ai-sdk-memory 0.0.1

package/README.md

@@ -0,0 +1 @@
+

package/dist/index.d.mts

@@ -0,0 +1,12 @@
+import { streamText, StreamTextResult, generateText, GenerateTextResult, generateObject, GenerateObjectResult, streamObject, StreamObjectResult } from 'ai';
+import { SemanticCacheConfig } from './schema.mjs';
+import 'zod';
+
+declare function createSemanticCache(config: SemanticCacheConfig): {
+    streamText: <TOOLS extends Record<string, any> = {}>(options: Parameters<typeof streamText<TOOLS>>[0]) => Promise<StreamTextResult<TOOLS, any>>;
+    generateText: <TOOLS extends Record<string, any> = {}, OUTPUT = undefined>(options: Parameters<typeof generateText<TOOLS, OUTPUT>>[0]) => Promise<GenerateTextResult<TOOLS, OUTPUT>>;
+    generateObject: <T = any>(options: Parameters<typeof generateObject>[0]) => Promise<GenerateObjectResult<T>>;
+    streamObject: <T = any>(options: Parameters<typeof streamObject>[0]) => Promise<StreamObjectResult<T, T, any>>;
+};
+
+export { createSemanticCache };

package/dist/index.d.ts

@@ -0,0 +1,12 @@
+import { streamText, StreamTextResult, generateText, GenerateTextResult, generateObject, GenerateObjectResult, streamObject, StreamObjectResult } from 'ai';
+import { SemanticCacheConfig } from './schema.js';
+import 'zod';
+
+declare function createSemanticCache(config: SemanticCacheConfig): {
+    streamText: <TOOLS extends Record<string, any> = {}>(options: Parameters<typeof streamText<TOOLS>>[0]) => Promise<StreamTextResult<TOOLS, any>>;
+    generateText: <TOOLS extends Record<string, any> = {}, OUTPUT = undefined>(options: Parameters<typeof generateText<TOOLS, OUTPUT>>[0]) => Promise<GenerateTextResult<TOOLS, OUTPUT>>;
+    generateObject: <T = any>(options: Parameters<typeof generateObject>[0]) => Promise<GenerateObjectResult<T>>;
+    streamObject: <T = any>(options: Parameters<typeof streamObject>[0]) => Promise<StreamObjectResult<T, T, any>>;
+};
+
+export { createSemanticCache };

