elsium-ai 0.2.2 → 0.3.0

package/dist/index.js

@@ -491,7 +491,16 @@ function envBool(name, fallback) {
   const raw = getEnvVar(name);
   if (raw !== undefined) {
     const normalized = raw.toLowerCase();
-    return normalized === "true" || normalized === "1" || normalized === "yes";
+    if (normalized === "true" || normalized === "1" || normalized === "yes")
+      return true;
+    if (normalized === "false" || normalized === "0" || normalized === "no")
+      return false;
+    throw new ElsiumError({
+      code: "CONFIG_ERROR",
+      message: `Environment variable ${name} has unrecognized boolean value: ${raw}. Expected one of: true, false, 1, 0, yes, no`,
+      retryable: false,
+      metadata: { variable: name, value: raw }
+    });
   }
   if (fallback !== undefined)
     return fallback;
@@ -502,6 +511,243 @@ function envBool(name, fallback) {
     metadata: { variable: name }
   });
 }
+// ../core/src/schema.ts
+var log = createLogger();
+function zodDefKind(def) {
+  return typeof def.type === "string" ? def.type : def.typeName;
+}
+function zodObjectToJsonSchema(schema, convert) {
+  const shape = typeof schema.shape === "function" ? schema.shape() : schema.shape;
+  const properties = {};
+  const required = [];
+  for (const [key, value] of Object.entries(shape)) {
+    const fieldSchema = value;
+    properties[key] = convert(fieldSchema);
+    const fieldDef = fieldSchema._def;
+    const fieldKind = zodDefKind(fieldDef);
+    if (fieldKind !== "optional" && fieldKind !== "ZodOptional" && fieldKind !== "default" && fieldKind !== "ZodDefault") {
+      required.push(key);
+    }
+    if (fieldDef.description) {
+      properties[key].description = fieldDef.description;
+    }
+  }
+  return { type: "object", properties, required };
+}
+function zodToJsonSchema(schema) {
+  if (!("_def" in schema))
+    return { type: "object" };
+  const def = schema._def;
+  const kind = zodDefKind(def);
+  switch (kind) {
+    case "object":
+    case "ZodObject":
+      return zodObjectToJsonSchema(def, zodToJsonSchema);
+    case "string":
+    case "ZodString":
+      return { type: "string" };
+    case "number":
+    case "ZodNumber":
+      return { type: "number" };
+    case "boolean":
+    case "ZodBoolean":
+      return { type: "boolean" };
+    case "array":
+    case "ZodArray":
+      return {
+        type: "array",
+        items: zodToJsonSchema(def.element ?? def.type)
+      };
+    case "enum":
+    case "ZodEnum": {
+      const values = def.values ?? (def.entries ? Object.values(def.entries) : []);
+      return { type: "string", enum: values };
+    }
+    case "optional":
+    case "ZodOptional":
+      return zodToJsonSchema(def.innerType);
+    case "default":
+    case "ZodDefault":
+      return zodToJsonSchema(def.innerType);
+    case "nullable":
+    case "ZodNullable": {
+      const inner = zodToJsonSchema(def.innerType);
+      return { ...inner, nullable: true };
+    }
+    case "ZodLiteral":
+      return { type: typeof def.value, const: def.value };
+    case "ZodUnion": {
+      const options = def.options.map(zodToJsonSchema);
+      return { anyOf: options };
+    }
+    case "ZodRecord":
+      return {
+        type: "object",
+        additionalProperties: def.valueType ? zodToJsonSchema(def.valueType) : { type: "string" }
+      };
+    case "ZodTuple": {
+      const items = (def.items ?? []).map(zodToJsonSchema);
+      return { type: "array", prefixItems: items, minItems: items.length, maxItems: items.length };
+    }
+    case "ZodDate":
+      return { type: "string", format: "date-time" };
+    default:
+      log.warn(`zodToJsonSchema: unsupported type ${kind}, defaulting to string`);
+      return { type: "string" };
+  }
+}
+// ../core/src/registry.ts
+var log2 = createLogger();
+var BLOCKED_KEYS = new Set(["__proto__", "constructor", "prototype"]);
+function createRegistry(label) {
+  const entries = new Map;
+  return {
+    register(name, factory) {
+      if (BLOCKED_KEYS.has(name)) {
+        log2.warn(`Registry(${label}): rejected blocked key "${name}"`);
+        return;
+      }
+      entries.set(name, factory);
+      log2.debug(`Registry(${label}): registered "${name}"`);
+    },
+    get(name) {
+      if (BLOCKED_KEYS.has(name))
+        return;
+      return entries.get(name);
+    },
+    list() {
+      return Array.from(entries.keys());
+    },
+    has(name) {
+      if (BLOCKED_KEYS.has(name))
+        return false;
+      return entries.has(name);
+    },
+    unregister(name) {
+      if (BLOCKED_KEYS.has(name))
+        return false;
+      const deleted = entries.delete(name);
+      if (deleted) {
+        log2.debug(`Registry(${label}): unregistered "${name}"`);
+      }
+      return deleted;
+    }
+  };
+}
+// ../core/src/tokens.ts
+var MODEL_RATIOS = {
+  cl100k_base: 4,
+  "gpt-4o": 4,
+  "gpt-4o-mini": 4,
+  o1: 4,
+  "o1-mini": 4,
+  "o3-mini": 4,
+  "claude-opus-4-6": 3.5,
+  "claude-sonnet-4-6": 3.5,
+  "claude-haiku-4-5-20251001": 3.5,
+  "gemini-2.0-flash": 4,
+  "gemini-2.0-flash-lite": 4,
+  "gemini-2.5-pro-preview-05-06": 4,
+  "gemini-2.5-flash-preview-04-17": 4
+};
+function getRatio(model) {
+  if (!model)
+    return 4;
+  if (MODEL_RATIOS[model])
+    return MODEL_RATIOS[model];
+  if (model.startsWith("claude"))
+    return 3.5;
+  if (model.startsWith("gemini"))
+    return 4;
+  if (model.startsWith("gpt") || model.startsWith("o1") || model.startsWith("o3"))
+    return 4;
+  return 4;
+}
+function extractMessageText(msg) {
+  if (typeof msg.content === "string")
+    return msg.content;
+  return msg.content.filter((p) => p.type === "text" && ("text" in p)).map((p) => p.text).join("");
+}
+function countTokens(text, model) {
+  const ratio = getRatio(model);
+  return Math.ceil(text.length / ratio) + 4;
+}
+function createContextManager(config) {
+  const { maxTokens, strategy, reserveTokens = 0 } = config;
+  const budget = maxTokens - reserveTokens;
+  function estimateMessageTokens(msg) {
+    return countTokens(extractMessageText(msg)) + 4;
+  }
+  function estimateTokens(messages) {
+    return messages.reduce((sum, msg) => sum + estimateMessageTokens(msg), 0);
+  }
+  async function fitTruncate(messages, system) {
+    let available = budget;
+    if (system)
+      available -= countTokens(system);
+    const result = [];
+    for (let i = messages.length - 1;i >= 0; i--) {
+      const tokens = estimateMessageTokens(messages[i]);
+      if (available - tokens < 0 && result.length > 0)
+        break;
+      available -= tokens;
+      result.unshift(messages[i]);
+    }
+    return result;
+  }
+  async function fitSummarize(messages, system) {
+    if (!config.summarizer)
+      return fitTruncate(messages, system);
+    let available = budget;
+    if (system)
+      available -= countTokens(system);
+    const total = estimateTokens(messages);
+    if (total <= available)
+      return messages;
+    const keepCount = Math.max(1, Math.floor(messages.length / 3));
+    const toSummarize = messages.slice(0, messages.length - keepCount);
+    const toKeep = messages.slice(messages.length - keepCount);
+    const summary = await config.summarizer(toSummarize);
+    const summaryMsg = {
+      role: "system",
+      content: `Previous conversation summary: ${summary}`
+    };
+    return [summaryMsg, ...toKeep];
+  }
+  async function fitSlidingWindow(messages, system) {
+    let available = budget;
+    if (system)
+      available -= countTokens(system);
+    const result = [];
+    for (let i = messages.length - 1;i >= 0; i--) {
+      const tokens = estimateMessageTokens(messages[i]);
+      if (available - tokens < 0 && result.length > 0)
+        break;
+      available -= tokens;
+      result.unshift(messages[i]);
+    }
+    return result;
+  }
+  return {
+    estimateTokens,
+    async fit(messages, system) {
+      const total = estimateTokens(messages);
+      let available = budget;
+      if (system)
+        available -= countTokens(system);
+      if (total <= available)
+        return messages;
+      switch (strategy) {
+        case "truncate":
+          return fitTruncate(messages, system);
+        case "summarize":
+          return fitSummarize(messages, system);
+        case "sliding-window":
+          return fitSlidingWindow(messages, system);
+      }
+    }
+  };
+}
 // ../core/src/circuit-breaker.ts
 function defaultShouldCount(error) {
   if (error && typeof error === "object" && "retryable" in error) {
@@ -630,6 +876,101 @@ function createCircuitBreaker(config) {
     }
   };
 }
+// ../core/src/shutdown.ts
+function createShutdownManager(config) {
+  const drainTimeoutMs = config?.drainTimeoutMs ?? 30000;
+  const signals = config?.signals ?? ["SIGTERM", "SIGINT"];
+  if (drainTimeoutMs < 0 || !Number.isFinite(drainTimeoutMs)) {
+    throw new ElsiumError({
+      code: "CONFIG_ERROR",
+      message: "drainTimeoutMs must be >= 0 and finite",
+      retryable: false
+    });
+  }
+  let shuttingDown = false;
+  let inFlightCount = 0;
+  let drainResolve = null;
+  let shutdownPromise = null;
+  const signalHandlers = [];
+  function checkDrained() {
+    if (inFlightCount === 0 && drainResolve) {
+      drainResolve();
+      drainResolve = null;
+    }
+  }
+  async function shutdown() {
+    if (shutdownPromise)
+      return shutdownPromise;
+    shuttingDown = true;
+    shutdownPromise = (async () => {
+      config?.onDrainStart?.();
+      if (inFlightCount === 0) {
+        config?.onDrainComplete?.();
+        return;
+      }
+      const drainPromise = new Promise((resolve) => {
+        drainResolve = resolve;
+      });
+      let drainTimer;
+      const timeoutPromise = new Promise((resolve) => {
+        drainTimer = setTimeout(() => resolve("timeout"), drainTimeoutMs);
+      });
+      const result = await Promise.race([
+        drainPromise.then(() => "drained"),
+        timeoutPromise
+      ]);
+      if (drainTimer !== undefined)
+        clearTimeout(drainTimer);
+      if (result === "timeout") {
+        config?.onForceShutdown?.();
+      } else {
+        config?.onDrainComplete?.();
+      }
+    })();
+    return shutdownPromise;
+  }
+  const manager = {
+    get inFlight() {
+      return inFlightCount;
+    },
+    get isShuttingDown() {
+      return shuttingDown;
+    },
+    async trackOperation(fn) {
+      if (shuttingDown) {
+        throw new ElsiumError({
+          code: "VALIDATION_ERROR",
+          message: "Server is shutting down, not accepting new operations",
+          retryable: true
+        });
+      }
+      inFlightCount++;
+      try {
+        return await fn();
+      } finally {
+        inFlightCount--;
+        checkDrained();
+      }
+    },
+    shutdown,
+    dispose() {
+      for (const { signal, handler } of signalHandlers) {
+        process.removeListener(signal, handler);
+      }
+      signalHandlers.length = 0;
+    }
+  };
+  if (typeof process !== "undefined" && process.on) {
+    for (const signal of signals) {
+      const handler = () => {
+        manager.shutdown();
+      };
+      signalHandlers.push({ signal, handler });
+      process.on(signal, handler);
+    }
+  }
+  return manager;
+}
 // ../gateway/src/provider.ts
 var providerRegistry = new Map;
 var metadataRegistry = new Map;
@@ -659,16 +1000,16 @@ function composeMiddleware(middlewares) {
   };
 }
 function loggingMiddleware(logger) {
-  const log = logger ?? createLogger({ level: "info" });
+  const log3 = logger ?? createLogger({ level: "info" });
   return async (ctx, next) => {
-    log.info("LLM request", {
+    log3.info("LLM request", {
       provider: ctx.provider,
       model: ctx.model,
       traceId: ctx.traceId,
       messageCount: ctx.request.messages.length
     });
     const response = await next(ctx);
-    log.info("LLM response", {
+    log3.info("LLM response", {
       provider: ctx.provider,
       model: ctx.model,
       traceId: ctx.traceId,
@@ -804,7 +1145,7 @@ function xrayMiddleware(options = {}) {
 }
 
 // ../gateway/src/pricing.ts
-var log = createLogger();
+var log3 = createLogger();
 var PRICING = {
   "claude-opus-4-6": { inputPerMillion: 15, outputPerMillion: 75 },
   "claude-sonnet-4-6": { inputPerMillion: 3, outputPerMillion: 15 },
@@ -842,7 +1183,7 @@ function resolveModelName(model) {
 function calculateCost(model, usage) {
   const pricing = PRICING[resolveModelName(model)];
   if (!pricing) {
-    log.warn(`Unknown model "${model}" — cost will be reported as $0. Register pricing with registerPricing().`);
+    log3.warn(`Unknown model "${model}" — cost will be reported as $0. Register pricing with registerPricing().`);
     return {
       inputCost: 0,
       outputCost: 0,
@@ -936,15 +1277,33 @@ function createAnthropicProvider(config) {
     if (part.type === "text")
       return { type: "text", text: part.text };
     if (part.type === "image" && part.source?.type === "base64") {
+      const src = part.source;
       return {
         type: "image",
         source: {
           type: "base64",
