IAAI,CAAC,oBAAoB,GAAG,CAAC,CAAC;QAC9B,IAAI,CAAC,oBAAoB,GAAG,CAAC,CAAC;QAC9B,IAAI,CAAC,cAAc,GAAG,CAAC,CAAC;QACxB,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC;QACxB,MAAM,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IACrC,CAAC;IAED;;;;OAIG;IACH,KAAK,CAAC,UAAU,CAAC,UAAkB,OAAO;QACxC,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAC3C,MAAM,GAAG,GAAG,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC;QAClE,MAAM,KAAK,GAAa;YACtB,uBAAuB;YACvB,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC;YACd,gBAAgB,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,cAAc,EAAE,SAAS;YAC7D,uBAAuB,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,cAAc,EAAE,KAAK,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,YAAY,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;YAC1I,2BAA2B,KAAK,CAAC,OAAO,CAAC,gBAAgB,CAAC,cAAc,EAAE,KAAK,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;YACtJ,qBAAqB,KAAK,CAAC,OAAO,CAAC,WAAW,CAAC,cAAc,EAAE,KAAK,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI;YACtI,EAAE;YACF,oBAAoB,KAAK,CAAC,cAAc,EAAE;YAC1C,wBAAwB,CAAC,KAAK,CAAC,uBAAuB,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC3E,4BAA4B,KAAK,CAAC,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,KAAK,WAAW;SAC7F,CAAC;QAEF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC"}

package/dist/services/monitoring/metrics.d.ts

@@ -0,0 +1,154 @@
+/**
+ * Monitoring and Metrics System for massa-ai MCP
+ *
+ * Tracks:
+ * - Embedding usage (calls, tokens, cache hits)
+ * - API costs per provider
+ * - Performance metrics (latency, throughput)
+ * - Error rates
+ * - Cache efficiency
+ */
+/**
+ * Pricing per provider (per 1M tokens)
+ */
+export declare const PROVIDER_PRICING: {
+    mistralText: {
+        model: string;
+        costPer1MTokens: number;
+        dimensions: number;
+    };
+    mistralCode: {
+        model: string;
+        costPer1MTokens: number;
+        dimensions: number;
+    };
+    mistralLLM: {
+        model: string;
+        inputCostPer1MTokens: number;
+        outputCostPer1MTokens: number;
+    };
+};
+/**
+ * Metrics data structure
+ */
+export interface Metrics {
+    embeddings: {
+        totalCalls: number;
+        totalTokens: number;
+        totalCost: number;
+        byProvider: Record<string, {
+            calls: number;
+            tokens: number;
+            cost: number;
+            avgLatency: number;
+            errors: number;
+        }>;
+    };
+    context: {
+        totalRequests: number;
+        totalInputTokens: number;
+        totalOptimizedTokens: number;
+        totalSentToLLM: number;
+        avgReductionRatio: number;
+        totalCostSaved: number;
+        avgLatency: number;
+    };
+    cache: {
+        hits: number;
+        misses: number;
+        hitRate: number;
+        savedCost: number;
+        avgHitLatency: number;
+        avgMissLatency: number;
+    };
+    compression: {
+        totalCalls: number;
+        totalInputTokens: number;
+        totalOutputTokens: number;
+        totalCost: number;
+        avgCompressionRatio: number;
+        avgLatency: number;
+    };
+    performance: {
+        avgEmbeddingLatency: number;
+        p95EmbeddingLatency: number;
+        p99EmbeddingLatency: number;
+        totalRequests: number;
+        errorRate: number;
+    };
+    period: {
+        start: string;
+        end: string;
+        durationHours: number;
+    };
+}
+/**
+ * Metrics collector singleton
+ */
+export declare class MetricsCollector {
+    private static instance;
+    private metricsPath;
+    private currentMetrics;
+    private latencies;
+    private constructor();
+    static getInstance(): MetricsCollector;
+    /**
+     * Load metrics from disk or initialize new
+     */
+    private loadMetrics;
+    /**
+     * Initialize empty metrics
+     */
+    private initializeMetrics;
+    /**
+     * Record embedding call
+     */
+    recordEmbedding(params: {
+        provider: string;
+        tokens: number;
+        latency: number;
+        cached: boolean;
+        error?: boolean;
+    }): void;
+    /**
+     * Record LLM compression call
+     */
+    recordCompression(params: {
+        inputTokens: number;
+        outputTokens: number;
+        latency: number;
+    }): void;
+    /**
+     * Record context optimization (CORE FEATURE)
+     * Tracks: input tokens → optimized tokens → sent to LLM
+     */
+    recordContextOptimization(params: {
+        inputTokens: number;
+        optimizedTokens: number;
+        sentToLLM: number;
+        latency: number;
+        costSaved?: number;
+    }): void;
+    /**
+     * Update performance metrics (percentiles)
+     */
+    private updatePerformanceMetrics;
+    /**
+     * Get current metrics
+     */
+    getMetrics(): Metrics;
+    /**
+     * Save metrics to disk
+     */
+    save(): void;
+    /**
+     * Reset metrics (start fresh period)
+     */
+    reset(): void;
+    /**
+     * Get formatted summary
+     */
+    getSummary(): string;
+}
+export declare const metrics: MetricsCollector;
+//# sourceMappingURL=metrics.d.ts.map

