@runnerpro/backend 1.21.8 → 1.21.9

package/lib/cjs/prompt/ai.js

@@ -16,6 +16,7 @@ const modelPricing_1 = require("./modelPricing");
 const constants_1 = require("./constants");
 const googleModel_1 = require("./googleModel");
 const slack_1 = require("../slack");
+const llmCacheStore_1 = require("./llmCacheStore");
 const MIN_RETRIES = 1;
 // ✅ Detección de Chain-of-Thought (CoT) que se cuela en la respuesta final del modelo
 const COT_HTML_TAG = /<\/?(thinking|thought|reasoning)\b/i;
@@ -204,6 +205,17 @@ function generateObject(options) {
     return __awaiter(this, void 0, void 0, function* () {
         let lastError;
         let currentOptions = options;
+        // [llm-cache] cache-aside: HIT → output guardado sin llamar al modelo (opt-out con options.cache === false)
+        const co = currentOptions;
+        const useCache = co.cache !== false;
+        const schemaSig = useCache ? (0, llmCacheStore_1.schemaSignature)(co.schema) : null;
+        let cacheKey = null;
+        if (useCache) {
+            cacheKey = (0, llmCacheStore_1.llmCacheKey)({ system: co.system, prompt: co.prompt, schemaSig, temperature: co.temperature });
+            const hit = yield (0, llmCacheStore_1.lookup)(cacheKey);
+            if (hit !== null && hit !== undefined)
+                return { object: hit, cost: { inputTokens: 0, outputTokens: 0, cost: 0 } };
+        }
         const fallbackNames = ((_a = currentOptions.model) === null || _a === void 0 ? void 0 : _a._fallbackModelNames) || [];
         const maxAttempts = (1 + fallbackNames.length) * (1 + MIN_RETRIES);
         let transientRetries = 0;
@@ -216,6 +228,8 @@ function generateObject(options) {
                 const tracker = costTrackingStorage.getStore();
                 if (tracker)
                     tracker.push(cost);
+                if (useCache && cacheKey)
+                    (0, llmCacheStore_1.store)(cacheKey, { modelName, system: co.system, prompt: co.prompt, schemaSig, temperature: co.temperature, output: object });
                 return { object, cost };
             }
             catch (error) {