package/dist/index.js

@@ -0,0 +1,334 @@
+'use strict';
+
+require('@ai-sdk/provider');
+var ai = require('ai');
+var vector = require('@upstash/vector');
+var redis = require('@upstash/redis');
+var crypto = require('crypto');
+var zod = require('zod');
+require('dotenv/config');
+
+function _interopNamespace(e) {
+  if (e && e.__esModule) return e;
+  var n = Object.create(null);
+  if (e) {
+    Object.keys(e).forEach(function (k) {
+      if (k !== 'default') {
+        var d = Object.getOwnPropertyDescriptor(e, k);
+        Object.defineProperty(n, k, d.get ? d : {
+          enumerable: true,
+          get: function () { return e[k]; }
+        });
+      }
+    });
+  }
+  n.default = e;
+  return Object.freeze(n);
+}
+
+var crypto__namespace = /*#__PURE__*/_interopNamespace(crypto);
+
+// index.ts
+var semanticCacheConfigSchema = zod.z.object({
+  model: zod.z.union([
+    zod.z.string(),
+    zod.z.custom((val) => {
+      return val && typeof val === "object";
+    })
+  ]),
+  vector: zod.z.object({
+    url: zod.z.url(),
+    token: zod.z.string().min(1)
+  }).optional().default({
+    url: process.env.VECTOR_REST_URL ?? "",
+    token: process.env.VECTOR_REST_TOKEN ?? ""
+  }),
+  redis: zod.z.object({
+    url: zod.z.url(),
+    token: zod.z.string().min(1)
+  }).optional().default({
+    url: process.env.REDIS_REST_URL ?? "",
+    token: process.env.REDIS_REST_TOKEN ?? ""
+  }),
+  threshold: zod.z.number().min(0).max(1).optional().default(0.92),
+  ttl: zod.z.number().positive().optional().default(60 * 60 * 24 * 14),
+  debug: zod.z.boolean().optional().default(false),
+  cacheMode: zod.z.enum(["default", "refresh"]).optional().default("default"),
+  simulateStream: zod.z.object({
+    enabled: zod.z.boolean().optional().default(true),
+    initialDelayInMs: zod.z.number().min(0).optional().default(0),
+    chunkDelayInMs: zod.z.number().min(0).optional().default(10)
+  }).optional().default({
+    enabled: true,
+    initialDelayInMs: 0,
+    chunkDelayInMs: 10
+  }),
+  useFullMessages: zod.z.boolean().optional().default(false)
+});
+function validateEnvConfig(config) {
+  const errors = [];
+  if (!config.vector.url) {
+    errors.push(
+      "Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable."
+    );
+  }
+  if (!config.vector.token) {
+    errors.push(
+      "Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable."
+    );
+  }
+  if (!config.redis.url) {
+    errors.push(
+      "Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable."
+    );
+  }
+  if (!config.redis.token) {
+    errors.push(
+      "Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable."
+    );
+  }
+  if (errors.length > 0) {
+    throw new Error(
+      `Semantic Cache Configuration Error:
+${errors.map((e) => `  - ${e}`).join("\n")}`
+    );
+  }
+}
+
+// index.ts
+var norm = (s) => s.trim().toLowerCase().replace(/\s+/g, " ");
+var sha = (s) => crypto__namespace.createHash("sha256").update(s).digest("hex");
+function canonicalizeMessages(msgs) {
+  return msgs.map((m) => ({
+    role: m.role,
+    content: typeof m.content === "string" ? m.content : JSON.stringify(m.content)
+  }));
+}
+function buildScope(options) {
+  return {
+    llmModel: options.model?.modelId ?? String(options.model ?? ""),
+    systemHash: sha(options.system ?? ""),
+    params: sha(
+      JSON.stringify({
+        temperature: options.temperature,
+        topP: options.topP
+      })
+    ),
+    toolsHash: sha(JSON.stringify(options.tools ?? {}))
+  };
+}
+function createSemanticCache(config) {
+  const parsed = semanticCacheConfigSchema.parse(config);
+  validateEnvConfig(parsed);
+  const {
+    vector: vector$1,
+    redis: redisConfig,
+    threshold,
+    ttl,
+    simulateStream,
+    debug,
+    cacheMode,
+    useFullMessages
+  } = parsed;
+  const model = parsed.model;
+  const index = new vector.Index({
+    url: vector$1.url,
+    token: vector$1.token
+  });
+  const redis$1 = new redis.Redis({
+    url: redisConfig.url,
+    token: redisConfig.token
+  });
+  function getCacheKey(options) {
+    if (options.messages) {
+      const messages = Array.isArray(options.messages) ? options.messages : [];
+      if (!useFullMessages && messages.length > 0) {
+        const lastMessage = messages[messages.length - 1];
+        return JSON.stringify({
+          role: lastMessage.role,
+          content: typeof lastMessage.content === "string" ? lastMessage.content : JSON.stringify(lastMessage.content)
+        });
+      }
+      return JSON.stringify(canonicalizeMessages(options.messages));
+    }
+    if (options.prompt) {
+      return String(options.prompt);
+    }
+    return "";
+  }
+  async function checkSemanticCache(cacheInput, scope) {
+    const promptNorm = norm(cacheInput);
+    const { embedding } = await ai.embed({
+      model,
+      value: promptNorm
+    });
+    const result = await index.query({
+      vector: embedding,
+      topK: 3,
+      includeMetadata: true
+    });
+    const hit = result.find((m) => {
+      if (m.score < threshold) return false;
+      const metadata = m.metadata;
+      if (!metadata) return false;
+      return metadata.llmModel === scope.llmModel && metadata.systemHash === scope.systemHash && metadata.params === scope.params && metadata.toolsHash === scope.toolsHash;
+    });
+    if (hit) {
+      const cached = await redis$1.get(hit.id.toString());
+      if (cached) {
+        if (debug) console.log("\u2705 cache hit", hit.score.toFixed(3));
+        return { cached, embedding, promptNorm };
+      }
+    }
+    if (debug) console.log("\u274C miss -> generating\u2026");
+    return { cached: null, embedding, promptNorm };
+  }
+  async function storeInCache(id, data, embedding, promptNorm, scope) {
+    const lockKey = "lock:" + id;
+    const ok = await redis$1.set(lockKey, "1", { nx: true, ex: 15 });
+    if (!ok) {
+      if (debug)
+        console.log("\u26A0\uFE0F Another process is writing to cache, skipping");
+      return;
+    }
+    try {
+      await redis$1.set(id, data, { ex: ttl });
+      await index.upsert([
+        {
+          id,
+          vector: embedding,
+          metadata: {
+            prompt: promptNorm,
+            ...scope
+          }
+        }
+      ]);
+    } finally {
+      await redis$1.del(lockKey);
+    }
+  }
+  const semanticCacheMiddleware = {
+    wrapStream: async ({ doStream, params }) => {
+      const cacheInput = getCacheKey(params);
+      const scope = buildScope(params);
+      const promptScope = Object.values(scope).join("|");
+      const { cached, embedding, promptNorm } = await checkSemanticCache(
+        cacheInput,
+        scope
+      );
+      if (cached && cacheMode !== "refresh") {
+        if (debug) console.log("\u2705 Returning cached stream");
+        let chunks = [];
+        if (cached.streamParts) {
+          chunks = cached.streamParts.map((p) => {
+            if (p.type === "response-metadata" && p.timestamp) {
+              return { ...p, timestamp: new Date(p.timestamp) };
+            }
+            return p;
+          });
+        } else if (cached.text) {
+          chunks = [
+            { type: "text-start", id: cached.id },
+            { type: "text-delta", delta: cached.text, id: cached.id },
+            { type: "finish", finishReason: "stop", usage: cached.usage }
+          ];
+        }
+        return {
+          stream: ai.simulateReadableStream({
+            initialDelayInMs: simulateStream.enabled ? simulateStream.initialDelayInMs : 0,
+            chunkDelayInMs: simulateStream.enabled ? simulateStream.chunkDelayInMs : 0,
+            chunks
+          })
+        };
+      }
+      const { stream, ...rest } = await doStream();
+      const fullResponse = [];
+      const transformStream = new TransformStream({
+        transform(chunk, controller) {
+          fullResponse.push(chunk);
+          controller.enqueue(chunk);
+        },
+        async flush() {
+          const id = "llm:" + sha(promptScope + "|" + promptNorm);
+          await storeInCache(
+            id,
+            { streamParts: fullResponse },
+            embedding,
+            promptNorm,
+            scope
+          );
+        }
+      });
+      return {
+        stream: stream.pipeThrough(transformStream),
+        ...rest
+      };
+    },
+    wrapGenerate: async ({ doGenerate, params }) => {
+      const cacheInput = getCacheKey(params);
+      const scope = buildScope(params);
+      const promptScope = Object.values(scope).join("|");
+      const { cached, embedding, promptNorm } = await checkSemanticCache(
+        cacheInput,
+        scope
+      );
+      if (cached && cacheMode !== "refresh") {
+        if (debug) console.log("\u2705 Returning cached generation");
+        if (cached?.response?.timestamp) {
+          cached.response.timestamp = new Date(cached.response.timestamp);
+        }
+        return cached;
+      }
+      const result = await doGenerate();
+      const id = "llm:" + sha(promptScope + "|" + promptNorm);
+      await storeInCache(id, result, embedding, promptNorm, scope);
+      return result;
+    }
+  };
+  return {
+    streamText: async (options) => {
+      const wrappedModel = ai.wrapLanguageModel({
+        model: typeof options.model === "string" ? ai.gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return ai.streamText({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    generateText: async (options) => {
+      const wrappedModel = ai.wrapLanguageModel({
+        model: typeof options.model === "string" ? ai.gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return ai.generateText({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    generateObject: async (options) => {
+      const wrappedModel = ai.wrapLanguageModel({
+        model: typeof options.model === "string" ? ai.gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return await ai.generateObject({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    streamObject: async (options) => {
+      const wrappedModel = ai.wrapLanguageModel({
+        model: typeof options.model === "string" ? ai.gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return ai.streamObject({
+        ...options,
+        model: wrappedModel
+      });
+    }
+  };
+}
+
+exports.createSemanticCache = createSemanticCache;
+//# sourceMappingURL=index.js.map
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../schema.ts","../index.ts"],"names":["z","crypto","vector","Index","redis","Redis","embed","simulateReadableStream","wrapLanguageModel","gateway","streamText","generateText","generateObject","streamObject"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAGO,IAAM,yBAAA,GAA4BA,MAAE,MAAA,CAAO;AAAA,EAChD,KAAA,EAAOA,MAAE,KAAA,CAAM;AAAA,IACbA,MAAE,MAAA,EAAO;AAAA,IACTA,KAAA,CAAE,MAAA,CAAO,CAAC,GAAA,KAAQ;AAChB,MAAA,OAAO,GAAA,IAAO,OAAO,GAAA,KAAQ,QAAA;AAAA,IAC/B,CAAC;AAAA,GACF,CAAA;AAAA,EACD,MAAA,EAAQA,MACL,MAAA,CAAO;AAAA,IACN,GAAA,EAAKA,MAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAOA,KAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,eAAA,IAAmB,EAAA;AAAA,IACpC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iBAAA,IAAqB;AAAA,GACzC,CAAA;AAAA,EACH,KAAA,EAAOA,MACJ,MAAA,CAAO;AAAA,IACN,GAAA,EAAKA,MAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAOA,KAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,cAAA,IAAkB,EAAA;AAAA,IACnC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gBAAA,IAAoB;AAAA,GACxC,CAAA;AAAA,EACH,SAAA,EAAWA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,EAC3D,GAAA,EAAKA,KAAA,CACF,MAAA,EAAO,CACP,QAAA,EAAS,CACT,QAAA,EAAS,CACT,OAAA,CAAQ,EAAA,GAAK,EAAA,GAAK,EAAA,GAAK,EAAE,CAAA;AAAA,EAC5B,OAAOA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK,CAAA;AAAA,EAC3C,SAAA,EAAWA,KAAA,CAAE,IAAA,CAAK,CAAC,SAAA,EAAW,SAAS,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,SAAS,CAAA;AAAA,EACtE,cAAA,EAAgBA,MACb,MAAA,CAAO;AAAA,IACN,SAASA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,IAC5C,gBAAA,EAAkBA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,CAAC,CAAA;AAAA,IACxD,cAAA,EAAgBA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,EAAE;AAAA,GACxD,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,OAAA,EAAS,IAAA;AAAA,IACT,gBAAA,EAAkB,CAAA;AAAA,IAClB,cAAA,EAAgB;AAAA,GACjB,CAAA;AAAA,EACH,iBAAiBA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK;AACvD,CAAC,CAAA;AAOM,SAAS,kBAAkB,MAAA,EAAmC;AACnE,EAAA,MAAM,SAAmB,EAAC;AAE1B,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,GAAA,EAAK;AACtB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,KAAA,EAAO;AACxB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,GAAA,EAAK;AACrB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,KAAA,EAAO;AACvB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,MAAA,CAAO,SAAS,CAAA,EAAG;AACrB,IAAA,MAAM,IAAI,KAAA;AAAA,MACR,CAAA;AAAA,EAAwC,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA,KAAM,CAAA,IAAA,EAAO,CAAC,CAAA,CAAE,CAAA,CAAE,IAAA,CAAK,IAAI,CAAC,CAAA;AAAA,KAClF;AAAA,EACF;AACF;;;AC3DA,IAAM,IAAA,GAAO,CAAC,CAAA,KAAc,CAAA,CAAE,IAAA,GAAO,WAAA,EAAY,CAAE,OAAA,CAAQ,MAAA,EAAQ,GAAG,CAAA;AACtE,IAAM,GAAA,GAAM,CAAC,CAAA,KAAqBC,iBAAA,CAAA,UAAA,CAAW,QAAQ,EAAE,MAAA,CAAO,CAAC,CAAA,CAAE,MAAA,CAAO,KAAK,CAAA;AAE7E,SAAS,qBAAqB,IAAA,EAAa;AACzC,EAAA,OAAO,IAAA,CAAK,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,IACtB,MAAM,CAAA,CAAE,IAAA;AAAA,IACR,OAAA,EACE,OAAO,CAAA,CAAE,OAAA,KAAY,QAAA,GAAW,EAAE,OAAA,GAAU,IAAA,CAAK,SAAA,CAAU,CAAA,CAAE,OAAO;AAAA,GACxE,CAAE,CAAA;AACJ;AAEA,SAAS,WAAW,OAAA,EAAc;AAChC,EAAA,OAAO;AAAA,IACL,UAAU,OAAA,CAAQ,KAAA,EAAO,WAAW,MAAA,CAAO,OAAA,CAAQ,SAAS,EAAE,CAAA;AAAA,IAC9D,UAAA,EAAY,GAAA,CAAI,OAAA,CAAQ,MAAA,IAAU,EAAE,CAAA;AAAA,IACpC,MAAA,EAAQ,GAAA;AAAA,MACN,KAAK,SAAA,CAAU;AAAA,QACb,aAAa,OAAA,CAAQ,WAAA;AAAA,QACrB,MAAM,OAAA,CAAQ;AAAA,OACf;AAAA,KACH;AAAA,IACA,SAAA,EAAW,IAAI,IAAA,CAAK,SAAA,CAAU,QAAQ,KAAA,IAAS,EAAE,CAAC;AAAA,GACpD;AACF;AAEO,SAAS,oBAAoB,MAAA,EAA6B;AAC/D,EAAA,MAAM,MAAA,GAAS,yBAAA,CAA0B,KAAA,CAAM,MAAM,CAAA;AAErD,EAAA,iBAAA,CAAkB,MAAM,CAAA;AAExB,EAAA,MAAM;AAAA,YACJC,QAAA;AAAA,IACA,KAAA,EAAO,WAAA;AAAA,IACP,SAAA;AAAA,IACA,GAAA;AAAA,IACA,cAAA;AAAA,IACA,KAAA;AAAA,IACA,SAAA;AAAA,IACA;AAAA,GACF,GAAI,MAAA;AACJ,EAAA,MAAM,QAAQ,MAAA,CAAO,KAAA;AAErB,EAAA,MAAM,KAAA,GAAQ,IAAIC,YAAA,CAAM;AAAA,IACtB,KAAKD,QAAA,CAAO,GAAA;AAAA,IACZ,OAAOA,QAAA,CAAO;AAAA,GACf,CAAA;AAED,EAAA,MAAME,OAAA,GAAQ,IAAIC,WAAA,CAAM;AAAA,IACtB,KAAK,WAAA,CAAY,GAAA;AAAA,IACjB,OAAO,WAAA,CAAY;AAAA,GACpB,CAAA;AAED,EAAA,SAAS,YAAY,OAAA,EAAsB;AACzC,IAAA,IAAI,QAAQ,QAAA,EAAU;AACpB,MAAA,MAAM,QAAA,GAAW,MAAM,OAAA,CAAQ,OAAA,CAAQ,QAAQ,CAAA,GAAI,OAAA,CAAQ,WAAW,EAAC;AAGvE,MAAA,IAAI,CAAC,eAAA,IAAmB,QAAA,CAAS,MAAA,GAAS,CAAA,EAAG;AAC3C,QAAA,MAAM,WAAA,GAAc,QAAA,CAAS,QAAA,CAAS,MAAA,GAAS,CAAC,CAAA;AAChD,QAAA,OAAO,KAAK,SAAA,CAAU;AAAA,UACpB,MAAM,WAAA,CAAY,IAAA;AAAA,UAClB,OAAA,EACE,OAAO,WAAA,CAAY,OAAA,KAAY,QAAA,GAC3B,YAAY,OAAA,GACZ,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,OAAO;AAAA,SACzC,CAAA;AAAA,MACH;AAEA,MAAA,OAAO,IAAA,CAAK,SAAA,CAAU,oBAAA,CAAqB,OAAA,CAAQ,QAAQ,CAAC,CAAA;AAAA,IAC9D;AACA,IAAA,IAAI,QAAQ,MAAA,EAAQ;AAClB,MAAA,OAAO,MAAA,CAAO,QAAQ,MAAM,CAAA;AAAA,IAC9B;AACA,IAAA,OAAO,EAAA;AAAA,EACT;AAEA,EAAA,eAAe,kBAAA,CAAmB,YAAoB,KAAA,EAAY;AAChE,IAAA,MAAM,UAAA,GAAa,KAAK,UAAU,CAAA;AAElC,IAAA,MAAM,EAAE,SAAA,EAAU,GAAI,MAAMC,QAAA,CAAM;AAAA,MAChC,KAAA;AAAA,MACA,KAAA,EAAO;AAAA,KACR,CAAA;AAED,IAAA,MAAM,MAAA,GAAS,MAAM,KAAA,CAAM,KAAA,CAAM;AAAA,MAC/B,MAAA,EAAQ,SAAA;AAAA,MACR,IAAA,EAAM,CAAA;AAAA,MACN,eAAA,EAAiB;AAAA,KAClB,CAAA;AAED,IAAA,MAAM,GAAA,GAAM,MAAA,CAAO,IAAA,CAAK,CAAC,CAAA,KAAM;AAC7B,MAAA,IAAI,CAAA,CAAE,KAAA,GAAQ,SAAA,EAAW,OAAO,KAAA;AAEhC,MAAA,MAAM,WAAW,CAAA,CAAE,QAAA;AACnB,MAAA,IAAI,CAAC,UAAU,OAAO,KAAA;AAEtB,MAAA,OACE,QAAA,CAAS,QAAA,KAAa,KAAA,CAAM,QAAA,IAC5B,SAAS,UAAA,KAAe,KAAA,CAAM,UAAA,IAC9B,QAAA,CAAS,MAAA,KAAW,KAAA,CAAM,MAAA,IAC1B,QAAA,CAAS,cAAc,KAAA,CAAM,SAAA;AAAA,IAEjC,CAAC,CAAA;AAED,IAAA,IAAI,GAAA,EAAK;AACP,MAAA,MAAM,SAAS,MAAMF,OAAA,CAAM,IAKxB,GAAA,CAAI,EAAA,CAAG,UAAU,CAAA;AAEpB,MAAA,IAAI,MAAA,EAAQ;AACV,QAAA,IAAI,KAAA,UAAe,GAAA,CAAI,kBAAA,EAAe,IAAI,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAC,CAAA;AAC1D,QAAA,OAAO,