-          media_type: part.source.mediaType,
-          data: part.source.data
+          media_type: src.mediaType,
+          data: src.data
         }
       };
     }
+    if (part.type === "document" && part.source) {
+      if (part.source.type === "base64") {
+        const src = part.source;
+        return {
+          type: "document",
+          source: {
+            type: "base64",
+            media_type: src.mediaType,
+            data: src.data
+          }
+        };
+      }
+      return { type: "text", text: "[document: url source not supported by Anthropic]" };
+    }
+    if (part.type === "audio") {
+      return { type: "text", text: "[audio content not supported by this provider]" };
+    }
     return { type: "text", text: "[unsupported content]" };
   }
   function formatMultipartContent(msg, role) {
@@ -1053,6 +1412,16 @@ function createAnthropicProvider(config) {
       const tools = formatTools(req.tools);
       if (tools)
         body.tools = tools;
+      if (req.schema) {
+        const jsonSchema = zodToJsonSchema(req.schema);
+        const structuredTool = {
+          name: "_structured_output",
+          description: "Return structured output matching the required schema",
+          input_schema: jsonSchema
+        };
+        body.tools = [...tools ?? [], structuredTool];
+        body.tool_choice = { type: "tool", name: "_structured_output" };
+      }
       const startTime = performance.now();
       const raw = await retry(async () => {
         const controller = new AbortController;
@@ -1262,7 +1631,31 @@ function createGoogleProvider(config) {
     return { role, parts };
   }
   function formatGeminiMultipartContent(msg, role) {
-    const parts = msg.content.filter((p) => p.type === "text").map((p) => ({ text: p.text }));
+    const parts = [];
+    for (const p of msg.content) {
+      if (p.type === "text") {
+        parts.push({ text: p.text });
+      } else if (p.type === "image") {
+        const img = p;
+        if (img.source.type === "base64") {
+          parts.push({ inlineData: { mimeType: img.source.mediaType, data: img.source.data } });
+        } else {
+          parts.push({ fileData: { mimeType: "image/jpeg", fileUri: img.source.url } });
+        }
+      } else if (p.type === "audio" || p.type === "document") {
+        const media = p;
+        if (media.source.type === "base64") {
+          parts.push({
+            inlineData: { mimeType: media.source.mediaType, data: media.source.data }
+          });
+        } else {
+          const urlSource = media.source;
+          parts.push({
+            fileData: { mimeType: "application/octet-stream", fileUri: urlSource.url }
+          });
+        }
+      }
+    }
     return { role, parts };
   }
   function formatMessages(messages) {
@@ -1359,6 +1752,10 @@ function createGoogleProvider(config) {
       config2.topP = req.topP;
     if (req.stopSequences?.length)
       config2.stopSequences = req.stopSequences;
+    if (req.schema) {
+      config2.responseMimeType = "application/json";
+      config2.responseSchema = zodToJsonSchema(req.schema);
+    }
     return config2;
   }
   function buildRequestBody(req) {
@@ -1437,7 +1834,8 @@ async function handleGoogleErrorResponse(response) {
     throw ElsiumError.authError("google");
   }
   if (response.status === 429) {
-    throw ElsiumError.rateLimit("google");
+    const retryAfter = response.headers.get("retry-after");
+    throw ElsiumError.rateLimit("google", retryAfter ? Number.parseInt(retryAfter) * 1000 : undefined);
   }
   throw ElsiumError.providerError(`Google API error ${response.status}: ${errorBody}`, {
     provider: "google",
@@ -1623,6 +2021,43 @@ function createOpenAIProvider(config) {
     }
     return openaiMsg;
   }
+  function formatUserContent(msg) {
+    if (typeof msg.content === "string")
+      return msg.content;
+    const parts = [];
+    for (const part of msg.content) {
+      if (part.type === "text") {
+        parts.push({ type: "text", text: part.text });
+      } else if (part.type === "image") {
+        if (part.source.type === "base64") {
+          const url = `data:${part.source.mediaType};base64,${part.source.data}`;
+          parts.push({ type: "image_url", image_url: { url } });
+        } else {
+          parts.push({ type: "image_url", image_url: { url: part.source.url } });
+        }
+      } else if (part.type === "audio") {
+        if (part.source.type === "base64") {
+          const format = part.source.mediaType.split("/")[1] ?? "wav";
+          parts.push({
+            type: "input_audio",
+            input_audio: { data: part.source.data, format }
+          });
+        } else {
+          parts.push({ type: "text", text: "[audio: url source requires file upload]" });
+        }
+      } else if (part.type === "document") {
+        if (part.source.type === "base64") {
+          parts.push({
+            type: "text",
+            text: `[document: ${part.source.mediaType} content attached as base64]`
+          });
+        } else {
+          parts.push({ type: "text", text: `[document: ${part.source.url}]` });
+        }
+      }
+    }
+    return parts;
+  }
   function formatMessages(messages) {
     const formatted = [];
     for (const msg of messages) {
@@ -1638,7 +2073,7 @@ function createOpenAIProvider(config) {
         formatted.push(formatAssistantMessage(msg));
         continue;
       }
-      formatted.push({ role: "user", content: extractTextContent(msg) });
+      formatted.push({ role: "user", content: formatUserContent(msg) });
     }
     return formatted;
   }
@@ -1715,6 +2150,17 @@ function createOpenAIProvider(config) {
       const tools = formatTools(req.tools);
       if (tools)
         body.tools = tools;
+      if (req.schema) {
+        const jsonSchema = zodToJsonSchema(req.schema);
+        body.response_format = {
+          type: "json_schema",
+          json_schema: {
+            name: "structured_output",
+            strict: true,
+            schema: jsonSchema
+          }
+        };
+      }
       const startTime = performance.now();
       const raw = await retry(async () => {
         const controller = new AbortController;
@@ -1877,7 +2323,7 @@ var PROVIDER_FACTORIES = {
 function registerProviderFactory(name, factory) {
   PROVIDER_FACTORIES[name] = factory;
 }
-function gateway(config) {
+function validateGatewayConfig(config) {
   const factory = PROVIDER_FACTORIES[config.provider];
   if (!factory) {
     throw new ElsiumError({
@@ -1886,21 +2332,92 @@ function gateway(config) {
       retryable: false
     });
   }
+  if (typeof config.apiKey !== "string" || config.apiKey.trim() === "") {
+    throw new ElsiumError({
+      code: "CONFIG_ERROR",
+      message: "apiKey must be a non-empty string",
+      retryable: false
+    });
+  }
+  if (config.timeout !== undefined && (!Number.isFinite(config.timeout) || config.timeout <= 0)) {
+    throw new ElsiumError({
+      code: "CONFIG_ERROR",
+      message: "timeout must be a positive finite number",
+      retryable: false
+    });
+  }
+  if (config.maxRetries !== undefined && (!Number.isFinite(config.maxRetries) || !Number.isInteger(config.maxRetries) || config.maxRetries < 0)) {
+    throw new ElsiumError({
+      code: "CONFIG_ERROR",
+      message: "maxRetries must be a non-negative finite integer",
+      retryable: false
+    });
+  }
+  return factory;
+}
+function autoRegisterProvider(provider) {
+  if (!provider.metadata)
+    return;
+  registerProviderMetadata(provider.name, provider.metadata);
+  if (!provider.metadata.pricing)
+    return;
+  for (const [model, pricing] of Object.entries(provider.metadata.pricing)) {
+    registerPricing(model, pricing);
+  }
+}
+function validateRequestLimits(request, maxMessages, maxInputTokens) {
+  if (request.messages.length > maxMessages) {
+    throw ElsiumError.validation(`Message count ${request.messages.length} exceeds limit of ${maxMessages}`);
+  }
+  let estimatedTokens = 0;
+  for (const msg of request.messages) {
+    const text = typeof msg.content === "string" ? msg.content : msg.content.map((p) => p.type === "text" ? p.text : "").join("");
+    estimatedTokens += Math.ceil(text.length / 4);
+  }
+  if (estimatedTokens > maxInputTokens) {
+    throw ElsiumError.validation(`Estimated input tokens (~${estimatedTokens}) exceeds limit of ${maxInputTokens}`);
+  }
+}
+function buildMiddlewareContext(req, providerName, defaultModel, metadata) {
+  return {
+    request: req,
+    provider: providerName,
+    model: req.model ?? defaultModel,
+    traceId: generateTraceId(),
+    startTime: performance.now(),
+    metadata
+  };
+}
+async function accumulateStreamEvents(stream, emit) {
+  let textContent = "";
+  let usage = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
+  let stopReason = "end_turn";
+  let id = "";
+  for await (const event of stream) {
+    emit(event);
+    if (event.type === "text_delta") {
+      textContent += event.text;
+    } else if (event.type === "message_end") {
+      usage = event.usage;
+      stopReason = event.stopReason;
+    } else if (event.type === "message_start") {
+      id = event.id;
+    }
+  }
+  return { textContent, usage, stopReason, id };
+}
+function gateway(config) {
+  const factory = validateGatewayConfig(config);
   const provider = factory({
     apiKey: config.apiKey,
     baseUrl: config.baseUrl,
     timeout: config.timeout,
     maxRetries: config.maxRetries
   });
-  if (provider.metadata) {
-    registerProviderMetadata(provider.name, provider.metadata);
-    if (provider.metadata.pricing) {
-      for (const [model, pricing] of Object.entries(provider.metadata.pricing)) {
-        registerPricing(model, pricing);
-      }
-    }
-  }
+  autoRegisterProvider(provider);
   const defaultModel = config.model ?? provider.defaultModel;
+  const maxMessages = config.maxMessages ?? 1000;
+  const maxInputTokens = config.maxInputTokens ?? 1e6;
   let xrayStore = null;
   const allMiddleware = [...config.middleware ?? []];
   if (config.xray) {
@@ -1915,14 +2432,7 @@ function gateway(config) {
     if (!composedMiddleware) {
       return provider.complete(req);
     }
-    const ctx = {
-      request: req,
-      provider: provider.name,
-      model: req.model ?? defaultModel,
-      traceId: generateTraceId(),
-      startTime: performance.now(),
-      metadata: request.metadata ?? {}
-    };
+    const ctx = buildMiddlewareContext(req, provider.name, defaultModel, request.metadata ?? {});
     return composedMiddleware(ctx, async (c) => provider.complete(c.request));
   }
   return {
@@ -1934,34 +2444,27 @@ function gateway(config) {
       return xrayStore?.callHistory(limit) ?? [];
     },
     async complete(request) {
+      validateRequestLimits(request, maxMessages, maxInputTokens);
       return executeWithMiddleware(request);
     },
     stream(request) {
+      validateRequestLimits(request, maxMessages, maxInputTokens);
       const req = { ...request, model: request.model ?? defaultModel };
       if (composedMiddleware) {
-        const ctx = {
-          request: req,
-          provider: provider.name,
-          model: req.model ?? defaultModel,
-          traceId: generateTraceId(),
-          startTime: performance.now(),
-          metadata: request.metadata ?? {}
-        };
+        const ctx = buildMiddlewareContext(req, provider.name, defaultModel, request.metadata ?? {});
         return createStream(async (emit) => {
           await composedMiddleware(ctx, async (c) => {
-            const stream = provider.stream(c.request);
-            for await (const event of stream) {
-              emit(event);
-            }
+            const result = await accumulateStreamEvents(provider.stream(c.request), emit);
+            const latencyMs = Math.round(performance.now() - ctx.startTime);
             return {
-              id: "",
-              message: { role: "assistant", content: "" },
-              usage: { inputTokens: 0, outputTokens: 0, totalTokens: 0 },
-              cost: { inputCost: 0, outputCost: 0, totalCost: 0, currency: "USD" },
+              id: result.id,
+              message: { role: "assistant", content: result.textContent },
+              usage: result.usage,
+              cost: calculateCost(c.model, result.usage),
               model: c.model,
               provider: provider.name,
-              stopReason: "end_turn",
-              latencyMs: 0,
+              stopReason: result.stopReason,
+              latencyMs,
               traceId: ctx.traceId
             };
           });
@@ -1971,107 +2474,46 @@ function gateway(config) {
     },
     async generate(request) {
       const { schema, ...rest } = request;
-      const jsonSchema = schemaToJsonSchema(schema);
-      const systemPrompt = [
-        rest.system ?? "",
-        "You MUST respond with valid JSON matching this schema:",
-        JSON.stringify(jsonSchema, null, 2),
-        "Respond ONLY with the JSON object, no markdown or explanation."