package/lib/cjs/prompt/llmCacheStore.js

@@ -0,0 +1,102 @@
+"use strict";
+var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
+    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
+    return new (P || (P = Promise))(function (resolve, reject) {
+        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
+        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
+        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
+        step((generator = generator.apply(thisArg, _arguments || [])).next());
+    });
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.store = exports.lookup = exports.llmCacheKey = exports.schemaSignature = void 0;
+/**
+ * Caché LLM (cache-aside) para `generateObject`.
+ *
+ * Antes de cada llamada se busca por (modelo + system + input + schema + temperatura); si existe se
+ * devuelve el output guardado SIN llamar al modelo; si no, se llama y se guarda. Una poda periódica
+ * (Cron-backend /cron/operations/llmCacheEvict) elimina las entradas de un solo uso rancias.
+ *
+ * Fail-safe: un único chequeo `to_regclass` desactiva la caché en silencio si la tabla "LLM CACHE" no
+ * existe (migración no aplicada) o la BBDD falla. Así nunca rompe ni hace spam de errores.
+ *
+ * Tabla: ver migración en IA-backend docs/llm-cache-prod/001-llm-cache-table.sql.
+ */
+const node_crypto_1 = require("node:crypto");
+const db_1 = require("../db");
+let _ready = null; // null=sin comprobar, true=tabla disponible, false=desactivada
+function ensureReady() {
+    return __awaiter(this, void 0, void 0, function* () {
+        if (_ready !== null)
+            return _ready;
+        try {
+            const rows = yield (0, db_1.query)('SELECT to_regclass(?) AS t', ['"LLM CACHE"']);
+            _ready = !!(rows && rows[0] && rows[0].t);
+        }
+        catch (_a) {
+            _ready = false;
+        }
+        return _ready;
+    });
+}
+/** Firma ligera del schema (zod) para distinguir llamadas con el mismo prompt pero distinta forma. */
+function schemaSignature(schema) {
+    var _a, _b;
+    try {
+        if (!schema)
+            return 'noschema';
+        const tn = ((_a = schema === null || schema === void 0 ? void 0 : schema._def) === null || _a === void 0 ? void 0 : _a.typeName) || ((_b = schema === null || schema === void 0 ? void 0 : schema.constructor) === null || _b === void 0 ? void 0 : _b.name) || 'schema';
+        let shape = schema === null || schema === void 0 ? void 0 : schema.shape;
+        if (!shape && (schema === null || schema === void 0 ? void 0 : schema._def))
+            shape = typeof schema._def.shape === 'function' ? schema._def.shape() : schema._def.shape;
+        const keys = shape ? Object.keys(shape).sort().join(',') : '';
+        return `${tn}:${keys}`;
+    }
+    catch (_c) {
+        return 'schema';
+    }
+}
+exports.schemaSignature = schemaSignature;
+/**
+ * Clave determinista de una llamada: SYSTEM + INPUT + SCHEMA + TEMPERATURE.
+ * NO incluye el modelo a propósito: si esos 4 coinciden, se reutiliza el output guardado (lo haya
+ * generado el modelo que sea). Implica que cambiar de modelo NO invalida la caché.
+ */
+function llmCacheKey(opts) {
+    var _a;
+    const payload = JSON.stringify({
+        system: opts.system || '',
+        prompt: opts.prompt || '',
+        schema: opts.schemaSig || 'noschema',
+        temperature: (_a = opts.temperature) !== null && _a !== void 0 ? _a : null,
+    });
+    return (0, node_crypto_1.createHash)('sha256').update(payload).digest('hex');
+}
+exports.llmCacheKey = llmCacheKey;
+/** HIT: incrementa USE COUNT + LAST USED y devuelve OUTPUT (atómico). null en MISS o ante error. */
+function lookup(key) {
+    return __awaiter(this, void 0, void 0, function* () {
+        if (!(yield ensureReady()))
+            return null;
+        try {
+            const rows = yield (0, db_1.query)('UPDATE "LLM CACHE" SET "USE COUNT" = "USE COUNT" + 1, "LAST USED AT" = now() WHERE "KEY" = ? RETURNING "OUTPUT"', [key]);
+            return rows && rows[0] ? rows[0].output : null;
+        }
+        catch (_a) {
+            return null;
+        }
+    });
+}
+exports.lookup = lookup;
+/** MISS: inserta (ON CONFLICT cubre la carrera). Fire-and-forget, nunca lanza. */
+function store(key, data) {
+    void ensureReady()
+        .then((ok) => {
+        var _a, _b;
+        if (!ok)
+            return undefined;
+        return (0, db_1.query)('INSERT INTO "LLM CACHE" ("KEY","MODEL","SYSTEM PROMPT","INPUT PROMPT","SCHEMA SIG","TEMPERATURE","OUTPUT") VALUES (?,?,?,?,?,?,?::jsonb) ON CONFLICT ("KEY") DO UPDATE SET "USE COUNT" = "LLM CACHE"."USE COUNT" + 1, "LAST USED AT" = now()', [key, data.modelName || null, data.system || null, data.prompt || null, data.schemaSig || null, (_a = data.temperature) !== null && _a !== void 0 ? _a : null, JSON.stringify((_b = data.output) !== null && _b !== void 0 ? _b : null)]);
+    })
+        .catch(() => { });
+}
+exports.store = store;