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,EAAW;AAAA,MACzC;AAAA,IACF;AAEA,IAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iCAAuB,CAAA;AAC9C,IAAA,OAAO,EAAE,MAAA,EAAQ,IAAA,EAAM,SAAA,EAAW,UAAA,EAAW;AAAA,EAC/C;AAEA,EAAA,eAAe,YAAA,CACb,EAAA,EACA,IAAA,EACA,SAAA,EACA,YACA,KAAA,EACA;AACA,IAAA,MAAM,UAAU,OAAA,GAAU,EAAA;AAC1B,IAAA,MAAM,EAAA,GAAK,MAAMA,OAAA,CAAM,GAAA,CAAI,OAAA,EAAS,GAAA,EAAK,EAAE,EAAA,EAAI,IAAA,EAAM,EAAA,EAAI,EAAA,EAAI,CAAA;AAE7D,IAAA,IAAI,CAAC,EAAA,EAAI;AACP,MAAA,IAAI,KAAA;AACF,QAAA,OAAA,CAAQ,IAAI,4DAAkD,CAAA;AAChE,MAAA;AAAA,IACF;AAEA,IAAA,IAAI;AACF,MAAA,MAAMA,QAAM,GAAA,CAAI,EAAA,EAAI,MAAM,EAAE,EAAA,EAAI,KAAK,CAAA;AACrC,MAAA,MAAM,MAAM,MAAA,CAAO;AAAA,QACjB;AAAA,UACE,EAAA;AAAA,UACA,MAAA,EAAQ,SAAA;AAAA,UACR,QAAA,EAAU;AAAA,YACR,MAAA,EAAQ,UAAA;AAAA,YACR,GAAG;AAAA;AACL;AACF,OACD,CAAA;AAAA,IACH,CAAA,SAAE;AACA,MAAA,MAAMA,OAAA,CAAM,IAAI,OAAO,CAAA;AAAA,IACzB;AAAA,EACF;AAEA,EAAA,MAAM,uBAAA,GAAqD;AAAA,IACzD,UAAA,EAAY,OAAO,EAAE,QAAA,EAAU,QAAO,KAAM;AAC1C,MAAA,MAAM,UAAA,GAAa,YAAY,MAAM,CAAA;AACrC,MAAA,MAAM,KAAA,GAAQ,WAAW,MAAM,CAAA;AAC/B,MAAA,MAAM,cAAc,MAAA,CAAO,MAAA,CAAO,KAAK,CAAA,CAAE,KAAK,GAAG,CAAA;AAEjD,MAAA,MAAM,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,KAAe,MAAM,kBAAA;AAAA,QAC9C,UAAA;AAAA,QACA;AAAA,OACF;AAEA,MAAA,IAAI,MAAA,IAAU,cAAc,SAAA,EAAW;AACrC,QAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gCAA2B,CAAA;AAElD,QAAA,IAAI,SAAsC,EAAC;AAE3C,QAAA,IAAI,OAAO,WAAA,EAAa;AACtB,UAAA,MAAA,GAAS,MAAA,CAAO,WAAA,CAAY,GAAA,CAAI,CAAC,CAAA,KAAW;AAC1C,YAAA,IAAI,CAAA,CAAE,IAAA,KAAS,mBAAA,IAAuB,CAAA,CAAE,SAAA,EAAW;AACjD,cAAA,OAAO,EAAE,GAAG,CAAA,EAAG,SAAA,EAAW,IAAI,IAAA,CAAK,CAAA,CAAE,SAAS,CAAA,EAAE;AAAA,YAClD;AACA,YAAA,OAAO,CAAA;AAAA,UACT,CAAC,CAAA;AAAA,QACH,CAAA,MAAA,IAAW,OAAO,IAAA,EAAM;AACtB,UAAA,MAAA,GAAS;AAAA,YACP,EAAE,IAAA,EAAM,YAAA,EAAc,EAAA,EAAI,OAAO,EAAA,EAAG;AAAA,YACpC,EAAE,MAAM,YAAA,EAAc,KAAA,EAAO,OAAO,IAAA,EAAM,EAAA,EAAI,OAAO,EAAA,EAAG;AAAA,YACxD,EAAE,IAAA,EAAM,QAAA,EAAU,cAAc,MAAA,EAAQ,KAAA,EAAO,OAAO,KAAA;AAAM,WAC9D;AAAA,QACF;AAEA,QAAA,OAAO;AAAA,UACL,QAAQG,yBAAA,CAAuB;AAAA,YAC7B,gBAAA,EAAkB,cAAA,CAAe,OAAA,GAC7B,cAAA,CAAe,gBAAA,GACf,CAAA;AAAA,YACJ,cAAA,EAAgB,cAAA,CAAe,OAAA,GAC3B,cAAA,CAAe,cAAA,GACf,CAAA;AAAA,YACJ;AAAA,WACD;AAAA,SACH;AAAA,MACF;AAEA,MAAA,MAAM,EAAE,MAAA,EAAQ,GAAG,IAAA,EAAK,GAAI,MAAM,QAAA,EAAS;AAE3C,MAAA,MAAM,eAA4C,EAAC;AAEnD,MAAA,MAAM,eAAA,GAAkB,IAAI,eAAA,CAG1B;AAAA,QACA,SAAA,CAAU,OAAO,UAAA,EAAY;AAC3B,UAAA,YAAA,CAAa,KAAK,KAAK,CAAA;AACvB,UAAA,UAAA,CAAW,QAAQ,KAAK,CAAA;AAAA,QAC1B,CAAA;AAAA,QACA,MAAM,KAAA,GAAQ;AACZ,UAAA,MAAM,EAAA,GAAK,MAAA,GAAS,GAAA,CAAI,WAAA,GAAc,MAAM,UAAU,CAAA;AACtD,UAAA,MAAM,YAAA;AAAA,YACJ,EAAA;AAAA,YACA,EAAE,aAAa,YAAA,EAAa;AAAA,YAC5B,SAAA;AAAA,YACA,UAAA;AAAA,YACA;AAAA,WACF;AAAA,QACF;AAAA,OACD,CAAA;AAED,MAAA,OAAO;AAAA,QACL,MAAA,EAAQ,MAAA,CAAO,WAAA,CAAY,eAAe,CAAA;AAAA,QAC1C,GAAG;AAAA,OACL;AAAA,IACF,CAAA;AAAA,IAEA,YAAA,EAAc,OAAO,EAAE,UAAA,EAAY,QAAO,KAAM;AAC9C,MAAA,MAAM,UAAA,GAAa,YAAY,MAAM,CAAA;AACrC,MAAA,MAAM,KAAA,GAAQ,WAAW,MAAM,CAAA;AAC/B,MAAA,MAAM,cAAc,MAAA,CAAO,MAAA,CAAO,KAAK,CAAA,CAAE,KAAK,GAAG,CAAA;AAEjD,MAAA,MAAM,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,KAAe,MAAM,kBAAA;AAAA,QAC9C,UAAA;AAAA,QACA;AAAA,OACF;AAEA,MAAA,IAAI,MAAA,IAAU,cAAc,SAAA,EAAW;AACrC,QAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,oCAA+B,CAAA;AAEtD,QAAA,IAAI,MAAA,EAAQ,UAAU,SAAA,EAAW;AAC/B,UAAA,MAAA,CAAO,SAAS,SAAA,GAAY,IAAI,IAAA,CAAK,MAAA,CAAO,SAAS,SAAS,CAAA;AAAA,QAChE;AAGA,QAAA,OAAO,MAAA;AAAA,MACT;AAEA,MAAA,MAAM,MAAA,GAAS,MAAM,UAAA,EAAW;AAEhC,MAAA,MAAM,EAAA,GAAK,MAAA,GAAS,GAAA,CAAI,WAAA,GAAc,MAAM,UAAU,CAAA;AACtD,MAAA,MAAM,YAAA,CAAa,EAAA,EAAI,MAAA,EAAQ,SAAA,EAAW,YAAY,KAAK,CAAA;AAE3D,MAAA,OAAO,MAAA;AAAA,IACT;AAAA,GACF;AAEA,EAAA,OAAO;AAAA,IACL,UAAA,EAAY,OACV,OAAA,KAC0C;AAC1C,MAAA,MAAM,eAAeC,oBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrBC,UAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAOC,aAAA,CAAW;AAAA,QAChB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,YAAA,EAAc,OAIZ,OAAA,KAC+C;AAC/C,MAAA,MAAM,eAAeF,oBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrBC,UAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAOE,eAAA,CAAa;AAAA,QAClB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,cAAA,EAAgB,OACd,OAAA,KACqC;AACrC,MAAA,MAAM,eAAeH,oBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrBC,UAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAQ,MAAMG,iBAAA,CAAe;AAAA,QAC3B,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,YAAA,EAAc,OACZ,OAAA,KAC2C;AAC3C,MAAA,MAAM,eAAeJ,oBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrBC,UAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAOI,eAAA,CAAa;AAAA,QAClB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH;AAAA,GACF;AACF","file":"index.js","sourcesContent":["import { z } from \"zod\";\nimport \"dotenv/config\";\n\nexport const semanticCacheConfigSchema = z.object({\n  model: z.union([\n    z.string(),\n    z.custom((val) => {\n      return val && typeof val === \"object\";\n    }),\n  ]),\n  vector: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.VECTOR_REST_URL ?? \"\",\n      token: process.env.VECTOR_REST_TOKEN ?? \"\",\n    }),\n  redis: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.REDIS_REST_URL ?? \"\",\n      token: process.env.REDIS_REST_TOKEN ?? \"\",\n    }),\n  threshold: z.number().min(0).max(1).optional().default(0.92),\n  ttl: z\n    .number()\n    .positive()\n    .optional()\n    .default(60 * 60 * 24 * 14),\n  debug: z.boolean().optional().default(false),\n  cacheMode: z.enum([\"default\", \"refresh\"]).optional().default(\"default\"),\n  simulateStream: z\n    .object({\n      enabled: z.boolean().optional().default(true),\n      initialDelayInMs: z.number().min(0).optional().default(0),\n      chunkDelayInMs: z.number().min(0).optional().default(10),\n    })\n    .optional()\n    .default({\n      enabled: true,\n      initialDelayInMs: 0,\n      chunkDelayInMs: 10,\n    }),\n  useFullMessages: z.boolean().optional().default(false),\n});\n\nexport type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;\nexport type SemanticCacheConfigParsed = z.output<\n  typeof semanticCacheConfigSchema\n>;\n\nexport function validateEnvConfig(config: SemanticCacheConfigParsed) {\n  const errors: string[] = [];\n\n  if (!config.vector.url) {\n    errors.push(\n      \"Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.vector.token) {\n    errors.push(\n      \"Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (!config.redis.url) {\n    errors.push(\n      \"Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.redis.token) {\n    errors.push(\n      \"Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (errors.length > 0) {\n    throw new Error(\n      `Semantic Cache Configuration Error:\\n${errors.map((e) => `  - ${e}`).join(\"\\n\")}`,\n    );\n  }\n}\n","import {\n  type LanguageModelV2Middleware,\n  type LanguageModelV2StreamPart,\n  type LanguageModelV2Content,\n  type LanguageModelV2CallOptions,\n} from \"@ai-sdk/provider\";\nimport {\n  embed,\n  streamText,\n  generateText,\n  generateObject,\n  streamObject,\n  simulateReadableStream,\n  wrapLanguageModel,\n  type EmbeddingModel,\n  type StreamTextResult,\n  type GenerateTextResult,\n  type GenerateObjectResult,\n  type StreamObjectResult,\n  gateway,\n} from \"ai\";\n\nimport { Index } from \"@upstash/vector\";\nimport { Redis } from \"@upstash/redis\";\nimport * as crypto from \"node:crypto\";\nimport {\n  semanticCacheConfigSchema,\n  type SemanticCacheConfig,\n  validateEnvConfig,\n} from \"./schema\";\n\nconst norm = (s: string) => s.trim().toLowerCase().replace(/\\s+/g, \" \");\nconst sha = (s: string) => crypto.createHash(\"sha256\").update(s).digest(\"hex\");\n\nfunction canonicalizeMessages(msgs: any[]) {\n  return msgs.map((m) => ({\n    role: m.role,\n    content:\n      typeof m.content === \"string\" ? m.content : JSON.stringify(m.content),\n  }));\n}\n\nfunction buildScope(options: any) {\n  return {\n    llmModel: options.model?.modelId ?? String(options.model ?? \"\"),\n    systemHash: sha(options.system ?? \"\"),\n    params: sha(\n      JSON.stringify({\n        temperature: options.temperature,\n        topP: options.topP,\n      }),\n    ),\n    toolsHash: sha(JSON.stringify(options.tools ?? {})),\n  };\n}\n\nexport function createSemanticCache(config: SemanticCacheConfig) {\n  const parsed = semanticCacheConfigSchema.parse(config);\n\n  validateEnvConfig(parsed);\n\n  const {\n    vector,\n    redis: redisConfig,\n    threshold,\n    ttl,\n    simulateStream,\n    debug,\n    cacheMode,\n    useFullMessages,\n  } = parsed;\n  const model = parsed.model as EmbeddingModel<string>;\n\n  const index = new Index({\n    url: vector.url as string,\n    token: vector.token as string,\n  });\n\n  const redis = new Redis({\n    url: redisConfig.url as string,\n    token: redisConfig.token as string,\n  });\n\n  function getCacheKey(options: any): string {\n    if (options.messages) {\n      const messages = Array.isArray(options.messages) ? options.messages : [];\n\n      // By default, use only the last message to avoid token limit issues\n      if (!useFullMessages && messages.length > 0) {\n        const lastMessage = messages[messages.length - 1];\n        return JSON.stringify({\n          role: lastMessage.role,\n          content:\n            typeof lastMessage.content === \"string\"\n              ? lastMessage.content\n              : JSON.stringify(lastMessage.content),\n        });\n      }\n\n      return JSON.stringify(canonicalizeMessages(options.messages));\n    }\n    if (options.prompt) {\n      return String(options.prompt);\n    }\n    return \"\";\n  }\n\n  async function checkSemanticCache(cacheInput: string, scope: any) {\n    const promptNorm = norm(cacheInput);\n\n    const { embedding } = await embed({\n      model,\n      value: promptNorm,\n    });\n\n    const result = await index.query({\n      vector: embedding,\n      topK: 3,\n      includeMetadata: true,\n    });\n\n    const hit = result.find((m) => {\n      if (m.score < threshold) return false;\n\n      const metadata = m.metadata as any;\n      if (!metadata) return false;\n\n      return (\n        metadata.llmModel === scope.llmModel &&\n        metadata.systemHash === scope.systemHash &&\n        metadata.params === scope.params &&\n        metadata.toolsHash === scope.toolsHash\n      );\n    });\n\n    if (hit) {\n      const cached = await redis.get<{\n        streamParts?: LanguageModelV2StreamPart[];\n        text?: string;\n        content?: LanguageModelV2Content;\n        [key: string]: any;\n      }>(hit.id.toString());\n\n      if (cached) {\n        if (debug) console.log(\"✅ cache hit\", hit.score.toFixed(3));\n        return { cached, embedding, promptNorm };\n      }\n    }\n\n    if (debug) console.log(\"❌ miss -> generating…\");\n    return { cached: null, embedding, promptNorm };\n  }\n\n  async function storeInCache(\n    id: string,\n    data: any,\n    embedding: number[],\n    promptNorm: string,\n    scope: any,\n  ) {\n    const lockKey = \"lock:\" + id;\n    const ok = await redis.set(lockKey, \"1\", { nx: true, ex: 15 });\n\n    if (!ok) {\n      if (debug)\n        console.log(\"⚠️ Another process is writing to cache, skipping\");\n      return;\n    }\n\n    try {\n      await redis.set(id, data, { ex: ttl });\n      await index.upsert([\n        {\n          id,\n          vector: embedding,\n          metadata: {\n            prompt: promptNorm,\n            ...scope,\n          },\n        },\n      ]);\n    } finally {\n      await redis.del(lockKey);\n    }\n  }\n\n  const semanticCacheMiddleware: LanguageModelV2Middleware = {\n    wrapStream: async ({ doStream, params }) => {\n      const cacheInput = getCacheKey(params);\n      const scope = buildScope(params);\n      const promptScope = Object.values(scope).join(\"|\");\n\n      const { cached, embedding, promptNorm } = await checkSemanticCache(\n        cacheInput,\n        scope,\n      );\n\n      if (cached && cacheMode !== \"refresh\") {\n        if (debug) console.log(\"✅ Returning cached stream\");\n\n        let chunks: LanguageModelV2StreamPart[] = [];\n\n        if (cached.streamParts) {\n          chunks = cached.streamParts.map((p: any) => {\n            if (p.type === \"response-metadata\" && p.timestamp) {\n              return { ...p, timestamp: new Date(p.timestamp) };\n            }\n            return p;\n          });\n        } else if (cached.text) {\n          chunks = [\n            { type: \"text-start\", id: cached.id },\n            { type: \"text-delta\", delta: cached.text, id: cached.id },\n            { type: \"finish\", finishReason: \"stop\", usage: cached.usage },\n          ];\n        }\n\n        return {\n          stream: simulateReadableStream({\n            initialDelayInMs: simulateStream.enabled\n              ? simulateStream.initialDelayInMs\n              : 0,\n            chunkDelayInMs: simulateStream.enabled\n              ? simulateStream.chunkDelayInMs\n              : 0,\n            chunks,\n          }),\n        };\n      }\n\n      const { stream, ...rest } = await doStream();\n\n      const fullResponse: LanguageModelV2StreamPart[] = [];\n\n      const transformStream = new TransformStream<\n        LanguageModelV2StreamPart,\n        LanguageModelV2StreamPart\n      >({\n        transform(chunk, controller) {\n          fullResponse.push(chunk);\n          controller.enqueue(chunk);\n        },\n        async flush() {\n          const id = \"llm:\" + sha(promptScope + \"|\" + promptNorm);\n          await storeInCache(\n            id,\n            { streamParts: fullResponse },\n            embedding,\n            promptNorm,\n            scope,\n          );\n        },\n      });\n\n      return {\n        stream: stream.pipeThrough(transformStream),\n        ...rest,\n      };\n    },\n\n    wrapGenerate: async ({ doGenerate, params }) => {\n      const cacheInput = getCacheKey(params);\n      const scope = buildScope(params);\n      const promptScope = Object.values(scope).join(\"|\");\n\n      const { cached, embedding, promptNorm } = await checkSemanticCache(\n        cacheInput,\n        scope,\n      );\n\n      if (cached && cacheMode !== \"refresh\") {\n        if (debug) console.log(\"✅ Returning cached generation\");\n\n        if (cached?.response?.timestamp) {\n          cached.response.timestamp = new Date(cached.response.timestamp);\n        }\n\n        type GenReturn<T> = T extends () => PromiseLike<infer R> ? R : never;\n        return cached as unknown as GenReturn<typeof doGenerate>;\n      }\n\n      const result = await doGenerate();\n\n      const id = \"llm:\" + sha(promptScope + \"|\" + promptNorm);\n      await storeInCache(id, result, embedding, promptNorm, scope);\n\n      return result;\n    },\n  };\n\n  return {\n    streamText: async <TOOLS extends Record<string, any> = {}>(\n      options: Parameters<typeof streamText<TOOLS>>[0],\n    ): Promise<StreamTextResult<TOOLS, any>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return streamText({\n        ...options,\n        model: wrappedModel,\n      });\n    },\n\n    generateText: async <\n      TOOLS extends Record<string, any> = {},\n      OUTPUT = undefined,\n    >(\n      options: Parameters<typeof generateText<TOOLS, OUTPUT>>[0],\n    ): Promise<GenerateTextResult<TOOLS, OUTPUT>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return generateText({\n        ...options,\n        model: wrappedModel,\n      });\n    },\n\n    generateObject: async <T = any>(\n      options: Parameters<typeof generateObject>[0],\n    ): Promise<GenerateObjectResult<T>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return (await generateObject({\n        ...options,\n        model: wrappedModel,\n      })) as GenerateObjectResult<T>;\n    },\n\n    streamObject: async <T = any>(\n      options: Parameters<typeof streamObject>[0],\n    ): Promise<StreamObjectResult<T, T, any>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return streamObject({\n        ...options,\n        model: wrappedModel,\n      }) as unknown as StreamObjectResult<T, T, any>;\n    },\n  };\n}\n"]}