-      ].filter(Boolean).join(`
-
-`);
+      const jsonSchema = zodToJsonSchema(schema);
       const response = await executeWithMiddleware({
         ...rest,
-        system: systemPrompt
+        schema,
+        system: [
+          rest.system ?? "",
+          "You MUST respond with valid JSON matching this schema:",
+          JSON.stringify(jsonSchema, null, 2),
+          "Respond ONLY with the JSON object, no markdown or explanation."
+        ].filter(Boolean).join(`
+
+`)
       });
-      const text = typeof response.message.content === "string" ? response.message.content : "";
-      const jsonMatch = text.match(/\{[\s\S]*\}/);
-      if (!jsonMatch) {
-        throw ElsiumError.validation("LLM response did not contain valid JSON", {
-          response: text
-        });
+      let parsed;
+      if (response.stopReason === "tool_use" && response.message.toolCalls?.length) {
+        const structuredCall = response.message.toolCalls.find((tc) => tc.name === "_structured_output");
+        if (structuredCall) {
+          parsed = structuredCall.arguments;
+        }
+      }
+      if (parsed === undefined) {
+        const text = typeof response.message.content === "string" ? response.message.content : "";
+        const jsonMatch = text.match(/\{[\s\S]*\}/);
+        if (!jsonMatch) {
+          throw ElsiumError.validation("LLM response did not contain valid JSON", {
+            response: text
+          });
+        }
+        parsed = JSON.parse(jsonMatch[0]);
       }
-      const parsed = JSON.parse(jsonMatch[0]);
       const result = schema.safeParse(parsed);
       if (!result.success) {
         throw ElsiumError.validation("LLM response did not match schema", {
-          errors: result.error.issues,
-          response: text
+          errors: result.error.issues
         });
       }
       return { data: result.data, response };
     }
   };
 }
-function schemaToJsonSchema(schema) {
-  try {
-    if ("_def" in schema) {
-      const def = schema._def;
-      const result = convertZodDef(def);
-      if (result)
-        return result;
-    }
-  } catch {}
-  return { type: "string" };
-}
-function zodDefKind(def) {
-  return typeof def.type === "string" ? def.type : def.typeName;
-}
-function convertZodDef(def) {
-  const kind = zodDefKind(def);
-  switch (kind) {
-    case "object":
-    case "ZodObject":
-      return convertZodObject(def);
-    case "string":
-    case "ZodString":
-      return { type: "string" };
-    case "number":
-    case "ZodNumber":
-      return { type: "number" };
-    case "boolean":
-    case "ZodBoolean":
-      return { type: "boolean" };
-    case "array":
-    case "ZodArray":
-      return convertZodArray(def);
-    case "enum":
-    case "ZodEnum": {
-      const values = def.values ?? (def.entries ? Object.values(def.entries) : []);
-      return { type: "string", enum: values };
-    }
-    case "optional":
-    case "ZodOptional":
-      return convertZodOptional(def);
-    default:
-      return null;
-  }
-}
-function convertZodObject(def) {
-  if (!def.shape)
-    return null;
-  const shape = typeof def.shape === "function" ? def.shape() : def.shape;
-  const properties = {};
-  const required = [];
-  for (const [key, value] of Object.entries(shape)) {
-    properties[key] = schemaToJsonSchema(value);
-    const valDef = value._def;
-    const valKind = zodDefKind(valDef);
-    if (valKind !== "optional" && valKind !== "ZodOptional") {
-      required.push(key);
-    }
-  }
-  return { type: "object", properties, required };
-}
-function convertZodArray(def) {
-  return {
-    type: "array",
-    items: schemaToJsonSchema(def.element ?? def.type)
-  };
-}
-function convertZodOptional(def) {
-  return schemaToJsonSchema(def.innerType ?? def.innerType);
-}
 // ../gateway/src/security.ts
 var INJECTION_PATTERNS = [
   {
@@ -2245,75 +2687,396 @@ function redactSecrets(text, piiTypes) {
     redacted = piiResult.redacted;
     found.push(...piiResult.found);
   }
-  return { redacted, found };
-}
-function checkBlockedPatterns(text, patterns) {
+  return { redacted, found };
+}
+function checkBlockedPatterns(text, patterns) {
+  const violations = [];
+  for (const pattern of patterns) {
+    pattern.lastIndex = 0;
+    if (pattern.test(text)) {
+      violations.push({
+        type: "blocked_pattern",
+        detail: `Blocked pattern matched: ${pattern.source}`,
+        severity: "medium"
+      });
+    }
+  }
+  return violations;
+}
+function scanMessageForViolations(text, config) {
+  const violations = [];
+  if (config.promptInjection !== false) {
+    violations.push(...detectPromptInjection(text));
+  }
+  if (config.jailbreakDetection) {
+    violations.push(...detectJailbreak(text));
+  }
+  if (config.blockedPatterns?.length) {
+    violations.push(...checkBlockedPatterns(text, config.blockedPatterns));
+  }
+  return violations;
+}
+function reportAndThrow(violations, config) {
+  for (const v of violations) {
+    config.onViolation?.(v);
+  }
+  throw ElsiumError.validation(`Security violation detected: ${violations.map((v) => v.detail).join("; ")}`);
+}
+function redactResponseSecrets(response, config) {
+  const responseText = extractText(response.message.content);
+  if (!responseText)
+    return response;
+  const { redacted, found } = redactSecrets(responseText, config.piiTypes);
+  if (found.length === 0)
+    return response;
+  for (const v of found) {
+    config.onViolation?.(v);
+  }
+  return {
+    ...response,
+    message: {
+      ...response.message,
+      content: redacted
+    }
+  };
+}
+function securityMiddleware(config) {
+  return async (ctx, next) => {
+    if (ctx.request.system) {
+      const sysViolations = scanMessageForViolations(ctx.request.system, config);
+      if (sysViolations.length > 0) {
+        reportAndThrow(sysViolations, config);
+      }
+    }
+    for (const message of ctx.request.messages) {
+      const text = extractText(message.content);
+      if (!text)
+        continue;
+      const violations = scanMessageForViolations(text, config);
+      if (violations.length > 0) {
+        reportAndThrow(violations, config);
+      }
+    }
+    const response = await next(ctx);
+    if (config.secretRedaction !== false) {
+      return redactResponseSecrets(response, config);
+    }
+    return response;
+  };
+}
+// ../gateway/src/cache.ts
+import { createHash } from "node:crypto";
+var log4 = createLogger();
+function createInMemoryCache(maxSize = 1000) {
+  const cache = new Map;
+  function evict() {
+    if (cache.size <= maxSize)
+      return;
+    const firstKey = cache.keys().next().value;
+    if (firstKey !== undefined)
+      cache.delete(firstKey);
+  }
+  return {
+    async get(key) {
+      const entry = cache.get(key);
+      if (!entry)
+        return null;
+      if (Date.now() > entry.expiresAt) {
+        cache.delete(key);
+        return null;
+      }
+      cache.delete(key);
+      cache.set(key, entry);
+      return entry.value;
+    },
+    async set(key, value, ttlMs) {
+      cache.set(key, { value, expiresAt: Date.now() + ttlMs });
+      evict();
+    },
+    async delete(key) {
+      cache.delete(key);
+    },
+    async clear() {
+      cache.clear();
+    }
+  };
+}
+function defaultCacheKey(ctx) {
+  const data = JSON.stringify({
+    provider: ctx.provider,
+    model: ctx.model,
+    messages: ctx.request.messages,
+    system: ctx.request.system,
+    temperature: ctx.request.temperature
+  });
+  return createHash("sha256").update(data).digest("hex");
+}
+function defaultShouldCache(_ctx, response) {
+  const temp = _ctx.request.temperature;
+  if (temp !== undefined && temp !== 0)
+    return false;
+  return response.stopReason === "end_turn";
+}
+function cacheMiddleware(config) {
+  const ttlMs = config?.ttlMs ?? 3600000;
+  const adapter = config?.adapter ?? createInMemoryCache(config?.maxSize ?? 1000);
+  const keyFn = config?.keyFn ?? defaultCacheKey;
+  const shouldCache = config?.shouldCache ?? defaultShouldCache;
+  let hits = 0;
+  let misses = 0;
+  const middleware = async (ctx, next) => {
+    if (ctx.request.stream) {
+      return next(ctx);
+    }
+    const key = keyFn(ctx);
+    const cached = await adapter.get(key);
+    if (cached) {
+      hits++;
+      log4.debug("Cache hit", { key: key.slice(0, 8), provider: ctx.provider });
+      return cached;
+    }
+    misses++;
+    const response = await next(ctx);
+    if (shouldCache(ctx, response)) {
+      await adapter.set(key, response, ttlMs);
+    }
+    return response;
+  };
+  return Object.assign(middleware, {
+    adapter,
+    stats() {
+      const total = hits + misses;
+      return {
+        hits,
+        misses,
+        size: 0,
+        hitRate: total > 0 ? hits / total : 0
+      };
+    }
+  });
+}
+// ../gateway/src/output-guardrails.ts
+var log5 = createLogger();
+var PII_PATTERNS2 = [
+  {
+    pattern: /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b/g,
+    label: "email",
+    replacement: "[REDACTED_EMAIL]"
+  },
+  {
+    pattern: /(?:\+?1[-.\s]?)?\(?\d{3}\)?[-.\s]?\d{3}[-.\s]?\d{4}\b/g,
+    label: "phone",
+    replacement: "[REDACTED_PHONE]"
+  },
+  {
+    pattern: /\b\d{3}-\d{2}-\d{4}\b/g,
+    label: "ssn",
+    replacement: "[REDACTED_SSN]"
+  },
+  {
+    pattern: /\b\d{4}[\s-]?\d{4}[\s-]?\d{4}[\s-]?\d{4}\b/g,
+    label: "credit_card",
+    replacement: "[REDACTED_CC]"
+  }
+];
+var SECRET_PATTERNS2 = [
+  {
+    pattern: /\bsk-[A-Za-z0-9]{20,}\b/g,
+    label: "api_key",
+    replacement: "[REDACTED_API_KEY]"
+  },
+  {
+    pattern: /\bpk-[A-Za-z0-9]{20,}\b/g,
+    label: "api_key",
+    replacement: "[REDACTED_API_KEY]"
+  },
+  {
+    pattern: /\bAKIA[A-Z0-9]{16}\b/g,
+    label: "aws_key",
+    replacement: "[REDACTED_AWS_KEY]"
+  }
+];
+function detectPII(text) {
   const violations = [];
-  for (const pattern of patterns) {
-    pattern.lastIndex = 0;
-    if (pattern.test(text)) {
+  const normalized = text.normalize("NFKC");
+  for (const { pattern, label } of [...PII_PATTERNS2, ...SECRET_PATTERNS2]) {
+    const regex = new RegExp(pattern.source, pattern.flags);
+    if (regex.test(normalized)) {
       violations.push({
-        type: "blocked_pattern",
-        detail: `Blocked pattern matched: ${pattern.source}`,
-        severity: "medium"
+        type: "pii",
+        detail: `Detected ${label} in output`,
+        pattern: label
       });
     }
   }
   return violations;
 }
-function scanMessageForViolations(text, config) {
-  const violations = [];
-  if (config.promptInjection !== false) {
-    violations.push(...detectPromptInjection(text));
-  }
-  if (config.jailbreakDetection) {
-    violations.push(...detectJailbreak(text));
+function redactPII(text) {
+  let result = text;
+  for (const { pattern, replacement } of [...PII_PATTERNS2, ...SECRET_PATTERNS2]) {
+    const regex = new RegExp(pattern.source, pattern.flags);
+    result = result.replace(regex, replacement);
   }
-  if (config.blockedPatterns?.length) {
-    violations.push(...checkBlockedPatterns(text, config.blockedPatterns));
+  return result;
+}
+function checkContentPolicy(text, policy) {
+  const violations = [];
+  if (policy.maxResponseLength && text.length > policy.maxResponseLength) {
+    violations.push({
+      type: "content_policy",
+      detail: `Response length ${text.length} exceeds max ${policy.maxResponseLength}`
+    });
+  }
+  if (policy.blockedPatterns) {
+    for (const pattern of policy.blockedPatterns) {
+      if (pattern.test(text)) {
+        violations.push({
+          type: "content_policy",