package/dist/services/monitoring/metrics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics.d.ts","sourceRoot":"","sources":["../../../src/services/monitoring/metrics.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAMH;;GAEG;AACH,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;CAgB5B,CAAC;AAEF;;GAEG;AACH,MAAM,WAAW,OAAO;IAEtB,UAAU,EAAE;QACV,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,CAAC;QACpB,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAChB,MAAM,EACN;YACE,KAAK,EAAE,MAAM,CAAC;YACd,MAAM,EAAE,MAAM,CAAC;YACf,IAAI,EAAE,MAAM,CAAC;YACb,UAAU,EAAE,MAAM,CAAC;YACnB,MAAM,EAAE,MAAM,CAAC;SAChB,CACF,CAAC;KACH,CAAC;IAGF,OAAO,EAAE;QACP,aAAa,EAAE,MAAM,CAAC;QACtB,gBAAgB,EAAE,MAAM,CAAC;QACzB,oBAAoB,EAAE,MAAM,CAAC;QAC7B,cAAc,EAAE,MAAM,CAAC;QACvB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,cAAc,EAAE,MAAM,CAAC;QACvB,UAAU,EAAE,MAAM,CAAC;KACpB,CAAC;IAGF,KAAK,EAAE;QACL,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,EAAE,MAAM,CAAC;QAClB,aAAa,EAAE,MAAM,CAAC;QACtB,cAAc,EAAE,MAAM,CAAC;KACxB,CAAC;IAGF,WAAW,EAAE;QACX,UAAU,EAAE,MAAM,CAAC;QACnB,gBAAgB,EAAE,MAAM,CAAC;QACzB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,SAAS,EAAE,MAAM,CAAC;QAClB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,UAAU,EAAE,MAAM,CAAC;KACpB,CAAC;IAGF,WAAW,EAAE;QACX,mBAAmB,EAAE,MAAM,CAAC;QAC5B,mBAAmB,EAAE,MAAM,CAAC;QAC5B,mBAAmB,EAAE,MAAM,CAAC;QAC5B,aAAa,EAAE,MAAM,CAAC;QACtB,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;IAGF,MAAM,EAAE;QACN,KAAK,EAAE,MAAM,CAAC;QACd,GAAG,EAAE,MAAM,CAAC;QACZ,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC;CACH;AAED;;GAEG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAmB;IAC1C,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,cAAc,CAAU;IAChC,OAAO,CAAC,SAAS,CAAgB;IAEjC,OAAO;IAgBP,MAAM,CAAC,WAAW,IAAI,gBAAgB;IAOtC;;OAEG;IACH,OAAO,CAAC,WAAW;IAanB;;OAEG;IACH,OAAO,CAAC,iBAAiB;IAgDzB;;OAEG;IACH,eAAe,CAAC,MAAM,EAAE;QACtB,QAAQ,EAAE,MAAM,CAAC;QACjB,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,OAAO,CAAC;QAChB,KAAK,CAAC,EAAE,OAAO,CAAC;KACjB;IAwGD;;OAEG;IACH,iBAAiB,CAAC,MAAM,EAAE;QACxB,WAAW,EAAE,MAAM,CAAC;QACpB,YAAY,EAAE,MAAM,CAAC;QACrB,OAAO,EAAE,MAAM,CAAC;KACjB;IAgCD;;;OAGG;IACH,yBAAyB,CAAC,MAAM,EAAE;QAChC,WAAW,EAAE,MAAM,CAAC;QACpB,eAAe,EAAE,MAAM,CAAC;QACxB,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB;IA+BD;;OAEG;IACH,OAAO,CAAC,wBAAwB;IA0BhC;;OAEG;IACH,UAAU,IAAI,OAAO;IAIrB;;OAEG;IACH,IAAI;IAWJ;;OAEG;IACH,KAAK;IAML;;OAEG;IACH,UAAU,IAAI,MAAM;CAkErB;AAGD,eAAO,MAAM,OAAO,kBAAiC,CAAC"}