package/dist/index.mjs

@@ -0,0 +1,312 @@
+import '@ai-sdk/provider';
+import { wrapLanguageModel, gateway, streamObject, generateObject, generateText, streamText, simulateReadableStream, embed } from 'ai';
+import { Index } from '@upstash/vector';
+import { Redis } from '@upstash/redis';
+import * as crypto from 'crypto';
+import { z } from 'zod';
+import 'dotenv/config';
+
+// index.ts
+var semanticCacheConfigSchema = z.object({
+  model: z.union([
+    z.string(),
+    z.custom((val) => {
+      return val && typeof val === "object";
+    })
+  ]),
+  vector: z.object({
+    url: z.url(),
+    token: z.string().min(1)
+  }).optional().default({
+    url: process.env.VECTOR_REST_URL ?? "",
+    token: process.env.VECTOR_REST_TOKEN ?? ""
+  }),
+  redis: z.object({
+    url: z.url(),
+    token: z.string().min(1)
+  }).optional().default({
+    url: process.env.REDIS_REST_URL ?? "",
+    token: process.env.REDIS_REST_TOKEN ?? ""
+  }),
+  threshold: z.number().min(0).max(1).optional().default(0.92),
+  ttl: z.number().positive().optional().default(60 * 60 * 24 * 14),
+  debug: z.boolean().optional().default(false),
+  cacheMode: z.enum(["default", "refresh"]).optional().default("default"),
+  simulateStream: z.object({
+    enabled: z.boolean().optional().default(true),
+    initialDelayInMs: z.number().min(0).optional().default(0),
+    chunkDelayInMs: z.number().min(0).optional().default(10)
+  }).optional().default({
+    enabled: true,
+    initialDelayInMs: 0,
+    chunkDelayInMs: 10
+  }),
+  useFullMessages: z.boolean().optional().default(false)
+});
+function validateEnvConfig(config) {
+  const errors = [];
+  if (!config.vector.url) {
+    errors.push(
+      "Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable."
+    );
+  }
+  if (!config.vector.token) {
+    errors.push(
+      "Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable."
+    );
+  }
+  if (!config.redis.url) {
+    errors.push(
+      "Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable."
+    );
+  }
+  if (!config.redis.token) {
+    errors.push(
+      "Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable."
+    );
+  }
+  if (errors.length > 0) {
+    throw new Error(
+      `Semantic Cache Configuration Error:
+${errors.map((e) => `  - ${e}`).join("\n")}`
+    );
+  }
+}
+
+// index.ts
+var norm = (s) => s.trim().toLowerCase().replace(/\s+/g, " ");
+var sha = (s) => crypto.createHash("sha256").update(s).digest("hex");
+function canonicalizeMessages(msgs) {
+  return msgs.map((m) => ({
+    role: m.role,
+    content: typeof m.content === "string" ? m.content : JSON.stringify(m.content)
+  }));
+}
+function buildScope(options) {
+  return {
+    llmModel: options.model?.modelId ?? String(options.model ?? ""),
+    systemHash: sha(options.system ?? ""),
+    params: sha(
+      JSON.stringify({
+        temperature: options.temperature,
+        topP: options.topP
+      })
+    ),
+    toolsHash: sha(JSON.stringify(options.tools ?? {}))
+  };
+}
+function createSemanticCache(config) {
+  const parsed = semanticCacheConfigSchema.parse(config);
+  validateEnvConfig(parsed);
+  const {
+    vector,
+    redis: redisConfig,
+    threshold,
+    ttl,
+    simulateStream,
+    debug,
+    cacheMode,
+    useFullMessages
+  } = parsed;
+  const model = parsed.model;
+  const index = new Index({
+    url: vector.url,
+    token: vector.token
+  });
+  const redis = new Redis({
+    url: redisConfig.url,
+    token: redisConfig.token
+  });
+  function getCacheKey(options) {
+    if (options.messages) {
+      const messages = Array.isArray(options.messages) ? options.messages : [];
+      if (!useFullMessages && messages.length > 0) {
+        const lastMessage = messages[messages.length - 1];
+        return JSON.stringify({
+          role: lastMessage.role,
+          content: typeof lastMessage.content === "string" ? lastMessage.content : JSON.stringify(lastMessage.content)
+        });
+      }
+      return JSON.stringify(canonicalizeMessages(options.messages));
+    }
+    if (options.prompt) {
+      return String(options.prompt);
+    }
+    return "";
+  }
+  async function checkSemanticCache(cacheInput, scope) {
+    const promptNorm = norm(cacheInput);
+    const { embedding } = await embed({
+      model,
+      value: promptNorm
+    });
+    const result = await index.query({
+      vector: embedding,
+      topK: 3,
+      includeMetadata: true
+    });
+    const hit = result.find((m) => {
+      if (m.score < threshold) return false;
+      const metadata = m.metadata;
+      if (!metadata) return false;
+      return metadata.llmModel === scope.llmModel && metadata.systemHash === scope.systemHash && metadata.params === scope.params && metadata.toolsHash === scope.toolsHash;
+    });
+    if (hit) {
+      const cached = await redis.get(hit.id.toString());
+      if (cached) {
+        if (debug) console.log("\u2705 cache hit", hit.score.toFixed(3));
+        return { cached, embedding, promptNorm };
+      }
+    }
+    if (debug) console.log("\u274C miss -> generating\u2026");
+    return { cached: null, embedding, promptNorm };
+  }
+  async function storeInCache(id, data, embedding, promptNorm, scope) {
+    const lockKey = "lock:" + id;
+    const ok = await redis.set(lockKey, "1", { nx: true, ex: 15 });
+    if (!ok) {
+      if (debug)
+        console.log("\u26A0\uFE0F Another process is writing to cache, skipping");
+      return;
+    }
+    try {
+      await redis.set(id, data, { ex: ttl });
+      await index.upsert([
+        {
+          id,
+          vector: embedding,
+          metadata: {
+            prompt: promptNorm,
+            ...scope
+          }
+        }
+      ]);
+    } finally {
+      await redis.del(lockKey);
+    }
+  }
+  const semanticCacheMiddleware = {
+    wrapStream: async ({ doStream, params }) => {
+      const cacheInput = getCacheKey(params);
+      const scope = buildScope(params);
+      const promptScope = Object.values(scope).join("|");
+      const { cached, embedding, promptNorm } = await checkSemanticCache(
+        cacheInput,
+        scope
+      );
+      if (cached && cacheMode !== "refresh") {
+        if (debug) console.log("\u2705 Returning cached stream");
+        let chunks = [];
+        if (cached.streamParts) {
+          chunks = cached.streamParts.map((p) => {
+            if (p.type === "response-metadata" && p.timestamp) {
+              return { ...p, timestamp: new Date(p.timestamp) };
+            }
+            return p;
+          });
+        } else if (cached.text) {
+          chunks = [
+            { type: "text-start", id: cached.id },
+            { type: "text-delta", delta: cached.text, id: cached.id },
+            { type: "finish", finishReason: "stop", usage: cached.usage }
+          ];
+        }
+        return {
+          stream: simulateReadableStream({
+            initialDelayInMs: simulateStream.enabled ? simulateStream.initialDelayInMs : 0,
+            chunkDelayInMs: simulateStream.enabled ? simulateStream.chunkDelayInMs : 0,
+            chunks
+          })
+        };
+      }
+      const { stream, ...rest } = await doStream();
+      const fullResponse = [];
+      const transformStream = new TransformStream({
+        transform(chunk, controller) {
+          fullResponse.push(chunk);
+          controller.enqueue(chunk);
+        },
+        async flush() {
+          const id = "llm:" + sha(promptScope + "|" + promptNorm);
+          await storeInCache(
+            id,
+            { streamParts: fullResponse },
+            embedding,
+            promptNorm,
+            scope
+          );
+        }
+      });
+      return {
+        stream: stream.pipeThrough(transformStream),
+        ...rest
+      };
+    },
+    wrapGenerate: async ({ doGenerate, params }) => {
+      const cacheInput = getCacheKey(params);
+      const scope = buildScope(params);
+      const promptScope = Object.values(scope).join("|");
+      const { cached, embedding, promptNorm } = await checkSemanticCache(
+        cacheInput,
+        scope
+      );
+      if (cached && cacheMode !== "refresh") {
+        if (debug) console.log("\u2705 Returning cached generation");
+        if (cached?.response?.timestamp) {
+          cached.response.timestamp = new Date(cached.response.timestamp);
+        }
+        return cached;
+      }
+      const result = await doGenerate();
+      const id = "llm:" + sha(promptScope + "|" + promptNorm);
+      await storeInCache(id, result, embedding, promptNorm, scope);
+      return result;
+    }
+  };
+  return {
+    streamText: async (options) => {
+      const wrappedModel = wrapLanguageModel({
+        model: typeof options.model === "string" ? gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return streamText({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    generateText: async (options) => {
+      const wrappedModel = wrapLanguageModel({
+        model: typeof options.model === "string" ? gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return generateText({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    generateObject: async (options) => {
+      const wrappedModel = wrapLanguageModel({
+        model: typeof options.model === "string" ? gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return await generateObject({
+        ...options,
+        model: wrappedModel
+      });
+    },
+    streamObject: async (options) => {
+      const wrappedModel = wrapLanguageModel({
+        model: typeof options.model === "string" ? gateway(options.model) : options.model,
+        middleware: semanticCacheMiddleware
+      });
+      return streamObject({
+        ...options,
+        model: wrappedModel
+      });
+    }
+  };
+}
+
+export { createSemanticCache };
+//# sourceMappingURL=index.mjs.map
+//# sourceMappingURL=index.mjs.map