+          detail: `Response matches blocked pattern: ${pattern.source}`,
+          pattern: pattern.source
+        });
+      }
+    }
   }
   return violations;
 }
-function reportAndThrow(violations, config) {
-  for (const v of violations) {
-    config.onViolation?.(v);
+function checkCustomRules(text, rules) {
+  const violations = [];
+  for (const rule of rules) {
+    if (rule.pattern.test(text)) {
+      violations.push({
+        type: "custom_rule",
+        detail: rule.message ?? `Output matched custom rule: ${rule.name}`,
+        pattern: rule.pattern.source
+      });
+    }
   }
-  throw ElsiumError.validation(`Security violation detected: ${violations.map((v) => v.detail).join("; ")}`);
+  return violations;
 }
-function redactResponseSecrets(response, config) {
-  const responseText = extractText(response.message.content);
-  if (!responseText)
-    return response;
-  const { redacted, found } = redactSecrets(responseText, config.piiTypes);
-  if (found.length === 0)
-    return response;
-  for (const v of found) {
-    config.onViolation?.(v);
-  }
-  return {
-    ...response,
-    message: {
-      ...response.message,
-      content: redacted
+function outputGuardrailMiddleware(config) {
+  const mode = config.onViolation ?? "block";
+  return async (ctx, next) => {
+    const response = await next(ctx);
+    const text = extractText(response.message.content);
+    const violations = [];
+    if (config.piiDetection) {
+      violations.push(...detectPII(text));
+    }
+    if (config.contentPolicy) {
+      violations.push(...checkContentPolicy(text, config.contentPolicy));
+    }
+    if (config.customRules?.length) {
+      violations.push(...checkCustomRules(text, config.customRules));
+    }
+    if (violations.length === 0)
+      return response;
+    for (const v of violations) {
+      config.onViolationCallback?.(v);
+    }
+    switch (mode) {
+      case "block":
+        throw ElsiumError.validation(`Output guardrail violation: ${violations.map((v) => v.detail).join("; ")}`, { violations });
+      case "redact": {
+        let redacted = text;
+        if (config.piiDetection) {
+          redacted = redactPII(redacted);
+        }
+        return {
+          ...response,
+          message: { ...response.message, content: redacted }
+        };
+      }
+      case "warn":
+        log5.warn("Output guardrail violations detected", { violations });
+        return response;
     }
   };
 }
-function securityMiddleware(config) {
-  return async (ctx, next) => {
-    for (const message of ctx.request.messages) {
-      const text = extractText(message.content);
-      if (!text)
-        continue;
-      const violations = scanMessageForViolations(text, config);
-      if (violations.length > 0) {
-        reportAndThrow(violations, config);
+// ../gateway/src/batch.ts
+var log6 = createLogger();
+function createBatch(gateway2, config) {
+  const concurrency = config?.concurrency ?? 5;
+  const retryPerItem = config?.retryPerItem ?? 0;
+  return {
+    async execute(requests) {
+      const startTime = performance.now();
+      const results = new Array(requests.length);
+      let completed = 0;
+      let totalSucceeded = 0;
+      let totalFailed = 0;
+      let running = 0;
+      let nextIndex = 0;
+      const signal = config?.signal;
+      async function processItem(index) {
+        if (signal?.aborted) {
+          results[index] = {
+            index,
+            success: false,
+            error: "Batch cancelled"
+          };
+          totalFailed++;
+          return;
+        }
+        let lastError;
+        for (let attempt = 0;attempt <= retryPerItem; attempt++) {
+          try {
+            const response = await gateway2.complete(requests[index]);
+            results[index] = { index, success: true, response };
+            totalSucceeded++;
+            return;
+          } catch (err2) {
+            lastError = err2 instanceof Error ? err2.message : String(err2);
+            if (attempt < retryPerItem && err2 instanceof ElsiumError && err2.retryable) {
+              continue;
+            }
+            break;
+          }
+        }
+        results[index] = { index, success: false, error: lastError };
+        totalFailed++;
       }
+      return new Promise((resolve) => {
+        function scheduleNext() {
+          while (running < concurrency && nextIndex < requests.length) {
+            if (signal?.aborted) {
+              for (let i = nextIndex;i < requests.length; i++) {
+                results[i] = { index: i, success: false, error: "Batch cancelled" };
+                totalFailed++;
+              }
+              nextIndex = requests.length;
+              break;
+            }
+            const idx = nextIndex++;
+            running++;
+            processItem(idx).then(() => {
+              running--;
+              completed++;
+              config?.onProgress?.(completed, requests.length);
+              if (completed === requests.length) {
+                resolve({
+                  results,
+                  totalSucceeded,
+                  totalFailed,
+                  totalDurationMs: Math.round(performance.now() - startTime)
+                });
+              } else {
+                scheduleNext();
+              }
+            });
+          }
+        }
+        if (requests.length === 0) {
+          resolve({
+            results: [],
+            totalSucceeded: 0,
+            totalFailed: 0,
+            totalDurationMs: 0
+          });
+          return;
+        }
+        scheduleNext();
+      });
     }
-    const response = await next(ctx);
-    if (config.secretRedaction !== false) {
-      return redactResponseSecrets(response, config);
-    }
-    return response;
   };
 }
 // ../gateway/src/router.ts
@@ -2544,12 +3307,26 @@ function createProviderMesh(config) {
       const available = sortedProviders.find((e) => isProviderAvailable(e.name));
       const entry = available ?? sortedProviders[0];
       const gw = getGateway(entry.name);
-      return gw.stream({ ...request, model: request.model ?? entry.model });
+      let resolvedStream = null;
+      callWithCircuitBreaker(entry.name, () => {
+        resolvedStream = gw.stream({ ...request, model: request.model ?? entry.model });
+        return Promise.resolve(resolvedStream);
+      }).catch(() => {});
+      if (resolvedStream === null) {
+        const err2 = new ElsiumError({
+          code: "PROVIDER_ERROR",
+          message: "Circuit breaker is open",
+          retryable: true
+        });
+        return new ElsiumStream(async function* () {
+          yield { type: "error", error: err2 };
+        }());
+      }
+      return resolvedStream;
     }
   };
 }
 // ../tools/src/define.ts
-var log2 = createLogger();
 function defineTool(config) {
   const { name, description, input, output, handler, timeoutMs = 30000 } = config;
   return {
@@ -2624,91 +3401,20 @@ function defineTool(config) {
     }
   };
 }
-function zodDefKind2(def) {
-  return typeof def.type === "string" ? def.type : def.typeName;
-}
-function zodObjectToJsonSchema(def) {
-  const shape = typeof def.shape === "function" ? def.shape() : def.shape;
-  const properties = {};
-  const required = [];
-  for (const [key, value] of Object.entries(shape)) {
-    const fieldSchema = value;
-    properties[key] = zodToJsonSchema(fieldSchema);
-    const fieldDef = fieldSchema._def;
-    const fieldKind = zodDefKind2(fieldDef);
-    if (fieldKind !== "optional" && fieldKind !== "ZodOptional" && fieldKind !== "default" && fieldKind !== "ZodDefault") {
-      required.push(key);
-    }
-    if (fieldDef.description) {
-      properties[key].description = fieldDef.description;
-    }
-  }
-  return { type: "object", properties, required };
-}
-function zodToJsonSchema(schema) {
-  if (!("_def" in schema))
-    return { type: "object" };
-  const def = schema._def;
-  const kind = zodDefKind2(def);
-  switch (kind) {
-    case "object":
-    case "ZodObject":
-      return zodObjectToJsonSchema(def);
-    case "string":
-    case "ZodString":
-      return { type: "string" };
-    case "number":
-    case "ZodNumber":
-      return { type: "number" };
-    case "boolean":
-    case "ZodBoolean":
-      return { type: "boolean" };
-    case "array":
-    case "ZodArray":
-      return {
-        type: "array",
-        items: zodToJsonSchema(def.element ?? def.type)
-      };
-    case "enum":
-    case "ZodEnum": {
-      const values = def.values ?? (def.entries ? Object.values(def.entries) : []);
-      return { type: "string", enum: values };
-    }
-    case "optional":
-    case "ZodOptional":
-      return zodToJsonSchema(def.innerType);
-    case "default":
-    case "ZodDefault":
-      return zodToJsonSchema(def.innerType);
-    case "nullable":
-    case "ZodNullable": {
-      const inner = zodToJsonSchema(def.innerType);
-      return { ...inner, nullable: true };
-    }
-    case "ZodLiteral":
-      return { type: typeof def.value, const: def.value };
-    case "ZodUnion": {
-      const options = def.options.map(zodToJsonSchema);
-      return { anyOf: options };
-    }
-    case "ZodRecord":
-      return {
-        type: "object",
-        additionalProperties: def.valueType ? zodToJsonSchema(def.valueType) : { type: "string" }
-      };
-    case "ZodTuple": {
-      const items = (def.items ?? []).map(zodToJsonSchema);
-      return { type: "array", prefixItems: items, minItems: items.length, maxItems: items.length };
-    }
-    case "ZodDate":
-      return { type: "string", format: "date-time" };
-    default:
-      log2.warn(`zodToJsonSchema: unsupported type ${kind}, defaulting to string`);
-      return { type: "string" };
-  }
-}
 // ../tools/src/toolkit.ts
 function createToolkit(name, tools) {
+  const seen = new Set;
+  for (const tool of tools) {
+    if (seen.has(tool.name)) {
+      throw new ElsiumError({
+        code: "CONFIG_ERROR",
+        message: `Duplicate tool name "${tool.name}" in toolkit "${name}"`,
+        retryable: false,
+        metadata: { toolkit: name, tool: tool.name }
+      });
+    }
+    seen.add(tool.name);
+  }
   const toolMap = new Map(tools.map((t) => [t.name, t]));
   return {
     name,
@@ -6727,7 +7433,26 @@ var coerce = {
 };
 var NEVER = INVALID;
 // ../tools/src/builtin.ts
-var BLOCKED_HOSTS = /^(localhost|127\.\d+\.\d+\.\d+|10\.\d+\.\d+\.\d+|172\.(1[6-9]|2\d|3[01])\.\d+\.\d+|192\.168\.\d+\.\d+|169\.254\.\d+\.\d+|0\.0\.0\.0|0+\.0+\.0+\.0+|\[::1\]|\[::ffff:127\.\d+\.\d+\.\d+\]|::ffff:127\.\d+\.\d+\.\d+|0177\.\d+\.\d+\.\d+|2130706433|\[fd[0-9a-f]{2}:)/i;
+var BLOCKED_HOSTS = /^(localhost|127\.\d+\.\d+\.\d+|10\.\d+\.\d+\.\d+|172\.(1[6-9]|2\d|3[01])\.\d+\.\d+|192\.168\.\d+\.\d+|169\.254\.\d+\.\d+|0\.0\.0\.0|0+\.0+\.0+\.0+|\[?::1\]?|\[?::ffff:127\.\d+\.\d+\.\d+\]?|\[?::ffff:10\.\d+\.\d+\.\d+\]?|\[?::ffff:192\.168\.\d+\.\d+\]?|\[?::ffff:172\.(1[6-9]|2\d|3[01])\.\d+\.\d+\]?|::ffff:127\.\d+\.\d+\.\d+|0177\.\d+\.\d+\.\d+|2130706433|\[?::\]?|\[?f[cd][0-9a-f]{2}:|\[?fe80:)/i;
+var BLOCKED_HEADER_NAMES = new Set([
+  "cookie",
+  "set-cookie",
+  "authorization",
+  "proxy-authorization",
+  "host",
+  "x-api-key",
+  "x-forwarded-for",
+  "x-real-ip"
+]);
+function sanitizeHeaders(headers) {
+  const safe = {};
+  for (const [key, value] of Object.entries(headers)) {
+    if (!BLOCKED_HEADER_NAMES.has(key.toLowerCase())) {
+      safe[key] = value;
+    }
+  }
+  return safe;
+}
 function validateUrl(urlString) {
   const parsed = new URL(urlString);
   if (!["http:", "https:"].includes(parsed.protocol)) {