package/lib/cjs/types/prompt/ai.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ai.d.ts","sourceRoot":"","sources":["../../../../src/prompt/ai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,IAAI,sBAAsB,EAAE,YAAY,IAAI,oBAAoB,EAAE,MAAM,IAAI,CAAC;AAEpG,OAAO,EAAiB,KAAK,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAuKhE;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,iBAAe,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,OAAO,sBAAsB,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,GAAG,CAAC;IAAC,IAAI,EAAE,UAAU,CAAA;CAAE,CAAC,CAyC/H;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,iBAAe,YAAY,CAAC,OAAO,EAAE,UAAU,CAAC,OAAO,oBAAoB,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,UAAU,CAAA;CAAE,CAAC,CA0C5H;AA2ED;;;;;;;;;;;;;;;;;GAiBG;AACH,iBAAe,mBAAmB,CAAC,CAAC,EAAE,EAAE,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,CAAC,CAAC;IAAC,SAAS,EAAE,UAAU,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAAC,CAc5H;AAED,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,mBAAmB,EAAE,CAAC"}
+{"version":3,"file":"ai.d.ts","sourceRoot":"","sources":["../../../../src/prompt/ai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,IAAI,sBAAsB,EAAE,YAAY,IAAI,oBAAoB,EAAE,MAAM,IAAI,CAAC;AAEpG,OAAO,EAAiB,KAAK,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAwKhE;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,iBAAe,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,OAAO,sBAAsB,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,GAAG,CAAC;IAAC,IAAI,EAAE,UAAU,CAAA;CAAE,CAAC,CAoD/H;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,iBAAe,YAAY,CAAC,OAAO,EAAE,UAAU,CAAC,OAAO,oBAAoB,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,UAAU,CAAA;CAAE,CAAC,CA0C5H;AA2ED;;;;;;;;;;;;;;;;;GAiBG;AACH,iBAAe,mBAAmB,CAAC,CAAC,EAAE,EAAE,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,CAAC,CAAC;IAAC,SAAS,EAAE,UAAU,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAAC,CAc5H;AAED,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,mBAAmB,EAAE,CAAC"}

package/lib/cjs/types/prompt/llmCacheStore.d.ts

@@ -0,0 +1,26 @@
+/** Firma ligera del schema (zod) para distinguir llamadas con el mismo prompt pero distinta forma. */
+declare function schemaSignature(schema: any): string;
+/**
+ * Clave determinista de una llamada: SYSTEM + INPUT + SCHEMA + TEMPERATURE.
+ * NO incluye el modelo a propósito: si esos 4 coinciden, se reutiliza el output guardado (lo haya
+ * generado el modelo que sea). Implica que cambiar de modelo NO invalida la caché.
+ */
+declare function llmCacheKey(opts: {
+    system?: string;
+    prompt?: string;
+    schemaSig?: string | null;
+    temperature?: number | null;
+}): string;
+/** HIT: incrementa USE COUNT + LAST USED y devuelve OUTPUT (atómico). null en MISS o ante error. */
+declare function lookup(key: string): Promise<any | null>;
+/** MISS: inserta (ON CONFLICT cubre la carrera). Fire-and-forget, nunca lanza. */
+declare function store(key: string, data: {
+    modelName?: string;
+    system?: string;
+    prompt?: string;
+    schemaSig?: string | null;
+    temperature?: number | null;
+    output: any;
+}): void;
+export { schemaSignature, llmCacheKey, lookup, store };
+//# sourceMappingURL=llmCacheStore.d.ts.map

package/lib/cjs/types/prompt/llmCacheStore.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llmCacheStore.d.ts","sourceRoot":"","sources":["../../../../src/prompt/llmCacheStore.ts"],"names":[],"mappings":"AA4BA,sGAAsG;AACtG,iBAAS,eAAe,CAAC,MAAM,EAAE,GAAG,GAAG,MAAM,CAW5C;AAED;;;;GAIG;AACH,iBAAS,WAAW,CAAC,IAAI,EAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,GAAG,MAAM,CAQ/H;AAED,oGAAoG;AACpG,iBAAe,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC,GAAG,GAAG,IAAI,CAAC,CAQtD;AAED,kFAAkF;AAClF,iBAAS,KAAK,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE;IAAE,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,MAAM,EAAE,GAAG,CAAA;CAAE,GAAG,IAAI,CAUrK;AAED,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@runnerpro/backend",
-  "version": "1.21.8",
+  "version": "1.21.9",
   "description": "A collection of common backend functions",
   "exports": {
     ".": "./lib/cjs/index.js"
@@ -63,6 +63,7 @@
   },
   "dependencies": {
     "@ai-sdk/amazon-bedrock": "^4.0.89",
+    "@ai-sdk/azure": "^3.0.74",
     "@ai-sdk/google": "^3.0.43",
     "@ai-sdk/google-vertex": "^4.0.102",
     "@google-cloud/speech": "^7.2.1",