package/dist/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../schema.ts","../index.ts"],"names":[],"mappings":";;;;;;;;;AAGO,IAAM,yBAAA,GAA4B,EAAE,MAAA,CAAO;AAAA,EAChD,KAAA,EAAO,EAAE,KAAA,CAAM;AAAA,IACb,EAAE,MAAA,EAAO;AAAA,IACT,CAAA,CAAE,MAAA,CAAO,CAAC,GAAA,KAAQ;AAChB,MAAA,OAAO,GAAA,IAAO,OAAO,GAAA,KAAQ,QAAA;AAAA,IAC/B,CAAC;AAAA,GACF,CAAA;AAAA,EACD,MAAA,EAAQ,EACL,MAAA,CAAO;AAAA,IACN,GAAA,EAAK,EAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAO,CAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,eAAA,IAAmB,EAAA;AAAA,IACpC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iBAAA,IAAqB;AAAA,GACzC,CAAA;AAAA,EACH,KAAA,EAAO,EACJ,MAAA,CAAO;AAAA,IACN,GAAA,EAAK,EAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAO,CAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,cAAA,IAAkB,EAAA;AAAA,IACnC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gBAAA,IAAoB;AAAA,GACxC,CAAA;AAAA,EACH,SAAA,EAAW,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,EAC3D,GAAA,EAAK,CAAA,CACF,MAAA,EAAO,CACP,QAAA,EAAS,CACT,QAAA,EAAS,CACT,OAAA,CAAQ,EAAA,GAAK,EAAA,GAAK,EAAA,GAAK,EAAE,CAAA;AAAA,EAC5B,OAAO,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK,CAAA;AAAA,EAC3C,SAAA,EAAW,CAAA,CAAE,IAAA,CAAK,CAAC,SAAA,EAAW,SAAS,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,SAAS,CAAA;AAAA,EACtE,cAAA,EAAgB,EACb,MAAA,CAAO;AAAA,IACN,SAAS,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,IAC5C,gBAAA,EAAkB,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,CAAC,CAAA;AAAA,IACxD,cAAA,EAAgB,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,EAAE;AAAA,GACxD,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,OAAA,EAAS,IAAA;AAAA,IACT,gBAAA,EAAkB,CAAA;AAAA,IAClB,cAAA,EAAgB;AAAA,GACjB,CAAA;AAAA,EACH,iBAAiB,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK;AACvD,CAAC,CAAA;AAOM,SAAS,kBAAkB,MAAA,EAAmC;AACnE,EAAA,MAAM,SAAmB,EAAC;AAE1B,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,GAAA,EAAK;AACtB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,KAAA,EAAO;AACxB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,GAAA,EAAK;AACrB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,KAAA,EAAO;AACvB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,MAAA,CAAO,SAAS,CAAA,EAAG;AACrB,IAAA,MAAM,IAAI,KAAA;AAAA,MACR,CAAA;AAAA,EAAwC,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA,KAAM,CAAA,IAAA,EAAO,CAAC,CAAA,CAAE,CAAA,CAAE,IAAA,CAAK,IAAI,CAAC,CAAA;AAAA,KAClF;AAAA,EACF;AACF;;;AC3DA,IAAM,IAAA,GAAO,CAAC,CAAA,KAAc,CAAA,CAAE,IAAA,GAAO,WAAA,EAAY,CAAE,OAAA,CAAQ,MAAA,EAAQ,GAAG,CAAA;AACtE,IAAM,GAAA,GAAM,CAAC,CAAA,KAAqB,MAAA,CAAA,UAAA,CAAW,QAAQ,EAAE,MAAA,CAAO,CAAC,CAAA,CAAE,MAAA,CAAO,KAAK,CAAA;AAE7E,SAAS,qBAAqB,IAAA,EAAa;AACzC,EAAA,OAAO,IAAA,CAAK,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,IACtB,MAAM,CAAA,CAAE,IAAA;AAAA,IACR,OAAA,EACE,OAAO,CAAA,CAAE,OAAA,KAAY,QAAA,GAAW,EAAE,OAAA,GAAU,IAAA,CAAK,SAAA,CAAU,CAAA,CAAE,OAAO;AAAA,GACxE,CAAE,CAAA;AACJ;AAEA,SAAS,WAAW,OAAA,EAAc;AAChC,EAAA,OAAO;AAAA,IACL,UAAU,OAAA,CAAQ,KAAA,EAAO,WAAW,MAAA,CAAO,OAAA,CAAQ,SAAS,EAAE,CAAA;AAAA,IAC9D,UAAA,EAAY,GAAA,CAAI,OAAA,CAAQ,MAAA,IAAU,EAAE,CAAA;AAAA,IACpC,MAAA,EAAQ,GAAA;AAAA,MACN,KAAK,SAAA,CAAU;AAAA,QACb,aAAa,OAAA,CAAQ,WAAA;AAAA,QACrB,MAAM,OAAA,CAAQ;AAAA,OACf;AAAA,KACH;AAAA,IACA,SAAA,EAAW,IAAI,IAAA,CAAK,SAAA,CAAU,QAAQ,KAAA,IAAS,EAAE,CAAC;AAAA,GACpD;AACF;AAEO,SAAS,oBAAoB,MAAA,EAA6B;AAC/D,EAAA,MAAM,MAAA,GAAS,yBAAA,CAA0B,KAAA,CAAM,MAAM,CAAA;AAErD,EAAA,iBAAA,CAAkB,MAAM,CAAA;AAExB,EAAA,MAAM;AAAA,IACJ,MAAA;AAAA,IACA,KAAA,EAAO,WAAA;AAAA,IACP,SAAA;AAAA,IACA,GAAA;AAAA,IACA,cAAA;AAAA,IACA,KAAA;AAAA,IACA,SAAA;AAAA,IACA;AAAA,GACF,GAAI,MAAA;AACJ,EAAA,MAAM,QAAQ,MAAA,CAAO,KAAA;AAErB,EAAA,MAAM,KAAA,GAAQ,IAAI,KAAA,CAAM;AAAA,IACtB,KAAK,MAAA,CAAO,GAAA;AAAA,IACZ,OAAO,MAAA,CAAO;AAAA,GACf,CAAA;AAED,EAAA,MAAM,KAAA,GAAQ,IAAI,KAAA,CAAM;AAAA,IACtB,KAAK,WAAA,CAAY,GAAA;AAAA,IACjB,OAAO,WAAA,CAAY;AAAA,GACpB,CAAA;AAED,EAAA,SAAS,YAAY,OAAA,EAAsB;AACzC,IAAA,IAAI,QAAQ,QAAA,EAAU;AACpB,MAAA,MAAM,QAAA,GAAW,MAAM,OAAA,CAAQ,OAAA,CAAQ,QAAQ,CAAA,GAAI,OAAA,CAAQ,WAAW,EAAC;AAGvE,MAAA,IAAI,CAAC,eAAA,IAAmB,QAAA,CAAS,MAAA,GAAS,CAAA,EAAG;AAC3C,QAAA,MAAM,WAAA,GAAc,QAAA,CAAS,QAAA,CAAS,MAAA,GAAS,CAAC,CAAA;AAChD,QAAA,OAAO,KAAK,SAAA,CAAU;AAAA,UACpB,MAAM,WAAA,CAAY,IAAA;AAAA,UAClB,OAAA,EACE,OAAO,WAAA,CAAY,OAAA,KAAY,QAAA,GAC3B,YAAY,OAAA,GACZ,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,OAAO;AAAA,SACzC,CAAA;AAAA,MACH;AAEA,MAAA,OAAO,IAAA,CAAK,SAAA,CAAU,oBAAA,CAAqB,OAAA,CAAQ,QAAQ,CAAC,CAAA;AAAA,IAC9D;AACA,IAAA,IAAI,QAAQ,MAAA,EAAQ;AAClB,MAAA,OAAO,MAAA,CAAO,QAAQ,MAAM,CAAA;AAAA,IAC9B;AACA,IAAA,OAAO,EAAA;AAAA,EACT;AAEA,EAAA,eAAe,kBAAA,CAAmB,YAAoB,KAAA,EAAY;AAChE,IAAA,MAAM,UAAA,GAAa,KAAK,UAAU,CAAA;AAElC,IAAA,MAAM,EAAE,SAAA,EAAU,GAAI,MAAM,KAAA,CAAM;AAAA,MAChC,KAAA;AAAA,MACA,KAAA,EAAO;AAAA,KACR,CAAA;AAED,IAAA,MAAM,MAAA,GAAS,MAAM,KAAA,CAAM,KAAA,CAAM;AAAA,MAC/B,MAAA,EAAQ,SAAA;AAAA,MACR,IAAA,EAAM,CAAA;AAAA,MACN,eAAA,EAAiB;AAAA,KAClB,CAAA;AAED,IAAA,MAAM,GAAA,GAAM,MAAA,CAAO,IAAA,CAAK,CAAC,CAAA,KAAM;AAC7B,MAAA,IAAI,CAAA,CAAE,KAAA,GAAQ,SAAA,EAAW,OAAO,KAAA;AAEhC,MAAA,MAAM,WAAW,CAAA,CAAE,QAAA;AACnB,MAAA,IAAI,CAAC,UAAU,OAAO,KAAA;AAEtB,MAAA,OACE,QAAA,CAAS,QAAA,KAAa,KAAA,CAAM,QAAA,IAC5B,SAAS,UAAA,KAAe,KAAA,CAAM,UAAA,IAC9B,QAAA,CAAS,MAAA,KAAW,KAAA,CAAM,MAAA,IAC1B,QAAA,CAAS,cAAc,KAAA,CAAM,SAAA;AAAA,IAEjC,CAAC,CAAA;AAED,IAAA,IAAI,GAAA,EAAK;AACP,MAAA,MAAM,SAAS,MAAM,KAAA,CAAM,IAKxB,GAAA,CAAI,EAAA,CAAG,UAAU,CAAA;AAEpB,MAAA,IAAI,MAAA,EAAQ;AACV,QAAA,IAAI,KAAA,UAAe,GAAA,CAAI,kBAAA,EAAe,IAAI,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAC,CAAA;AAC1D,QAAA,OAAO,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,EAAW;AAAA,MACzC;AAAA,IACF;AAEA,IAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iCAAuB,CAAA;AAC9C,IAAA,OAAO,EAAE,MAAA,EAAQ,IAAA,EAAM,SAAA,EAAW,UAAA,EAAW;AAAA,EAC/C;AAEA,EAAA,eAAe,YAAA,CACb,EAAA,EACA,IAAA,EACA,SAAA,EACA,YACA,KAAA,EACA;AACA,IAAA,MAAM,UAAU,OAAA,GAAU,EAAA;AAC1B,IAAA,MAAM,EAAA,GAAK,MAAM,KAAA,CAAM,GAAA,CAAI,OAAA,EAAS,GAAA,EAAK,EAAE,EAAA,EAAI,IAAA,EAAM,EAAA,EAAI,EAAA,EAAI,CAAA;AAE7D,IAAA,IAAI,CAAC,EAAA,EAAI;AACP,MAAA,IAAI,KAAA;AACF,QAAA,OAAA,CAAQ,IAAI,4DAAkD,CAAA;AAChE,MAAA;AAAA,IACF;AAEA,IAAA,IAAI;AACF,MAAA,MAAM,MAAM,GAAA,CAAI,EAAA,EAAI,MAAM,EAAE,EAAA,EAAI,KAAK,CAAA;AACrC,MAAA,MAAM,MAAM,MAAA,CAAO;AAAA,QACjB;AAAA,UACE,EAAA;AAAA,UACA,MAAA,EAAQ,SAAA;AAAA,UACR,QAAA,EAAU;AAAA,YACR,MAAA,EAAQ,UAAA;AAAA,YACR,GAAG;AAAA;AACL;AACF,OACD,CAAA;AAAA,IACH,CAAA,SAAE;AACA,MAAA,MAAM,KAAA,CAAM,IAAI,OAAO,CAAA;AAAA,IACzB;AAAA,EACF;AAEA,EAAA,MAAM,uBAAA,GAAqD;AAAA,IACzD,UAAA,EAAY,OAAO,EAAE,QAAA,EAAU,QAAO,KAAM;AAC1C,MAAA,MAAM,UAAA,GAAa,YAAY,MAAM,CAAA;AACrC,MAAA,MAAM,KAAA,GAAQ,WAAW,MAAM,CAAA;AAC/B,MAAA,MAAM,cAAc,MAAA,CAAO,MAAA,CAAO,KAAK,CAAA,CAAE,KAAK,GAAG,CAAA;AAEjD,MAAA,MAAM,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,KAAe,MAAM,kBAAA;AAAA,QAC9C,UAAA;AAAA,QACA;AAAA,OACF;AAEA,MAAA,IAAI,MAAA,IAAU,cAAc,SAAA,EAAW;AACrC,QAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gCAA2B,CAAA;AAElD,QAAA,IAAI,SAAsC,EAAC;AAE3C,QAAA,IAAI,OAAO,WAAA,EAAa;AACtB,UAAA,MAAA,GAAS,MAAA,CAAO,WAAA,CAAY,GAAA,CAAI,CAAC,CAAA,KAAW;AAC1C,YAAA,IAAI,CAAA,CAAE,IAAA,KAAS,mBAAA,IAAuB,CAAA,CAAE,SAAA,EAAW;AACjD,cAAA,OAAO,EAAE,GAAG,CAAA,EAAG,SAAA,EAAW,IAAI,IAAA,CAAK,CAAA,CAAE,SAAS,CAAA,EAAE;AAAA,YAClD;AACA,YAAA,OAAO,CAAA;AAAA,UACT,CAAC,CAAA;AAAA,QACH,CAAA,MAAA,IAAW,OAAO,IAAA,EAAM;AACtB,UAAA,MAAA,GAAS;AAAA,YACP,EAAE,IAAA,EAAM,YAAA,EAAc,EAAA,EAAI,OAAO,EAAA,EAAG;AAAA,YACpC,EAAE,MAAM,YAAA,EAAc,KAAA,EAAO,OAAO,IAAA,EAAM,EAAA,EAAI,OAAO,EAAA,EAAG;AAAA,YACxD,EAAE,IAAA,EAAM,QAAA,EAAU,cAAc,MAAA,EAAQ,KAAA,EAAO,OAAO,KAAA;AAAM,WAC9D;AAAA,QACF;AAEA,QAAA,OAAO;AAAA,UACL,QAAQ,sBAAA,CAAuB;AAAA,YAC7B,gBAAA,EAAkB,cAAA,CAAe,OAAA,GAC7B,cAAA,CAAe,gBAAA,GACf,CAAA;AAAA,YACJ,cAAA,EAAgB,cAAA,CAAe,OAAA,GAC3B,cAAA,CAAe,cAAA,GACf,CAAA;AAAA,YACJ;AAAA,WACD;AAAA,SACH;AAAA,MACF;AAEA,MAAA,MAAM,EAAE,MAAA,EAAQ,GAAG,IAAA,EAAK,GAAI,MAAM,QAAA,EAAS;AAE3C,MAAA,MAAM,eAA4C,EAAC;AAEnD,MAAA,MAAM,eAAA,GAAkB,IAAI,eAAA,CAG1B;AAAA,QACA,SAAA,CAAU,OAAO,UAAA,EAAY;AAC3B,UAAA,YAAA,CAAa,KAAK,KAAK,CAAA;AACvB,UAAA,UAAA,CAAW,QAAQ,KAAK,CAAA;AAAA,QAC1B,CAAA;AAAA,QACA,MAAM,KAAA,GAAQ;AACZ,UAAA,MAAM,EAAA,GAAK,MAAA,GAAS,GAAA,CAAI,WAAA,GAAc,MAAM,UAAU,CAAA;AACtD,UAAA,MAAM,YAAA;AAAA,YACJ,EAAA;AAAA,YACA,EAAE,aAAa,YAAA,EAAa;AAAA,YAC5B,SAAA;AAAA,YACA,UAAA;AAAA,YACA;AAAA,WACF;AAAA,QACF;AAAA,OACD,CAAA;AAED,MAAA,OAAO;AAAA,QACL,MAAA,EAAQ,MAAA,CAAO,WAAA,CAAY,eAAe,CAAA;AAAA,QAC1C,GAAG;AAAA,OACL;AAAA,IACF,CAAA;AAAA,IAEA,YAAA,EAAc,OAAO,EAAE,UAAA,EAAY,QAAO,KAAM;AAC9C,MAAA,MAAM,UAAA,GAAa,YAAY,MAAM,CAAA;AACrC,MAAA,MAAM,KAAA,GAAQ,WAAW,MAAM,CAAA;AAC/B,MAAA,MAAM,cAAc,MAAA,CAAO,MAAA,CAAO,KAAK,CAAA,CAAE,KAAK,GAAG,CAAA;AAEjD,MAAA,MAAM,EAAE,MAAA,EAAQ,SAAA,EAAW,UAAA,KAAe,MAAM,kBAAA;AAAA,QAC9C,UAAA;AAAA,QACA;AAAA,OACF;AAEA,MAAA,IAAI,MAAA,IAAU,cAAc,SAAA,EAAW;AACrC,QAAA,IAAI,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,oCAA+B,CAAA;AAEtD,QAAA,IAAI,MAAA,EAAQ,UAAU,SAAA,EAAW;AAC/B,UAAA,MAAA,CAAO,SAAS,SAAA,GAAY,IAAI,IAAA,CAAK,MAAA,CAAO,SAAS,SAAS,CAAA;AAAA,QAChE;AAGA,QAAA,OAAO,MAAA;AAAA,MACT;AAEA,MAAA,MAAM,MAAA,GAAS,MAAM,UAAA,EAAW;AAEhC,MAAA,MAAM,EAAA,GAAK,MAAA,GAAS,GAAA,CAAI,WAAA,GAAc,MAAM,UAAU,CAAA;AACtD,MAAA,MAAM,YAAA,CAAa,EAAA,EAAI,MAAA,EAAQ,SAAA,EAAW,YAAY,KAAK,CAAA;AAE3D,MAAA,OAAO,MAAA;AAAA,IACT;AAAA,GACF;AAEA,EAAA,OAAO;AAAA,IACL,UAAA,EAAY,OACV,OAAA,KAC0C;AAC1C,MAAA,MAAM,eAAe,iBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrB,OAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAO,UAAA,CAAW;AAAA,QAChB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,YAAA,EAAc,OAIZ,OAAA,KAC+C;AAC/C,MAAA,MAAM,eAAe,iBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrB,OAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAO,YAAA,CAAa;AAAA,QAClB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,cAAA,EAAgB,OACd,OAAA,KACqC;AACrC,MAAA,MAAM,eAAe,iBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrB,OAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAQ,MAAM,cAAA,CAAe;AAAA,QAC3B,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH,CAAA;AAAA,IAEA,YAAA,EAAc,OACZ,OAAA,KAC2C;AAC3C,MAAA,MAAM,eAAe,iBAAA,CAAkB;AAAA,QACrC,KAAA,EACE,OAAO,OAAA,CAAQ,KAAA,KAAU,WACrB,OAAA,CAAQ,OAAA,CAAQ,KAAK,CAAA,GACrB,OAAA,CAAQ,KAAA;AAAA,QACd,UAAA,EAAY;AAAA,OACb,CAAA;AAED,MAAA,OAAO,YAAA,CAAa;AAAA,QAClB,GAAG,OAAA;AAAA,QACH,KAAA,EAAO;AAAA,OACR,CAAA;AAAA,IACH;AAAA,GACF;AACF","file":"index.mjs","sourcesContent":["import { z } from \"zod\";\nimport \"dotenv/config\";\n\nexport const semanticCacheConfigSchema = z.object({\n  model: z.union([\n    z.string(),\n    z.custom((val) => {\n      return val && typeof val === \"object\";\n    }),\n  ]),\n  vector: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.VECTOR_REST_URL ?? \"\",\n      token: process.env.VECTOR_REST_TOKEN ?? \"\",\n    }),\n  redis: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.REDIS_REST_URL ?? \"\",\n      token: process.env.REDIS_REST_TOKEN ?? \"\",\n    }),\n  threshold: z.number().min(0).max(1).optional().default(0.92),\n  ttl: z\n    .number()\n    .positive()\n    .optional()\n    .default(60 * 60 * 24 * 14),\n  debug: z.boolean().optional().default(false),\n  