@@ -6752,8 +7477,9 @@ var httpFetchTool = defineTool({
   timeoutMs: 15000,
   handler: async ({ url, headers }, context) => {
     validateUrl(url);
+    const safeHeaders = headers ? sanitizeHeaders(headers) : undefined;
     const response = await fetch(url, {
-      headers,
+      headers: safeHeaders,
       signal: context.signal,
       redirect: "manual"
     });
@@ -7228,15 +7954,33 @@ function createMemory(config) {
         case "unlimited":
           break;
       }
+      if (config.store && config.agentId) {
+        config.store.save(config.agentId, [...messages]).catch(() => {});
+      }
     },
     getMessages() {
       return [...messages];
     },
     clear() {
       messages.length = 0;
+      if (config.store && config.agentId) {
+        config.store.clear(config.agentId).catch(() => {});
+      }
     },
     getTokenEstimate() {
       return messages.reduce((sum, m) => sum + estimateTokens(m), 0);
+    },
+    async loadFromStore() {
+      if (!config.store || !config.agentId)
+        return;
+      const stored = await config.store.load(config.agentId);
+      messages.length = 0;
+      messages.push(...stored);
+    },
+    async saveToStore() {
+      if (!config.store || !config.agentId)
+        return;
+      await config.store.save(config.agentId, [...messages]);
     }
   };
 }
@@ -7282,7 +8026,7 @@ var JAILBREAK_PATTERNS2 = [
     detail: "Opposite mode jailbreak attempt"
   }
 ];
-var SECRET_PATTERNS2 = [
+var SECRET_PATTERNS3 = [
   {
     pattern: /\bsk-[a-zA-Z0-9_-]{20,}\b/g,
     detail: "API secret key detected",
@@ -7360,7 +8104,7 @@ function createAgentSecurity(config) {
     const violations = [];
     let redactedOutput = output;
     if (config.redactSecrets !== false) {
-      for (const { pattern, detail, replacement } of SECRET_PATTERNS2) {
+      for (const { pattern, detail, replacement } of SECRET_PATTERNS3) {
         const regex = new RegExp(pattern.source, pattern.flags);
         if (regex.test(redactedOutput)) {
           violations.push({ type: "secret_detected", detail, severity: "medium" });
@@ -7421,9 +8165,9 @@ Only respond with JSON, nothing else.`
     if (!parsed) {
       return null;
     }
-    const score = parsed.score ?? 0.5;
+    const score = typeof parsed.score === "number" ? parsed.score : 0.5;
     const threshold = config.hallucination?.threshold ?? 0.7;
-    const claims = parsed.hallucinated_claims ?? [];
+    const claims = Array.isArray(parsed.hallucinated_claims) ? parsed.hallucinated_claims : [];
     return {
       passed: score >= threshold,
       score,
@@ -7486,12 +8230,12 @@ Only respond with JSON, nothing else.`
     if (!parsed) {
       return null;
     }
-    const score = parsed.score ?? 0.5;
+    const score = typeof parsed.score === "number" ? parsed.score : 0.5;
     const threshold = config.relevance?.threshold ?? 0.5;
     return {
       passed: score >= threshold,
       score,
-      reason: parsed.reason ?? (score >= threshold ? "Output is relevant" : "Output lacks relevance")
+      reason: typeof parsed.reason === "string" ? parsed.reason : score >= threshold ? "Output is relevant" : "Output lacks relevance"
     };
   }
   function checkRelevanceHeuristic(input, output) {
@@ -7543,8 +8287,8 @@ Only respond with JSON, nothing else.`
     if (!parsed) {
       return null;
     }
-    const score = parsed.score ?? 0.5;
-    const claims = parsed.ungrounded_claims ?? [];
+    const score = typeof parsed.score === "number" ? parsed.score : 0.5;
+    const claims = Array.isArray(parsed.ungrounded_claims) ? parsed.ungrounded_claims : [];
     return {
       passed: score >= 0.7,
       score,
@@ -7605,16 +8349,23 @@ Only respond with JSON, nothing else.`
 }
 
 // ../agents/src/state-machine.ts
+async function safeHook(fn) {
+  if (!fn)
+    return;
+  try {
+    await fn();
+  } catch (_) {}
+}
 function executeStateMachine(baseConfig, stateConfig, deps, input, options) {
   return runStateMachine(baseConfig, stateConfig, deps, input, options ?? {});
 }
-function handleToolCallsAndContinue(response, toolMap, toolCallHistory, conversationMessages, signal) {
+function handleToolCallsAndContinue(response, toolMap, toolCallHistory, conversationMessages, signal, hooks, approvalGate, approvalConfig) {
   const toolCalls = response.message.toolCalls;
   if (!toolCalls?.length || response.stopReason !== "tool_use") {
     return null;
   }
   return (async () => {
-    const toolResults = await executeToolCalls(toolCalls, toolMap, toolCallHistory, signal);
+    const toolResults = await executeToolCalls(toolCalls, toolMap, toolCallHistory, signal, hooks, approvalGate, approvalConfig);
     const toolMessage = {
       role: "tool",
       content: "",
@@ -7718,6 +8469,7 @@ async function runStateMachine(baseConfig, stateConfig, deps, input, options) {
     });
   }
   const agentSecurity = baseConfig.guardrails?.security ? createAgentSecurity(baseConfig.guardrails.security) : null;
+  const approvalGate = baseConfig.guardrails?.approval ? createApprovalGate(baseConfig.guardrails.approval) : null;
   const maxTokenBudget = guardrails?.maxTokenBudget ?? 500000;
   const outputValidator = guardrails?.outputValidator ?? (() => true);
   const conversationMessages = [{ role: "user", content: input }];
@@ -7735,7 +8487,8 @@ async function runStateMachine(baseConfig, stateConfig, deps, input, options) {
     checkTokenBudget(totalInputTokens + totalOutputTokens, maxTokenBudget);
     response = applyOutputGuardrails(response, outputValidator, agentSecurity);
     conversationMessages.push(response.message);
-    const toolCallAction = handleToolCallsAndContinue(response, toolMap, toolCallHistory, conversationMessages, options.signal);
+    await safeHook(() => baseConfig.hooks?.onMessage?.(response.message));
+    const toolCallAction = handleToolCallsAndContinue(response, toolMap, toolCallHistory, conversationMessages, options.signal, baseConfig.hooks, approvalGate, baseConfig.guardrails);
     if (toolCallAction) {
       await toolCallAction;
       continue;
@@ -7756,9 +8509,30 @@ async function runStateMachine(baseConfig, stateConfig, deps, input, options) {
     metadata: { iterations, maxIterations, lastState: currentStateName }
   });
 }
-async function executeToolCalls(toolCalls, toolMap, history, signal) {
+async function executeToolCalls(toolCalls, toolMap, history, signal, hooks, approvalGate, approvalConfig) {
   const results = [];
   for (const tc of toolCalls) {
+    await safeHook(() => hooks?.onToolCall?.({ name: tc.name, arguments: tc.arguments }));
+    if (approvalGate && shouldRequireApproval(approvalConfig?.approval?.requireApprovalFor, {
+      toolName: tc.name
+    })) {
+      const decision = await approvalGate.requestApproval("tool_call", `Execute tool: ${tc.name}`, {
+        toolName: tc.name,
+        arguments: tc.arguments
+      });
+      if (!decision.approved) {
+        const deniedResult = {
+          success: false,
+          error: `Tool call denied: ${decision.reason ?? "Approval denied"}`,
+          toolCallId: tc.id,
+          durationMs: 0
+        };
+        await safeHook(() => hooks?.onToolResult?.(deniedResult));
+        history.push({ name: tc.name, arguments: tc.arguments, result: deniedResult });
+        results.push(formatToolResult(deniedResult));
+        continue;
+      }
+    }
     const tool = toolMap.get(tc.name);
     if (!tool) {
       const errorResult = {
@@ -7767,11 +8541,13 @@ async function executeToolCalls(toolCalls, toolMap, history, signal) {
         toolCallId: tc.id,
         durationMs: 0
       };
+      await safeHook(() => hooks?.onToolResult?.(errorResult));
       history.push({ name: tc.name, arguments: tc.arguments, result: errorResult });
       results.push(formatToolResult(errorResult));
       continue;
     }
     const result = await tool.execute(tc.arguments, { toolCallId: tc.id, signal });
+    await safeHook(() => hooks?.onToolResult?.(result));
     history.push({ name: tc.name, arguments: tc.arguments, result });
     results.push(formatToolResult(result));
   }
@@ -7779,7 +8555,7 @@ async function executeToolCalls(toolCalls, toolMap, history, signal) {
 }
 
 // ../agents/src/agent.ts
-async function safeHook(fn) {
+async function safeHook2(fn) {
   if (!fn)
     return;
   try {
@@ -7885,7 +8661,7 @@ function defineAgent(config, deps) {
       traceId,
       confidence
     };
-    await safeHook(() => config.hooks?.onComplete?.(agentResult));
+    await safeHook2(() => config.hooks?.onComplete?.(agentResult));
     return { action: "return", result: agentResult };
   }
   function checkBudget(totalInputTokens, totalOutputTokens) {
@@ -7943,7 +8719,7 @@ function defineAgent(config, deps) {
       totalInputTokens += response.usage.inputTokens;
       totalOutputTokens += response.usage.outputTokens;
       totalCost += response.cost.totalCost;
-      await safeHook(() => config.hooks?.onMessage?.(response.message));
+      await safeHook2(() => config.hooks?.onMessage?.(response.message));
       conversationMessages.push(response.message);
       if (!response.message.toolCalls?.length || response.stopReason !== "tool_use") {
         const result = await handleNonToolResponse(response, messages, iterations, totalInputTokens, totalOutputTokens, totalCost, toolCallHistory, traceId, conversationMessages);
@@ -7971,7 +8747,7 @@ function defineAgent(config, deps) {
   async function executeToolCalls2(toolCalls, history, options = {}) {
     const results = [];
     for (const tc of toolCalls) {
-      await safeHook(() => config.hooks?.onToolCall?.({ name: tc.name, arguments: tc.arguments }));
+      await safeHook2(() => config.hooks?.onToolCall?.({ name: tc.name, arguments: tc.arguments }));
       if (approvalGate && shouldRequireApproval(config.guardrails?.approval?.requireApprovalFor, {
         toolName: tc.name
       })) {
@@ -7983,7 +8759,7 @@ function defineAgent(config, deps) {
             toolCallId: tc.id,
             durationMs: 0
           };
-          await safeHook(() => config.hooks?.onToolResult?.(deniedResult));
+          await safeHook2(() => config.hooks?.onToolResult?.(deniedResult));
           history.push({ name: tc.name, arguments: tc.arguments, result: deniedResult });
           results.push(formatToolResult(deniedResult));
           continue;
@@ -7997,7 +8773,7 @@ function defineAgent(config, deps) {
           toolCallId: tc.id,
           durationMs: 0
         };
-        await safeHook(() => config.hooks?.onToolResult?.(errorResult));
+        await safeHook2(() => config.hooks?.onToolResult?.(errorResult));
         history.push({ name: tc.name, arguments: tc.arguments, result: errorResult });
         results.push(formatToolResult(errorResult));
         continue;
@@ -8006,7 +8782,7 @@ function defineAgent(config, deps) {
         toolCallId: tc.id,
         signal: options.signal
       });
-      await safeHook(() => config.hooks?.onToolResult?.(result));
+      await safeHook2(() => config.hooks?.onToolResult?.(result));
       history.push({ name: tc.name, arguments: tc.arguments, result });
       results.push(formatToolResult(result));
     }
@@ -8029,15 +8805,113 @@ function defineAgent(config, deps) {
           continue;
         validateInputText(extractText(msg.content));
       }
-      if (config.states && config.initialState) {
-        const inputText = messages.filter((m) => m.role === "user").map((m) => extractText(m.content)).join(`
-`);
-        return executeStateMachine(config, { states: config.states, initialState: config.initialState }, deps, inputText || "", options);