cacheMode: z.enum([\"default\", \"refresh\"]).optional().default(\"default\"),\n  simulateStream: z\n    .object({\n      enabled: z.boolean().optional().default(true),\n      initialDelayInMs: z.number().min(0).optional().default(0),\n      chunkDelayInMs: z.number().min(0).optional().default(10),\n    })\n    .optional()\n    .default({\n      enabled: true,\n      initialDelayInMs: 0,\n      chunkDelayInMs: 10,\n    }),\n  useFullMessages: z.boolean().optional().default(false),\n});\n\nexport type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;\nexport type SemanticCacheConfigParsed = z.output<\n  typeof semanticCacheConfigSchema\n>;\n\nexport function validateEnvConfig(config: SemanticCacheConfigParsed) {\n  const errors: string[] = [];\n\n  if (!config.vector.url) {\n    errors.push(\n      \"Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.vector.token) {\n    errors.push(\n      \"Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (!config.redis.url) {\n    errors.push(\n      \"Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.redis.token) {\n    errors.push(\n      \"Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (errors.length > 0) {\n    throw new Error(\n      `Semantic Cache Configuration Error:\\n${errors.map((e) => `  - ${e}`).join(\"\\n\")}`,\n    );\n  }\n}\n","import {\n  type LanguageModelV2Middleware,\n  type LanguageModelV2StreamPart,\n  type LanguageModelV2Content,\n  type LanguageModelV2CallOptions,\n} from \"@ai-sdk/provider\";\nimport {\n  embed,\n  streamText,\n  generateText,\n  generateObject,\n  streamObject,\n  simulateReadableStream,\n  wrapLanguageModel,\n  type EmbeddingModel,\n  type StreamTextResult,\n  type GenerateTextResult,\n  type GenerateObjectResult,\n  type StreamObjectResult,\n  gateway,\n} from \"ai\";\n\nimport { Index } from \"@upstash/vector\";\nimport { Redis } from \"@upstash/redis\";\nimport * as crypto from \"node:crypto\";\nimport {\n  semanticCacheConfigSchema,\n  type SemanticCacheConfig,\n  validateEnvConfig,\n} from \"./schema\";\n\nconst norm = (s: string) => s.trim().toLowerCase().replace(/\\s+/g, \" \");\nconst sha = (s: string) => crypto.createHash(\"sha256\").update(s).digest(\"hex\");\n\nfunction canonicalizeMessages(msgs: any[]) {\n  return msgs.map((m) => ({\n    role: m.role,\n    content:\n      typeof m.content === \"string\" ? m.content : JSON.stringify(m.content),\n  }));\n}\n\nfunction buildScope(options: any) {\n  return {\n    llmModel: options.model?.modelId ?? String(options.model ?? \"\"),\n    systemHash: sha(options.system ?? \"\"),\n    params: sha(\n      JSON.stringify({\n        temperature: options.temperature,\n        topP: options.topP,\n      }),\n    ),\n    toolsHash: sha(JSON.stringify(options.tools ?? {})),\n  };\n}\n\nexport function createSemanticCache(config: SemanticCacheConfig) {\n  const parsed = semanticCacheConfigSchema.parse(config);\n\n  validateEnvConfig(parsed);\n\n  const {\n    vector,\n    redis: redisConfig,\n    threshold,\n    ttl,\n    simulateStream,\n    debug,\n    cacheMode,\n    useFullMessages,\n  } = parsed;\n  const model = parsed.model as EmbeddingModel<string>;\n\n  const index = new Index({\n    url: vector.url as string,\n    token: vector.token as string,\n  });\n\n  const redis = new Redis({\n    url: redisConfig.url as string,\n    token: redisConfig.token as string,\n  });\n\n  function getCacheKey(options: any): string {\n    if (options.messages) {\n      const messages = Array.isArray(options.messages) ? options.messages : [];\n\n      // By default, use only the last message to avoid token limit issues\n      if (!useFullMessages && messages.length > 0) {\n        const lastMessage = messages[messages.length - 1];\n        return JSON.stringify({\n          role: lastMessage.role,\n          content:\n            typeof lastMessage.content === \"string\"\n              ? lastMessage.content\n              : JSON.stringify(lastMessage.content),\n        });\n      }\n\n      return JSON.stringify(canonicalizeMessages(options.messages));\n    }\n    if (options.prompt) {\n      return String(options.prompt);\n    }\n    return \"\";\n  }\n\n  async function checkSemanticCache(cacheInput: string, scope: any) {\n    const promptNorm = norm(cacheInput);\n\n    const { embedding } = await embed({\n      model,\n      value: promptNorm,\n    });\n\n    const result = await index.query({\n      vector: embedding,\n      topK: 3,\n      includeMetadata: true,\n    });\n\n    const hit = result.find((m) => {\n      if (m.score < threshold) return false;\n\n      const metadata = m.metadata as any;\n      if (!metadata) return false;\n\n      return (\n        metadata.llmModel === scope.llmModel &&\n        metadata.systemHash === scope.systemHash &&\n        metadata.params === scope.params &&\n        metadata.toolsHash === scope.toolsHash\n      );\n    });\n\n    if (hit) {\n      const cached = await redis.get<{\n        streamParts?: LanguageModelV2StreamPart[];\n        text?: string;\n        content?: LanguageModelV2Content;\n        [key: string]: any;\n      }>(hit.id.toString());\n\n      if (cached) {\n        if (debug) console.log(\"✅ cache hit\", hit.score.toFixed(3));\n        return { cached, embedding, promptNorm };\n      }\n    }\n\n    if (debug) console.log(\"❌ miss -> generating…\");\n    return { cached: null, embedding, promptNorm };\n  }\n\n  async function storeInCache(\n    id: string,\n    data: any,\n    embedding: number[],\n    promptNorm: string,\n    scope: any,\n  ) {\n    const lockKey = \"lock:\" + id;\n    const ok = await redis.set(lockKey, \"1\", { nx: true, ex: 15 });\n\n    if (!ok) {\n      if (debug)\n        console.log(\"⚠️ Another process is writing to cache, skipping\");\n      return;\n    }\n\n    try {\n      await redis.set(id, data, { ex: ttl });\n      await index.upsert([\n        {\n          id,\n          vector: embedding,\n          metadata: {\n            prompt: promptNorm,\n            ...scope,\n          },\n        },\n      ]);\n    } finally {\n      await redis.del(lockKey);\n    }\n  }\n\n  const semanticCacheMiddleware: LanguageModelV2Middleware = {\n    wrapStream: async ({ doStream, params }) => {\n      const cacheInput = getCacheKey(params);\n      const scope = buildScope(params);\n      const promptScope = Object.values(scope).join(\"|\");\n\n      const { cached, embedding, promptNorm } = await checkSemanticCache(\n        cacheInput,\n        scope,\n      );\n\n      if (cached && cacheMode !== \"refresh\") {\n        if (debug) console.log(\"✅ Returning cached stream\");\n\n        let chunks: LanguageModelV2StreamPart[] = [];\n\n        if (cached.streamParts) {\n          chunks = cached.streamParts.map((p: any) => {\n            if (p.type === \"response-metadata\" && p.timestamp) {\n              return { ...p, timestamp: new Date(p.timestamp) };\n            }\n            return p;\n          });\n        } else if (cached.text) {\n          chunks = [\n            { type: \"text-start\", id: cached.id },\n            { type: \"text-delta\", delta: cached.text, id: cached.id },\n            { type: \"finish\", finishReason: \"stop\", usage: cached.usage },\n          ];\n        }\n\n        return {\n          stream: simulateReadableStream({\n            initialDelayInMs: simulateStream.enabled\n              ? simulateStream.initialDelayInMs\n              : 0,\n            chunkDelayInMs: simulateStream.enabled\n              ? simulateStream.chunkDelayInMs\n              : 0,\n            chunks,\n          }),\n        };\n      }\n\n      const { stream, ...rest } = await doStream();\n\n      const fullResponse: LanguageModelV2StreamPart[] = [];\n\n      const transformStream = new TransformStream<\n        LanguageModelV2StreamPart,\n        LanguageModelV2StreamPart\n      >({\n        transform(chunk, controller) {\n          fullResponse.push(chunk);\n          controller.enqueue(chunk);\n        },\n        async flush() {\n          const id = \"llm:\" + sha(promptScope + \"|\" + promptNorm);\n          await storeInCache(\n            id,\n            { streamParts: fullResponse },\n            embedding,\n            promptNorm,\n            scope,\n          );\n        },\n      });\n\n      return {\n        stream: stream.pipeThrough(transformStream),\n        ...rest,\n      };\n    },\n\n    wrapGenerate: async ({ doGenerate, params }) => {\n      const cacheInput = getCacheKey(params);\n      const scope = buildScope(params);\n      const promptScope = Object.values(scope).join(\"|\");\n\n      const { cached, embedding, promptNorm } = await checkSemanticCache(\n        cacheInput,\n        scope,\n      );\n\n      if (cached && cacheMode !== \"refresh\") {\n        if (debug) console.log(\"✅ Returning cached generation\");\n\n        if (cached?.response?.timestamp) {\n          cached.response.timestamp = new Date(cached.response.timestamp);\n        }\n\n        type GenReturn<T> = T extends () => PromiseLike<infer R> ? R : never;\n        return cached as unknown as GenReturn<typeof doGenerate>;\n      }\n\n      const result = await doGenerate();\n\n      const id = \"llm:\" + sha(promptScope + \"|\" + promptNorm);\n      await storeInCache(id, result, embedding, promptNorm, scope);\n\n      return result;\n    },\n  };\n\n  return {\n    streamText: async <TOOLS extends Record<string, any> = {}>(\n      options: Parameters<typeof streamText<TOOLS>>[0],\n    ): Promise<StreamTextResult<TOOLS, any>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return streamText({\n        ...options,\n        model: wrappedModel,\n      });\n    },\n\n    generateText: async <\n      TOOLS extends Record<string, any> = {},\n      OUTPUT = undefined,\n    >(\n      options: Parameters<typeof generateText<TOOLS, OUTPUT>>[0],\n    ): Promise<GenerateTextResult<TOOLS, OUTPUT>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return generateText({\n        ...options,\n        model: wrappedModel,\n      });\n    },\n\n    generateObject: async <T = any>(\n      options: Parameters<typeof generateObject>[0],\n    ): Promise<GenerateObjectResult<T>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return (await generateObject({\n        ...options,\n        model: wrappedModel,\n      })) as GenerateObjectResult<T>;\n    },\n\n    streamObject: async <T = any>(\n      options: Parameters<typeof streamObject>[0],\n    ): Promise<StreamObjectResult<T, T, any>> => {\n      const wrappedModel = wrapLanguageModel({\n        model:\n          typeof options.model === \"string\"\n            ? gateway(options.model)\n            : options.model,\n        middleware: semanticCacheMiddleware,\n      });\n\n      return streamObject({\n        ...options,\n        model: wrappedModel,\n      }) as unknown as StreamObjectResult<T, T, any>;\n    },\n  };\n}\n"]}

package/dist/schema.d.mts

@@ -0,0 +1,31 @@
+import { z } from 'zod';
+
+declare const semanticCacheConfigSchema: z.ZodObject<{
+    model: z.ZodUnion<readonly [z.ZodString, z.ZodCustom<unknown, unknown>]>;
+    vector: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        url: z.ZodURL;
+        token: z.ZodString;
+    }, z.core.$strip>>>;
+    redis: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        url: z.ZodURL;
+        token: z.ZodString;
+    }, z.core.$strip>>>;
+    threshold: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    ttl: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    debug: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+    cacheMode: z.ZodDefault<z.ZodOptional<z.ZodEnum<{
+        default: "default";
+        refresh: "refresh";
+    }>>>;
+    simulateStream: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        enabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+        initialDelayInMs: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+        chunkDelayInMs: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    }, z.core.$strip>>>;
+    useFullMessages: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+}, z.core.$strip>;
+type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;
+type SemanticCacheConfigParsed = z.output<typeof semanticCacheConfigSchema>;
+declare function validateEnvConfig(config: SemanticCacheConfigParsed): void;
+
+export { type SemanticCacheConfig, type SemanticCacheConfigParsed, semanticCacheConfigSchema, validateEnvConfig };

package/dist/schema.d.ts

@@ -0,0 +1,31 @@
+import { z } from 'zod';
+
+declare const semanticCacheConfigSchema: z.ZodObject<{
+    model: z.ZodUnion<readonly [z.ZodString, z.ZodCustom<unknown, unknown>]>;
+    vector: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        url: z.ZodURL;
+        token: z.ZodString;
+    }, z.core.$strip>>>;
+    redis: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        url: z.ZodURL;
+        token: z.ZodString;
+    }, z.core.$strip>>>;
+    threshold: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    ttl: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    debug: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+    cacheMode: z.ZodDefault<z.ZodOptional<z.ZodEnum<{
+        default: "default";
+        refresh: "refresh";
+    }>>>;
+    simulateStream: z.ZodDefault<z.ZodOptional<z.ZodObject<{
+        enabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+        initialDelayInMs: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+        chunkDelayInMs: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+    }, z.core.$strip>>>;
+    useFullMessages: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
+}, z.core.$strip>;
+type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;
+type SemanticCacheConfigParsed = z.output<typeof semanticCacheConfigSchema>;
+declare function validateEnvConfig(config: SemanticCacheConfigParsed): void;
+
+export { type SemanticCacheConfig, type SemanticCacheConfigParsed, semanticCacheConfigSchema, validateEnvConfig };

package/dist/schema.js

@@ -0,0 +1,76 @@
+'use strict';
+
+var zod = require('zod');
+require('dotenv/config');
+
+// schema.ts
+var semanticCacheConfigSchema = zod.z.object({
+  model: zod.z.union([
+    zod.z.string(),
+    zod.z.custom((val) => {
+      return val && typeof val === "object";
+    })
+  ]),
+  vector: zod.z.object({
+    url: zod.z.url(),
+    token: zod.z.string().min(1)
+  }).optional().default({
+    url: process.env.VECTOR_REST_URL ?? "",
+    token: process.env.VECTOR_REST_TOKEN ?? ""
+  }),
+  redis: zod.z.object({
+    url: zod.z.url(),
+    token: zod.z.string().min(1)
+  }).optional().default({
+    url: process.env.REDIS_REST_URL ?? "",
+    token: process.env.REDIS_REST_TOKEN ?? ""
+  }),
+  threshold: zod.z.number().min(0).max(1).optional().default(0.92),
+  ttl: zod.z.number().positive().optional().default(60 * 60 * 24 * 14),
+  debug: zod.z.boolean().optional().default(false),
+  cacheMode: zod.z.enum(["default", "refresh"]).optional().default("default"),
+  simulateStream: zod.z.object({
+    enabled: zod.z.boolean().optional().default(true),
+    initialDelayInMs: zod.z.number().min(0).optional().default(0),
+    chunkDelayInMs: zod.z.number().min(0).optional().default(10)
+  }).optional().default({
+    enabled: true,
+    initialDelayInMs: 0,
+    chunkDelayInMs: 10
+  }),
+  useFullMessages: zod.z.boolean().optional().default(false)
+});
+function validateEnvConfig(config) {
+  const errors = [];
+  if (!config.vector.url) {
+    errors.push(
+      "Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable."
+    );
+  }
+  if (!config.vector.token) {
+    errors.push(
+      "Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable."
+    );
+  }
+  if (!config.redis.url) {
+    errors.push(
+      "Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable."
+    );
+  }
+  if (!config.redis.token) {
+    errors.push(
+      "Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable."
+    );
+  }
+  if (errors.length > 0) {
+    throw new Error(
+      `Semantic Cache Configuration Error:
+${errors.map((e) => `  - ${e}`).join("\n")}`
+    );
+  }
+}
+
+exports.semanticCacheConfigSchema = semanticCacheConfigSchema;
+exports.validateEnvConfig = validateEnvConfig;
+//# sourceMappingURL=schema.js.map
+//# sourceMappingURL=schema.js.map

package/dist/schema.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../schema.ts"],"names":["z"],"mappings":";;;;;;AAGO,IAAM,yBAAA,GAA4BA,MAAE,MAAA,CAAO;AAAA,EAChD,KAAA,EAAOA,MAAE,KAAA,CAAM;AAAA,IACbA,MAAE,MAAA,EAAO;AAAA,IACTA,KAAA,CAAE,MAAA,CAAO,CAAC,GAAA,KAAQ;AAChB,MAAA,OAAO,GAAA,IAAO,OAAO,GAAA,KAAQ,QAAA;AAAA,IAC/B,CAAC;AAAA,GACF,CAAA;AAAA,EACD,MAAA,EAAQA,MACL,MAAA,CAAO;AAAA,IACN,GAAA,EAAKA,MAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAOA,KAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,eAAA,IAAmB,EAAA;AAAA,IACpC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iBAAA,IAAqB;AAAA,GACzC,CAAA;AAAA,EACH,KAAA,EAAOA,MACJ,MAAA,CAAO;AAAA,IACN,GAAA,EAAKA,MAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAOA,KAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,cAAA,IAAkB,EAAA;AAAA,IACnC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gBAAA,IAAoB;AAAA,GACxC,CAAA;AAAA,EACH,SAAA,EAAWA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,EAC3D,GAAA,EAAKA,KAAA,CACF,MAAA,EAAO,CACP,QAAA,EAAS,CACT,QAAA,EAAS,CACT,OAAA,CAAQ,EAAA,GAAK,EAAA,GAAK,EAAA,GAAK,EAAE,CAAA;AAAA,EAC5B,OAAOA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK,CAAA;AAAA,EAC3C,SAAA,EAAWA,KAAA,CAAE,IAAA,CAAK,CAAC,SAAA,EAAW,SAAS,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,SAAS,CAAA;AAAA,EACtE,cAAA,EAAgBA,MACb,MAAA,CAAO;AAAA,IACN,SAASA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,IAC5C,gBAAA,EAAkBA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,CAAC,CAAA;AAAA,IACxD,cAAA,EAAgBA,KAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,EAAE;AAAA,GACxD,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,OAAA,EAAS,IAAA;AAAA,IACT,gBAAA,EAAkB,CAAA;AAAA,IAClB,cAAA,EAAgB;AAAA,GACjB,CAAA;AAAA,EACH,iBAAiBA,KAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK;AACvD,CAAC;AAOM,SAAS,kBAAkB,MAAA,EAAmC;AACnE,EAAA,MAAM,SAAmB,EAAC;AAE1B,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,GAAA,EAAK;AACtB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,KAAA,EAAO;AACxB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,GAAA,EAAK;AACrB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,KAAA,EAAO;AACvB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,MAAA,CAAO,SAAS,CAAA,EAAG;AACrB,IAAA,MAAM,IAAI,KAAA;AAAA,MACR,CAAA;AAAA,EAAwC,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA,KAAM,CAAA,IAAA,EAAO,CAAC,CAAA,CAAE,CAAA,CAAE,IAAA,CAAK,IAAI,CAAC,CAAA;AAAA,KAClF;AAAA,EACF;AACF","file":"schema.js","sourcesContent":["import { z } from \"zod\";\nimport \"dotenv/config\";\n\nexport const semanticCacheConfigSchema = z.object({\n  model: z.union([\n    z.string(),\n    z.custom((val) => {\n      return val && typeof val === \"object\";\n    }),\n  ]),\n  vector: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.VECTOR_REST_URL ?? \"\",\n      token: process.env.VECTOR_REST_TOKEN ?? \"\",\n    }),\n  redis: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.REDIS_REST_URL ?? \"\",\n      token: process.env.REDIS_REST_TOKEN ?? \"\",\n    }),\n  threshold: z.number().min(0).max(1).optional().default(0.92),\n  ttl: z\n    .number()\n    .positive()\n    .optional()\n    .default(60 * 60 * 24 * 14),\n  debug: z.boolean().optional().default(false),\n  cacheMode: z.enum([\"default\", \"refresh\"]).optional().default(\"default\"),\n  simulateStream: z\n    .object({\n      enabled: z.boolean().optional().default(true),\n      initialDelayInMs: z.number().min(0).optional().default(0),\n      chunkDelayInMs: z.number().min(0).optional().default(10),\n    })\n    .optional()\n    .default({\n      enabled: true,\n      initialDelayInMs: 0,\n      chunkDelayInMs: 10,\n    }),\n  useFullMessages: z.boolean().optional().default(false),\n});\n\nexport type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;\nexport type SemanticCacheConfigParsed = z.output<\n  typeof semanticCacheConfigSchema\n>;\n\nexport function validateEnvConfig(config: SemanticCacheConfigParsed) {\n  const errors: string[] = [];\n\n  if (!config.vector.url) {\n    errors.push(\n      \"Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.vector.token) {\n    errors.push(\n      \"Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (!config.redis.url) {\n    errors.push(\n      \"Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.redis.token) {\n    errors.push(\n      \"Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (errors.length > 0) {\n    throw new Error(\n      `Semantic Cache Configuration Error:\\n${errors.map((e) => `  - ${e}`).join(\"\\n\")}`,\n    );\n  }\n}\n"]}