+      if (config.states && config.initialState) {
+        const inputText = messages.filter((m) => m.role === "user").map((m) => extractText(m.content)).join(`
+`);
+        return executeStateMachine(config, { states: config.states, initialState: config.initialState }, deps, inputText || "", options);
+      }
+      return executeLoop(messages, options);
+    },
+    resetMemory() {
+      memory.clear();
+    }
+  };
+}
+// ../agents/src/stores/memory-store.ts
+function createInMemoryMemoryStore() {
+  const store = new Map;
+  return {
+    async load(agentId) {
+      return [...store.get(agentId) ?? []];
+    },
+    async save(agentId, messages) {
+      store.set(agentId, [...messages]);
+    },
+    async clear(agentId) {
+      store.delete(agentId);
+    }
+  };
+}
+// ../agents/src/stores/sqlite-store.ts
+import { createRequire } from "node:module";
+var require2 = createRequire(import.meta.url);
+var log7 = createLogger();
+var BLOCKED_KEYS2 = new Set(["__proto__", "constructor", "prototype"]);
+var TABLE_NAME_PATTERN = /^[a-zA-Z_][a-zA-Z0-9_]*$/;
+function createSqliteMemoryStore(config) {
+  const { path, tableName = "agent_memory" } = config;
+  if (BLOCKED_KEYS2.has(tableName)) {
+    throw new Error(`Invalid table name: ${tableName}`);
+  }
+  if (!TABLE_NAME_PATTERN.test(tableName)) {
+    throw new Error(`Invalid table name format: ${tableName}`);
+  }
+  let db = null;
+  let initPromise = null;
+  async function getDb() {
+    if (db)
+      return db;
+    if (initPromise)
+      return initPromise;
+    initPromise = (async () => {
+      try {
+        const Database = require2("better-sqlite3");
+        db = new Database(path);
+        db.exec(`
+					CREATE TABLE IF NOT EXISTS ${tableName} (
+						agent_id TEXT NOT NULL,
+						idx INTEGER NOT NULL,
+						role TEXT NOT NULL,
+						content TEXT NOT NULL,
+						metadata TEXT,
+						created_at INTEGER NOT NULL DEFAULT (unixepoch()),
+						PRIMARY KEY (agent_id, idx)
+					)
+				`);
+        return db;
+      } catch (err2) {
+        initPromise = null;
+        log7.error("Failed to initialize SQLite memory store", {
+          error: err2 instanceof Error ? err2.message : String(err2)
+        });
+        throw new Error("better-sqlite3 is required for SQLite memory store. Install it as a dependency.");
+      }
+    })();
+    return initPromise;
+  }
+  return {
+    async load(agentId) {
+      if (BLOCKED_KEYS2.has(agentId))
+        return [];
+      const database = await getDb();
+      const rows = database.prepare(`SELECT role, content, metadata FROM ${tableName} WHERE agent_id = ? ORDER BY idx`).all(agentId);
+      return rows.map((row) => {
+        const msg = {
+          role: row.role,
+          content: JSON.parse(row.content)
+        };
+        if (row.metadata) {
+          msg.metadata = JSON.parse(row.metadata);
+        }
+        return msg;
+      });
+    },
+    async save(agentId, messages) {
+      if (BLOCKED_KEYS2.has(agentId))
+        return;
+      const database = await getDb();
+      database.prepare(`DELETE FROM ${tableName} WHERE agent_id = ?`).run(agentId);
+      const insert = database.prepare(`INSERT INTO ${tableName} (agent_id, idx, role, content, metadata) VALUES (?, ?, ?, ?, ?)`);
+      for (let i = 0;i < messages.length; i++) {
+        const msg = messages[i];
+        insert.run(agentId, i, msg.role, JSON.stringify(msg.content), msg.metadata ? JSON.stringify(msg.metadata) : null);
       }
-      return executeLoop(messages, options);
     },
-    resetMemory() {
-      memory.clear();
+    async clear(agentId) {
+      if (BLOCKED_KEYS2.has(agentId))
+        return;
+      const database = await getDb();
+      database.prepare(`DELETE FROM ${tableName} WHERE agent_id = ?`).run(agentId);
     }
   };
 }
@@ -8538,7 +9412,11 @@ function createMockEmbeddings(dims = 128) {
     }
   };
 }
+var embeddingProviderRegistry = createRegistry("embeddingProvider");
 function getEmbeddingProvider(config) {
+  const registered = embeddingProviderRegistry.get(config.provider);
+  if (registered)
+    return registered(config);
   switch (config.provider) {
     case "openai":
       return createOpenAIEmbeddings(config);
@@ -8547,12 +9425,13 @@ function getEmbeddingProvider(config) {
     default:
       throw new ElsiumError({
         code: "CONFIG_ERROR",
-        message: `Unknown embedding provider: ${config.provider}`,
+        message: `Unknown embedding provider: ${config.provider}. Available: openai, mock${embeddingProviderRegistry.list().length ? `, ${embeddingProviderRegistry.list().join(", ")}` : ""}`,
         retryable: false
       });
   }
 }
 // ../rag/src/vectorstore.ts
+var vectorStoreRegistry = createRegistry("vectorStore");
 function cosineSimilarity(a, b) {
   if (a.length !== b.length)
     return 0;
@@ -8686,6 +9565,123 @@ function rag(config) {
     }
   };
 }
+// ../rag/src/stores/pgvector.ts
+import { createRequire as createRequire2 } from "node:module";
+var require3 = createRequire2(import.meta.url);
+var log8 = createLogger();
+var BLOCKED_KEYS3 = new Set(["__proto__", "constructor", "prototype"]);
+var TABLE_NAME_PATTERN2 = /^[a-zA-Z_][a-zA-Z0-9_]*$/;
+function createPgVectorStore(config) {
+  const { connectionString, tableName = "vector_chunks", dimensions = 1536 } = config;
+  if (BLOCKED_KEYS3.has(tableName)) {
+    throw new Error(`Invalid table name: ${tableName}`);
+  }
+  if (!TABLE_NAME_PATTERN2.test(tableName)) {
+    throw new Error(`Invalid table name format: ${tableName}`);
+  }
+  let client = null;
+  let initialized = false;
+  async function getClient() {
+    if (client)
+      return client;
+    try {
+      const pg = require3("pg");
+      client = new pg.Client({ connectionString });
+      await client.connect();
+      if (!initialized) {
+        await client.query("CREATE EXTENSION IF NOT EXISTS vector");
+        await client.query(`
+					CREATE TABLE IF NOT EXISTS ${tableName} (
+						id TEXT PRIMARY KEY,
+						content TEXT NOT NULL,
+						document_id TEXT NOT NULL,
+						chunk_index INTEGER NOT NULL,
+						metadata JSONB DEFAULT '{}',
+						embedding vector(${dimensions})
+					)
+				`);
+        initialized = true;
+      }
+      return client;
+    } catch (err2) {
+      log8.error("Failed to initialize PgVector store", {
+        error: err2 instanceof Error ? err2.message : String(err2)
+      });
+      throw new Error("pg is required for PgVector store. Install it as a dependency.");
+    }
+  }
+  return {
+    name: "pgvector",
+    async upsert(chunks) {
+      const pg = await getClient();
+      for (const chunk of chunks) {
+        if (BLOCKED_KEYS3.has(chunk.id))
+          continue;
+        const embedding = `[${chunk.embedding.values.join(",")}]`;
+        await pg.query(`INSERT INTO ${tableName} (id, content, document_id, chunk_index, metadata, embedding)
+					 VALUES ($1, $2, $3, $4, $5, $6)
+					 ON CONFLICT (id) DO UPDATE SET
+						content = EXCLUDED.content,
+						document_id = EXCLUDED.document_id,
+						chunk_index = EXCLUDED.chunk_index,
+						metadata = EXCLUDED.metadata,
+						embedding = EXCLUDED.embedding`, [
+          chunk.id,
+          chunk.content,
+          chunk.documentId,
+          chunk.index,
+          JSON.stringify(chunk.metadata),
+          embedding
+        ]);
+      }
+    },
+    async query(embedding, options) {
+      const pg = await getClient();
+      const topK = options?.topK ?? 5;
+      const minScore = options?.minScore ?? 0;
+      const embeddingStr = `[${embedding.values.join(",")}]`;
+      const result = await pg.query(`SELECT id, content, document_id, chunk_index, metadata,
+								1 - (embedding <=> $1::vector) as score
+				 FROM ${tableName}
+				 WHERE 1 - (embedding <=> $1::vector) >= $2
+				 ORDER BY embedding <=> $1::vector
+				 LIMIT $3`, [embeddingStr, minScore, topK]);
+      return result.rows.map((row) => ({
+        chunk: {
+          id: row.id,
+          content: row.content,
+          documentId: row.document_id,
+          index: row.chunk_index,
+          metadata: {
+            startChar: 0,
+            endChar: 0,
+            tokenEstimate: 0,
+            ...row.metadata ?? {}
+          }
+        },
+        score: row.score,
+        distance: 1 - row.score
+      }));
+    },
+    async delete(ids) {
+      const pg = await getClient();
+      const filtered = ids.filter((id) => !BLOCKED_KEYS3.has(id));
+      if (filtered.length === 0)
+        return;
+      const placeholders = filtered.map((_, i) => `$${i + 1}`).join(", ");
+      await pg.query(`DELETE FROM ${tableName} WHERE id IN (${placeholders})`, filtered);
+    },
+    async clear() {
+      const pg = await getClient();
+      await pg.query(`DELETE FROM ${tableName}`);
+    },
+    async count() {
+      const pg = await getClient();
+      const result = await pg.query(`SELECT COUNT(*)::int as count FROM ${tableName}`);
+      return result.rows[0]?.count ?? 0;
+    }
+  };
+}
 // ../workflows/src/step.ts
 function step(name, config) {
   return { name, ...config };
@@ -9267,7 +10263,8 @@ function createCostEngine(config = {}) {
   };
 }
 // ../observe/src/tracer.ts
-var log3 = createLogger();
+import { writeFileSync } from "node:fs";
+var log9 = createLogger();
 function observe(config = {}) {
   const {
     output = ["console"],
@@ -9282,7 +10279,21 @@ function observe(config = {}) {
   for (const out of output) {
     if (out === "console") {
       handlers.push(consoleHandler);
-    } else if (out === "json-file") {} else {
+    } else if (out === "json-file") {
+      exporters.push({
+        name: "json-file",
+        export(spansToExport) {
+          const filename = `.elsium/traces-${Date.now()}.json`;
+          try {
+            writeFileSync(filename, JSON.stringify(spansToExport, null, 2));
+          } catch (err2) {
+            log9.error("Failed to write trace file", {
+              error: err2 instanceof Error ? err2.message : String(err2)
+            });
+          }
+        }
+      });
+    } else {
       exporters.push(out);
     }
   }
@@ -9351,7 +10362,7 @@ function observe(config = {}) {
 function consoleHandler(span) {
   const duration = span.durationMs !== undefined ? `${span.durationMs}ms` : "running";
   const status = span.status === "error" ? "[ERROR]" : span.status === "ok" ? "[OK]" : "[...]";
-  log3.info("span", {
+  log9.info("span", {
     trace: span.traceId,
     span: span.name,
     kind: span.kind,
@@ -9449,8 +10460,86 @@ function createMetrics(options) {
     }
   };
 }
+// ../observe/src/experiment.ts
+import { createHash as createHash2 } from "node:crypto";
+var log10 = createLogger();
+function createExperiment(config) {
+  const { name, variants } = config;
+  if (variants.length === 0) {
+    throw new Error("Experiment must have at least one variant");
+  }
+  const totalWeight = variants.reduce((sum, v) => sum + v.weight, 0);
+  const stats = {};
+  for (const v of variants) {
+    stats[v.name] = { assignments: 0, metrics: {} };
+  }
+  function hashAssign(userId) {
+    const hash = createHash2("sha256").update(`${name}:${userId}`).digest();
+    const value = hash.readUInt32BE(0) % 1e4 / 1e4;
+    return pickVariant(value);
+  }
+  function randomAssign() {
+    const value = Math.random();
+    return pickVariant(value);
+  }
+  function pickVariant(value) {
+    let cumulative = 0;
+    for (const v of variants) {
+      cumulative += v.weight / totalWeight;
+      if (value < cumulative)
+        return v;
+    }