package/dist/schema.mjs

@@ -0,0 +1,73 @@
+import { z } from 'zod';
+import 'dotenv/config';
+
+// schema.ts
+var semanticCacheConfigSchema = z.object({
+  model: z.union([
+    z.string(),
+    z.custom((val) => {
+      return val && typeof val === "object";
+    })
+  ]),
+  vector: z.object({
+    url: z.url(),
+    token: z.string().min(1)
+  }).optional().default({
+    url: process.env.VECTOR_REST_URL ?? "",
+    token: process.env.VECTOR_REST_TOKEN ?? ""
+  }),
+  redis: z.object({
+    url: z.url(),
+    token: z.string().min(1)
+  }).optional().default({
+    url: process.env.REDIS_REST_URL ?? "",
+    token: process.env.REDIS_REST_TOKEN ?? ""
+  }),
+  threshold: z.number().min(0).max(1).optional().default(0.92),
+  ttl: z.number().positive().optional().default(60 * 60 * 24 * 14),
+  debug: z.boolean().optional().default(false),
+  cacheMode: z.enum(["default", "refresh"]).optional().default("default"),
+  simulateStream: z.object({
+    enabled: z.boolean().optional().default(true),
+    initialDelayInMs: z.number().min(0).optional().default(0),
+    chunkDelayInMs: z.number().min(0).optional().default(10)
+  }).optional().default({
+    enabled: true,
+    initialDelayInMs: 0,
+    chunkDelayInMs: 10
+  }),
+  useFullMessages: z.boolean().optional().default(false)
+});
+function validateEnvConfig(config) {
+  const errors = [];
+  if (!config.vector.url) {
+    errors.push(
+      "Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable."
+    );
+  }
+  if (!config.vector.token) {
+    errors.push(
+      "Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable."
+    );
+  }
+  if (!config.redis.url) {
+    errors.push(
+      "Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable."
+    );
+  }
+  if (!config.redis.token) {
+    errors.push(
+      "Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable."
+    );
+  }
+  if (errors.length > 0) {
+    throw new Error(
+      `Semantic Cache Configuration Error:
+${errors.map((e) => `  - ${e}`).join("\n")}`
+    );
+  }
+}
+
+export { semanticCacheConfigSchema, validateEnvConfig };
+//# sourceMappingURL=schema.mjs.map
+//# sourceMappingURL=schema.mjs.map

package/dist/schema.mjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../schema.ts"],"names":[],"mappings":";;;;AAGO,IAAM,yBAAA,GAA4B,EAAE,MAAA,CAAO;AAAA,EAChD,KAAA,EAAO,EAAE,KAAA,CAAM;AAAA,IACb,EAAE,MAAA,EAAO;AAAA,IACT,CAAA,CAAE,MAAA,CAAO,CAAC,GAAA,KAAQ;AAChB,MAAA,OAAO,GAAA,IAAO,OAAO,GAAA,KAAQ,QAAA;AAAA,IAC/B,CAAC;AAAA,GACF,CAAA;AAAA,EACD,MAAA,EAAQ,EACL,MAAA,CAAO;AAAA,IACN,GAAA,EAAK,EAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAO,CAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,eAAA,IAAmB,EAAA;AAAA,IACpC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,iBAAA,IAAqB;AAAA,GACzC,CAAA;AAAA,EACH,KAAA,EAAO,EACJ,MAAA,CAAO;AAAA,IACN,GAAA,EAAK,EAAE,GAAA,EAAI;AAAA,IACX,KAAA,EAAO,CAAA,CAAE,MAAA,EAAO,CAAE,IAAI,CAAC;AAAA,GACxB,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,GAAA,EAAK,OAAA,CAAQ,GAAA,CAAI,cAAA,IAAkB,EAAA;AAAA,IACnC,KAAA,EAAO,OAAA,CAAQ,GAAA,CAAI,gBAAA,IAAoB;AAAA,GACxC,CAAA;AAAA,EACH,SAAA,EAAW,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,EAC3D,GAAA,EAAK,CAAA,CACF,MAAA,EAAO,CACP,QAAA,EAAS,CACT,QAAA,EAAS,CACT,OAAA,CAAQ,EAAA,GAAK,EAAA,GAAK,EAAA,GAAK,EAAE,CAAA;AAAA,EAC5B,OAAO,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK,CAAA;AAAA,EAC3C,SAAA,EAAW,CAAA,CAAE,IAAA,CAAK,CAAC,SAAA,EAAW,SAAS,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,SAAS,CAAA;AAAA,EACtE,cAAA,EAAgB,EACb,MAAA,CAAO;AAAA,IACN,SAAS,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,IAAI,CAAA;AAAA,IAC5C,gBAAA,EAAkB,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,CAAC,CAAA;AAAA,IACxD,cAAA,EAAgB,CAAA,CAAE,MAAA,EAAO,CAAE,GAAA,CAAI,CAAC,CAAA,CAAE,QAAA,EAAS,CAAE,OAAA,CAAQ,EAAE;AAAA,GACxD,CAAA,CACA,QAAA,EAAS,CACT,OAAA,CAAQ;AAAA,IACP,OAAA,EAAS,IAAA;AAAA,IACT,gBAAA,EAAkB,CAAA;AAAA,IAClB,cAAA,EAAgB;AAAA,GACjB,CAAA;AAAA,EACH,iBAAiB,CAAA,CAAE,OAAA,GAAU,QAAA,EAAS,CAAE,QAAQ,KAAK;AACvD,CAAC;AAOM,SAAS,kBAAkB,MAAA,EAAmC;AACnE,EAAA,MAAM,SAAmB,EAAC;AAE1B,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,GAAA,EAAK;AACtB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,KAAA,EAAO;AACxB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,GAAA,EAAK;AACrB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,CAAC,MAAA,CAAO,KAAA,CAAM,KAAA,EAAO;AACvB,IAAA,MAAA,CAAO,IAAA;AAAA,MACL;AAAA,KACF;AAAA,EACF;AAEA,EAAA,IAAI,MAAA,CAAO,SAAS,CAAA,EAAG;AACrB,IAAA,MAAM,IAAI,KAAA;AAAA,MACR,CAAA;AAAA,EAAwC,MAAA,CAAO,GAAA,CAAI,CAAC,CAAA,KAAM,CAAA,IAAA,EAAO,CAAC,CAAA,CAAE,CAAA,CAAE,IAAA,CAAK,IAAI,CAAC,CAAA;AAAA,KAClF;AAAA,EACF;AACF","file":"schema.mjs","sourcesContent":["import { z } from \"zod\";\nimport \"dotenv/config\";\n\nexport const semanticCacheConfigSchema = z.object({\n  model: z.union([\n    z.string(),\n    z.custom((val) => {\n      return val && typeof val === \"object\";\n    }),\n  ]),\n  vector: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.VECTOR_REST_URL ?? \"\",\n      token: process.env.VECTOR_REST_TOKEN ?? \"\",\n    }),\n  redis: z\n    .object({\n      url: z.url(),\n      token: z.string().min(1),\n    })\n    .optional()\n    .default({\n      url: process.env.REDIS_REST_URL ?? \"\",\n      token: process.env.REDIS_REST_TOKEN ?? \"\",\n    }),\n  threshold: z.number().min(0).max(1).optional().default(0.92),\n  ttl: z\n    .number()\n    .positive()\n    .optional()\n    .default(60 * 60 * 24 * 14),\n  debug: z.boolean().optional().default(false),\n  cacheMode: z.enum([\"default\", \"refresh\"]).optional().default(\"default\"),\n  simulateStream: z\n    .object({\n      enabled: z.boolean().optional().default(true),\n      initialDelayInMs: z.number().min(0).optional().default(0),\n      chunkDelayInMs: z.number().min(0).optional().default(10),\n    })\n    .optional()\n    .default({\n      enabled: true,\n      initialDelayInMs: 0,\n      chunkDelayInMs: 10,\n    }),\n  useFullMessages: z.boolean().optional().default(false),\n});\n\nexport type SemanticCacheConfig = z.input<typeof semanticCacheConfigSchema>;\nexport type SemanticCacheConfigParsed = z.output<\n  typeof semanticCacheConfigSchema\n>;\n\nexport function validateEnvConfig(config: SemanticCacheConfigParsed) {\n  const errors: string[] = [];\n\n  if (!config.vector.url) {\n    errors.push(\n      \"Vector URL is required. Provide 'vector.url' or set VECTOR_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.vector.token) {\n    errors.push(\n      \"Vector token is required. Provide 'vector.token' or set VECTOR_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (!config.redis.url) {\n    errors.push(\n      \"Redis URL is required. Provide 'redis.url' or set REDIS_REST_URL environment variable.\",\n    );\n  }\n\n  if (!config.redis.token) {\n    errors.push(\n      \"Redis token is required. Provide 'redis.token' or set REDIS_REST_TOKEN environment variable.\",\n    );\n  }\n\n  if (errors.length > 0) {\n    throw new Error(\n      `Semantic Cache Configuration Error:\\n${errors.map((e) => `  - ${e}`).join(\"\\n\")}`,\n    );\n  }\n}\n"]}

package/package.json

@@ -0,0 +1,65 @@
+{
+  "name": "ai-sdk-memory",
+  "version": "0.0.1",
+  "description": "Semantic caching layer for AI SDK. Automatically generates embeddings for prompts, stores them in Upstash Vector and Redis, and reuses semantically similar responses to reduce token costs.",
+  "main": "./dist/index.js",
+  "module": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js",
+      "require": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "prepublishOnly": "bun run build",
+    "test": "echo \"Error: no test specified\" && exit 1"
+  },
+  "keywords": [
+    "ai",
+    "ai-sdk",
+    "vercel",
+    "cache",
+    "semantic",
+    "semantic-cache",
+    "vector",
+    "upstash",
+    "redis",
+    "embedding",
+    "llm"
+  ],
+  "author": "",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/yourusername/ai-sdk-memory.git"
+  },
+  "bugs": {
+    "url": "https://github.com/yourusername/ai-sdk-memory/issues"
+  },
+  "homepage": "https://github.com/yourusername/ai-sdk-memory#readme",
+  "devDependencies": {
+    "@types/bun": "latest",
+    "@types/node": "^24.7.0",
+    "tsup": "^8.5.0",
+    "typescript": "^5.0.0",
+    "vitest": "^3.2.4"
+  },
+  "peerDependencies": {
+    "ai": "^5.0.0"
+  },
+  "dependencies": {
+    "@ai-sdk/provider": "^2.0.0",
+    "@upstash/redis": "^1.35.5",
+    "@upstash/vector": "^1.2.2",
+    "dotenv": "^17.2.3",
+    "zod": "4.1.12"
+  }
+}