+    return variants[variants.length - 1];
+  }
+  return {
+    assign(userId) {
+      const variant = userId ? hashAssign(userId) : randomAssign();
+      const s = stats[variant.name];
+      if (s)
+        s.assignments++;
+      log10.debug("Experiment assignment", {
+        experiment: name,
+        variant: variant.name,
+        userId
+      });
+      return variant;
+    },
+    record(variant, metrics) {
+      const s = stats[variant];
+      if (!s)
+        return;
+      for (const [key, value] of Object.entries(metrics)) {
+        if (!s.metrics[key]) {
+          s.metrics[key] = { sum: 0, count: 0 };
+        }
+        s.metrics[key].sum += value;
+        s.metrics[key].count++;
+      }
+    },
+    results() {
+      let totalAssignments = 0;
+      const variantResults = {};
+      for (const [vName, s] of Object.entries(stats)) {
+        totalAssignments += s.assignments;
+        const metricsResult = {};
+        for (const [key, m] of Object.entries(s.metrics)) {
+          metricsResult[key] = {
+            sum: m.sum,
+            count: m.count,
+            avg: m.count > 0 ? m.sum / m.count : 0
+          };
+        }
+        variantResults[vName] = {
+          assignments: s.assignments,
+          metrics: metricsResult
+        };
+      }
+      return { name, totalAssignments, variants: variantResults };
+    }
+  };
+}
 // ../observe/src/otel.ts
-var log4 = createLogger();
+var log11 = createLogger();
 var SPAN_KIND_MAP = {
   llm: 3,
   tool: 1,
@@ -9591,10 +10680,10 @@ function createOTLPExporter(config) {
         body: JSON.stringify(payload)
       });
       if (!response.ok) {
-        log4.error(`OTLP export failed: ${response.status} ${response.statusText}`);
+        log11.error(`OTLP export failed: ${response.status} ${response.statusText}`);
       }
     } catch (err2) {
-      log4.error("OTLP export error", { error: err2 instanceof Error ? err2.message : String(err2) });
+      log11.error("OTLP export error", { error: err2 instanceof Error ? err2.message : String(err2) });
     }
   }
   function startAutoFlush() {
@@ -9617,6 +10706,16 @@ function createOTLPExporter(config) {
       } else {
         startAutoFlush();
       }
+    },
+    async shutdown() {
+      if (flushTimer) {
+        clearInterval(flushTimer);
+        flushTimer = null;
+      }
+      if (buffer.length > 0) {
+        const batch = buffer.splice(0, buffer.length);
+        await sendBatch(batch);
+      }
     }
   };
 }
@@ -11692,7 +12791,7 @@ var Hono2 = class extends Hono {
 // ../app/src/middleware.ts
 import { timingSafeEqual } from "node:crypto";
 function corsMiddleware(config = true) {
-  const opts = typeof config === "boolean" ? { origin: [], methods: ["GET", "POST", "OPTIONS"] } : config;
+  const opts = typeof config === "boolean" ? { origin: "*", methods: ["GET", "POST", "OPTIONS"] } : config;
   return async (c, next) => {
     const requestOrigin = c.req.header("Origin") ?? "";
     let allowedOrigin;
@@ -11769,8 +12868,203 @@ function rateLimitMiddleware(config) {
     await next();
   };
 }
+function requestIdMiddleware() {
+  return async (c, next) => {
+    const raw2 = c.req.header("X-Request-ID");
+    const id = raw2 && /^[\w\-.:]{1,128}$/.test(raw2) ? raw2 : generateId("req");
+    c.set("requestId", id);
+    await next();
+    c.res.headers.set("X-Request-ID", id);
+  };
+}
+function requestLoggerMiddleware(logger) {
+  const log12 = logger ?? createLogger();
+  return async (c, next) => {
+    const start = Date.now();
+    await next();
+    const duration = Date.now() - start;
+    log12.info(`${c.req.method} ${c.req.path}`, {
+      method: c.req.method,
+      path: c.req.path,
+      status: c.res.status,
+      durationMs: duration,
+      requestId: c.get("requestId")
+    });
+  };
+}
+
+// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/utils/stream.js
+var StreamingApi = class {
+  writer;
+  encoder;
+  writable;
+  abortSubscribers = [];
+  responseReadable;
+  aborted = false;
+  closed = false;
+  constructor(writable, _readable) {
+    this.writable = writable;
+    this.writer = writable.getWriter();
+    this.encoder = new TextEncoder;
+    const reader = _readable.getReader();
+    this.abortSubscribers.push(async () => {
+      await reader.cancel();
+    });
+    this.responseReadable = new ReadableStream({
+      async pull(controller) {
+        const { done, value } = await reader.read();
+        done ? controller.close() : controller.enqueue(value);
+      },
+      cancel: () => {
+        this.abort();
+      }
+    });
+  }
+  async write(input) {
+    try {
+      if (typeof input === "string") {
+        input = this.encoder.encode(input);
+      }
+      await this.writer.write(input);
+    } catch {}
+    return this;
+  }
+  async writeln(input) {
+    await this.write(input + `
+`);
+    return this;
+  }
+  sleep(ms) {
+    return new Promise((res) => setTimeout(res, ms));
+  }
+  async close() {
+    try {
+      await this.writer.close();
+    } catch {}
+    this.closed = true;
+  }
+  async pipe(body) {
+    this.writer.releaseLock();
+    await body.pipeTo(this.writable, { preventClose: true });
+    this.writer = this.writable.getWriter();
+  }
+  onAbort(listener) {
+    this.abortSubscribers.push(listener);
+  }
+  abort() {
+    if (!this.aborted) {
+      this.aborted = true;
+      this.abortSubscribers.forEach((subscriber) => subscriber());
+    }
+  }
+};
+
+// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/helper/streaming/utils.js
+var isOldBunVersion = () => {
+  const version = typeof Bun !== "undefined" ? Bun.version : undefined;
+  if (version === undefined) {
+    return false;
+  }
+  const result = version.startsWith("1.1") || version.startsWith("1.0") || version.startsWith("0.");
+  isOldBunVersion = () => result;
+  return result;
+};
+
+// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/helper/streaming/stream.js
+var contextStash = /* @__PURE__ */ new WeakMap;
+var stream = (c, cb, onError) => {
+  const { readable, writable } = new TransformStream;
+  const stream2 = new StreamingApi(writable, readable);
+  if (isOldBunVersion()) {
+    c.req.raw.signal.addEventListener("abort", () => {
+      if (!stream2.closed) {
+        stream2.abort();
+      }
+    });
+  }
+  contextStash.set(stream2.responseReadable, c);
+  (async () => {
+    try {
+      await cb(stream2);
+    } catch (e) {
+      if (e === undefined) {} else if (e instanceof Error && onError) {
+        await onError(e, stream2);
+      } else {
+        console.error(e);
+      }
+    } finally {
+      stream2.close();
+    }
+  })();
+  return c.newResponse(stream2.responseReadable);
+};
+
+// ../app/src/sse.ts
+function sseHeaders() {
+  return {
+    "Content-Type": "text/event-stream",
+    "Cache-Control": "no-cache",
+    Connection: "keep-alive",
+    "X-Accel-Buffering": "no"
+  };
+}
+function formatSSE(event, data) {
+  const json = JSON.stringify(data);
+  if (event === "message") {
+    return `data: ${json}
+
+`;
+  }
+  return `event: ${event}
+data: ${json}
+
+`;
+}
+function streamResponse(c, source) {
+  const headers = sseHeaders();
+  for (const [key, value] of Object.entries(headers)) {
+    c.header(key, value);
+  }
+  return stream(c, async (s) => {
+    try {
+      for await (const event of source) {
+        const sseData = formatSSE("message", event);
+        await s.write(sseData);
+      }
+    } catch (err2) {
+      const errorEvent = {
+        type: "error",
+        error: err2 instanceof Error ? err2 : new Error(String(err2))
+      };
+      const sseData = formatSSE("error", {
+        type: "error",
+        message: errorEvent.error.message
+      });
+      await s.write(sseData);
+    }
+  });
+}
 
 // ../app/src/routes.ts
+function parseJsonBody(raw2) {
+  try {
+    return { ok: true, data: JSON.parse(raw2) };
+  } catch {
+    return { ok: false };
+  }
+}
+function elsiumErrorResponse(c, err2, fallbackMessage) {
+  if (err2 instanceof ElsiumError) {
+    return c.json({ error: err2.message, code: err2.code }, err2.statusCode ?? 500);
+  }
+  return c.json({ error: fallbackMessage }, 500);
+}
+function resolveAgent(name, agents, defaultAgent) {
+  const agent = name ? agents.get(name) : defaultAgent;
+  if (agent)
+    return { agent };
+  return { error: name ? `Agent "${name}" not found` : "No default agent configured" };
+}
 function createRoutes(deps) {
   const app = new Hono2;
   let totalRequests = 0;
@@ -11811,17 +13105,32 @@ function createRoutes(deps) {
     if (rawText.length > MAX_BODY_SIZE) {
       return c.json({ error: "Request body too large (max 1MB)" }, 413);
     }
-    const body = JSON.parse(rawText);
+    const parsed = parseJsonBody(rawText);
+    if (!parsed.ok) {
+      return c.json({ error: "Invalid JSON in request body" }, 400);
+    }
+    const body = parsed.data;
     if (!body.message) {
       return c.json({ error: "message is required" }, 400);
     }
-    const agent = body.agent ? deps.agents.get(body.agent) : deps.defaultAgent;
-    if (!agent) {
-      return c.json({
-        error: body.agent ? `Agent "${body.agent}" not found` : "No default agent configured"
-      }, 404);
+    const resolved = resolveAgent(body.agent, deps.agents, deps.defaultAgent);
+    if ("error" in resolved) {
+      return c.json({ error: resolved.error }, 404);
+    }
+    if (body.stream) {
+      const stream2 = deps.gateway.stream({
+        messages: [{ role: "user", content: body.message }],
+        system: resolved.agent.config.system,
+        model: resolved.agent.config.model
+      });
+      return streamResponse(c, stream2);
+    }
+    let result;
+    try {
+      result = await resolved.agent.run(body.message);
+    } catch (err2) {
+      return elsiumErrorResponse(c, err2, "Agent execution failed");
     }
-    const result = await agent.run(body.message);
     deps.tracer?.trackLLMCall({
       model: "unknown",
       inputTokens: result.usage.totalInputTokens,
@@ -11838,7 +13147,7 @@ function createRoutes(deps) {
         totalTokens: result.usage.totalTokens,
         cost: result.usage.totalCost
       },
-      model: agent.config.model ?? "default",
+      model: resolved.agent.config.model ?? "default",
       traceId: result.traceId
     };
     return c.json(response);
@@ -11850,7 +13159,11 @@ function createRoutes(deps) {
     if (rawText.length > MAX_BODY_SIZE) {
       return c.json({ error: "Request body too large (max 1MB)" }, 413);
     }
-    const body = JSON.parse(rawText);
+    const parsed = parseJsonBody(rawText);
+    if (!parsed.ok) {
+      return c.json({ error: "Invalid JSON in request body" }, 400);
+    }
+    const body = parsed.data;
     if (!body.messages?.length) {
       return c.json({ error: "messages array is required" }, 400);
     }
@@ -11858,13 +13171,28 @@ function createRoutes(deps) {
       role: m.role,
       content: m.content
     }));
-    const response = await deps.gateway.complete({
-      messages,
-      model: body.model,
-      system: body.system,
-      maxTokens: body.maxTokens,
-      temperature: body.temperature
-    });
+    if (body.stream) {
+      const stream2 = deps.gateway.stream({
+        messages,
+        model: body.model,
+        system: body.system,
+        maxTokens: body.maxTokens,
+        temperature: body.temperature
+      });
+      return streamResponse(c, stream2);
+    }
+    let response;
+    try {
+      response = await deps.gateway.complete({
+        messages,
+        model: body.model,
+        system: body.system,
+        maxTokens: body.maxTokens,
+        temperature: body.temperature
+      });
+    } catch (err2) {
+      return elsiumErrorResponse(c, err2, "Completion failed");
+    }
     deps.tracer?.trackLLMCall({
       model: response.model,
       inputTokens: response.usage.inputTokens,
@@ -11893,9 +13221,18 @@ function createRoutes(deps) {
 }
 
 // ../app/src/app.ts
-var log5 = createLogger();
+var log12 = createLogger();
 function createApp(config) {
   const app = new Hono2;
+  app.onError((err2, c) => {
+    const statusCode = err2 instanceof ElsiumError ? err2.statusCode ?? 500 : 500;
+    const code = err2 instanceof ElsiumError ? err2.code : "UNKNOWN";
+    log12.error("Unhandled error", { error: err2.message, code, path: c.req.path });
+    return c.json({ error: err2.message, code }, statusCode);
+  });
+  app.notFound((c) => {
+    return c.json({ error: "Not found" }, 404);
+  });
   const providerNames = Object.keys(config.gateway.providers);
   const primaryProvider = providerNames[0];
   const primaryConfig = config.gateway.providers[primaryProvider];
@@ -11910,6 +13247,8 @@ function createApp(config) {
     costTracking: config.observe?.costTracking ?? true
   });
   const serverConfig = config.server ?? {};
+  app.use("*", requestIdMiddleware());
+  app.use("*", requestLoggerMiddleware(log12));
   if (serverConfig.cors) {
     app.use("*", corsMiddleware(serverConfig.cors));
   }
@@ -11932,7 +13271,7 @@ function createApp(config) {
     defaultAgent,
     tracer,
     startTime: Date.now(),
-    version: "0.1.0",
+    version: config.version ?? "0.2.2",
     providers: providerNames
   });
   app.route("/", routes);
@@ -11948,13 +13287,25 @@ function createApp(config) {
         port: listenPort,
         hostname
       });
-      log5.info("ElsiumAI server started", {
+      let shutdownManager;
+      if (serverConfig.gracefulShutdown) {
+        const drainTimeoutMs = typeof serverConfig.gracefulShutdown === "object" ? serverConfig.gracefulShutdown.drainTimeoutMs : undefined;
+        shutdownManager = createShutdownManager({
+          drainTimeoutMs,
+          onDrainStart: () => log12.info("Draining connections..."),
+          onDrainComplete: () => log12.info("Drain complete")
+        });
+      }
+      log12.info("ElsiumAI server started", {
         url: `http://${hostname}:${listenPort}`,
         routes: ["POST /chat", "POST /complete", "GET /health", "GET /metrics", "GET /agents"]
       });
       return {
         port: listenPort,
-        stop: () => {
+        stop: async () => {
+          if (shutdownManager) {
+            await shutdownManager.shutdown();
+          }
           server.close();
         }
       };
@@ -11962,7 +13313,54 @@ function createApp(config) {
   };
 }
 // ../app/src/rbac.ts
-var log6 = createLogger();
+var log13 = createLogger();
+// ../app/src/tenant.ts
+var log14 = createLogger();
+function tenantMiddleware(config) {
+  const { extractTenant, onUnknownTenant = "reject", defaultTenant } = config;
+  return async (c, next) => {
+    const tenant = extractTenant(c);
+    if (!tenant) {
+      if (onUnknownTenant === "default" && defaultTenant) {
+        c.set("tenant", defaultTenant);
+        log14.debug("Using default tenant", { tenantId: defaultTenant.tenantId });
+      } else {
+        return c.json({ error: "Tenant identification required" }, 401);
+      }
+    } else {
+      c.set("tenant", tenant);
+      log14.debug("Tenant identified", { tenantId: tenant.tenantId });
+    }
+    await next();
+  };
+}
+function tenantRateLimitMiddleware() {
+  const windows = new Map;
+  return async (c, next) => {
+    const tenant = c.get("tenant");
+    if (!tenant?.limits?.maxRequestsPerMinute) {
+      await next();
+      return;
+    }
+    const limit = tenant.limits.maxRequestsPerMinute;
+    const now = Date.now();
+    const windowMs = 60000;
+    const key = tenant.tenantId;
+    let entry = windows.get(key);
+    if (!entry || now - entry.windowStart > windowMs) {
+      entry = { count: 0, windowStart: now };
+      windows.set(key, entry);
+    }
+    entry.count++;
+    if (entry.count > limit) {
+      return c.json({
+        error: "Rate limit exceeded",
+        retryAfterMs: windowMs - (now - entry.windowStart)
+      }, 429);
+    }
+    await next();
+  };
+}
 // ../mcp/src/client.ts
 import { spawn } from "node:child_process";
 function createMCPClient(config) {
@@ -12174,7 +13572,7 @@ function createMCPClient(config) {
   };
 }
 // ../mcp/src/server.ts
-var log7 = createLogger();
+var log15 = createLogger();
 function createMCPServer(config) {
   let running = false;
   const toolMap = new Map(config.tools.map((t) => [t.name, t]));
@@ -12309,7 +13707,7 @@ function createMCPServer(config) {
           pendingChunks.shift();
           buffer += chunk;
           if (buffer.length > MAX_BUFFER_SIZE2) {
-            log7.error("MCP server: buffer size limit exceeded, resetting");
+            log15.error("MCP server: buffer size limit exceeded, resetting");
             buffer = "";
             continue;
           }
@@ -12344,6 +13742,119 @@ function createMCPServer(config) {
     }
   };
 }
+// ../client/dist/index.js
+async function* parseSSEStream(response) {
+  if (!response.body) {
+    throw new Error("Response body is null");
+  }
+  const reader = response.body.getReader();
+  const decoder = new TextDecoder;
+  let buffer = "";
+  try {
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done)
+        break;
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split(`
+`);
+      buffer = lines.pop() ?? "";
+      for (const line of lines) {
+        if (line.startsWith("event: error")) {
+          continue;
+        }
+        if (!line.startsWith("data: "))
+          continue;
+        const data = line.slice(6).trim();
+        if (!data || data === "[DONE]")
+          continue;
+        try {
+          const event = JSON.parse(data);
+          yield event;
+        } catch {}
+      }
+    }
+  } finally {
+    reader.releaseLock();
+  }
+}
+function createClient(config) {
+  const { baseUrl, apiKey, timeout = 30000 } = config;
+  function headers() {
+    const h = {
+      "Content-Type": "application/json"
+    };
+    if (apiKey) {
+      h.Authorization = `Bearer ${apiKey}`;
+    }
+    return h;
+  }
+  async function request(method, path, body) {
+    const controller = new AbortController;
+    const timer = setTimeout(() => controller.abort(), timeout);
+    try {
+      const response = await fetch(`${baseUrl}${path}`, {
+        method,
+        headers: headers(),
+        body: body ? JSON.stringify(body) : undefined,
+        signal: controller.signal
+      });
+      if (!response.ok) {
+        const errorBody = await response.text().catch(() => "Unknown error");
+        throw new Error(`HTTP ${response.status}: ${errorBody}`);
+      }
+      return await response.json();
+    } finally {
+      clearTimeout(timer);
+    }
+  }
+  async function streamRequest(path, body) {
+    const controller = new AbortController;
+    const timer = setTimeout(() => controller.abort(), timeout);
+    try {
+      const response = await fetch(`${baseUrl}${path}`, {
+        method: "POST",
+        headers: headers(),
+        body: JSON.stringify(body),
+        signal: controller.signal
+      });
+      if (!response.ok) {
+        const errorBody = await response.text().catch(() => "Unknown error");
+        clearTimeout(timer);
+        throw new Error(`HTTP ${response.status}: ${errorBody}`);
+      }
+      return response;
+    } catch (err2) {
+      clearTimeout(timer);
+      throw err2;
+    }
+  }
+  return {
+    async chat(req) {
+      return request("POST", "/chat", { ...req, stream: false });
+    },
+    async* chatStream(req) {
+      const response = await streamRequest("/chat", { ...req, stream: true });
+      yield* parseSSEStream(response);
+    },
+    async complete(req) {
+      return request("POST", "/complete", { ...req, stream: false });
+    },
+    async* completeStream(req) {
+      const response = await streamRequest("/complete", { ...req, stream: true });
+      yield* parseSSEStream(response);
+    },
+    async health() {
+      return request("GET", "/health");
+    },
+    async metrics() {
+      return request("GET", "/metrics");
+    },
+    async agents() {
+      return request("GET", "/agents");
+    }
+  };
+}
 // ../testing/src/mock-provider.ts
 function mockProvider(options = {}) {
   const { responses = [], defaultResponse, onRequest } = options;
@@ -12452,10 +13963,10 @@ function mockProvider(options = {}) {
   };
 }
 // ../testing/src/fixtures.ts
-import { createHash } from "node:crypto";
+import { createHash as createHash3 } from "node:crypto";
 function hashMessages(messages) {
   const content = messages.map((m) => `${m.role}:${m.content}`).join("|");
-  return createHash("sha256").update(content).digest("hex").slice(0, 16);
+  return createHash3("sha256").update(content).digest("hex").slice(0, 16);
 }
 function createFixture(name, entries) {
   return {
@@ -12965,7 +14476,7 @@ function createPromptRegistry() {
   };
 }
 // ../testing/src/regression.ts
-import { mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { mkdirSync, readFileSync, writeFileSync as writeFileSync2 } from "node:fs";
 import { dirname } from "node:path";
 function makeEmptyResult(name) {
   return {
@@ -13050,7 +14561,7 @@ function createRegressionSuite(name) {
         };
       }
       mkdirSync(dirname(path), { recursive: true });
-      writeFileSync(path, JSON.stringify(baseline, null, 2));
+      writeFileSync2(path, JSON.stringify(baseline, null, 2));
     },
     addCase(input, output, score) {
       if (!baseline) {
@@ -13121,7 +14632,9 @@ function createReplayPlayer(entriesOrJson) {
   };
 }
 export {
+  zodToJsonSchema,
   xrayMiddleware,
+  vectorStoreRegistry,
   unwrapOr,
   unwrap,
   tryCatchSync,
@@ -13129,7 +14642,11 @@ export {
   toTraceparent,
   toOTelSpan,
   toOTelExportRequest,
+  tenantRateLimitMiddleware,
+  tenantMiddleware,
+  streamResponse,
   step,
+  sseHeaders,
   sleep,
   securityMiddleware,
   runSupervisor,
@@ -13142,6 +14659,7 @@ export {
   redactSecrets,
   rag,
   parseTraceparent,
+  outputGuardrailMiddleware,
   ok,
   observe,
   mockProvider,
@@ -13157,6 +14675,7 @@ export {
   gateway,
   formatToolResultAsText,
   formatToolResult,
+  formatSSE,
   formatEvalReport,
   extractTraceContext,
   extractText,
@@ -13165,6 +14684,7 @@ export {
   envNumber,
   envBool,
   env,
+  embeddingProviderRegistry,
   detectPromptInjection,
   detectJailbreak,
   defineWorkflow,
@@ -13176,15 +14696,18 @@ export {
   currentTimeTool,
   createToolkit,
   createStream,
+  createSqliteMemoryStore,
   createSpan,
   createSnapshotStore,
   createSemanticValidator,
   createReplayRecorder,
   createReplayPlayer,
   createRegressionSuite,
+  createRegistry,
   createRecorder,
   createProviderMesh,
   createPromptRegistry,
+  createPgVectorStore,
   createOpenAIProvider,
   createOpenAIEmbeddings,
   createOTLPExporter,
@@ -13195,17 +14718,25 @@ export {
   createMCPClient,
   createLogger,
   createInMemoryStore,
+  createInMemoryMemoryStore,
+  createInMemoryCache,
   createGoogleProvider,
   createFixture,
+  createExperiment,
   createCostEngine,
+  createContextManager,
   createConfidenceScorer,
+  createClient,
+  createBatch,
   createApp,
   createAnthropicProvider,
   createAgentSecurity,
+  countTokens,
   costTrackingMiddleware,
   composeMiddleware,
   checkBlockedPatterns,
   calculatorTool,
   calculateCost,
+  cacheMiddleware,
   ElsiumError
 };