@elsium-ai/app 0.2.3 → 0.4.0

package/dist/index.js

@@ -389,6 +389,94 @@ function createLogger(options = {}) {
     }
   };
 }
+// ../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"]);
 // ../core/src/shutdown.ts
 function createShutdownManager(config) {
   const drainTimeoutMs = config?.drainTimeoutMs ?? 30000;
@@ -487,6 +575,12 @@ function createShutdownManager(config) {
 // ../gateway/src/provider.ts
 var providerRegistry = new Map;
 var metadataRegistry = new Map;
+function getProviderFactory(name) {
+  return providerRegistry.get(name);
+}
+function listProviders() {
+  return Array.from(providerRegistry.keys());
+}
 function registerProviderMetadata(name, metadata) {
   metadataRegistry.set(name, metadata);
 }
@@ -512,6 +606,17 @@ function composeMiddleware(middlewares) {
     return dispatch(0);
   };
 }
+function composeStreamMiddleware(middlewares) {
+  return (ctx, source, finalNext) => {
+    function dispatch(i, currentCtx, currentSource) {
+      if (i >= middlewares.length) {
+        return finalNext(currentCtx, currentSource);
+      }
+      return middlewares[i](currentCtx, currentSource, (c, s) => dispatch(i + 1, c, s));
+    }
+    return dispatch(0, ctx, source);
+  };
+}
 var SENSITIVE_HEADERS = ["x-api-key", "authorization", "api-key"];
 function redactHeaders(headers) {
   const redacted = {};
@@ -615,7 +720,7 @@ 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 },
@@ -653,7 +758,7 @@ 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,
@@ -747,15 +852,33 @@ 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) {
@@ -798,6 +921,52 @@ function createAnthropicProvider(config) {
       input_schema: t.inputSchema
     }));
   }
+  function buildOptionalParams(req) {
+    const params = {};
+    if (req.temperature !== undefined)
+      params.temperature = req.temperature;
+    if (req.topP !== undefined)
+      params.top_p = req.topP;
+    if (req.stopSequences?.length)
+      params.stop_sequences = req.stopSequences;
+    return params;
+  }
+  function applyStructuredOutput(body, req, tools) {
+    if (!req.schema)
+      return;
+    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" };
+  }
+  function buildRequestBody(req) {
+    const { system, messages } = formatMessages(req.messages);
+    const model = req.model ?? "claude-sonnet-4-6";
+    const body = {
+      model,
+      messages,
+      max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS,
+      ...system || req.system ? { system: req.system ?? system } : {},
+      ...buildOptionalParams(req),
+      ...buildSeedMetadata(req)
+    };
+    const tools = formatTools(req.tools);
+    if (tools)
+      body.tools = tools;
+    applyStructuredOutput(body, req, tools);
+    return body;
+  }
+  function executeWithTimeout(fn, reqSignal) {
+    const controller = new AbortController;
+    const timer = setTimeout(() => controller.abort(), timeout);
+    const signals = [controller.signal, reqSignal].filter(Boolean);
+    const mergedSignal = signals.length > 1 ? AbortSignal.any(signals) : signals[0];
+    return fn(mergedSignal).finally(() => clearTimeout(timer));
+  }
   function extractContentBlocks(content) {
     const toolCalls = [];
     const textParts = [];
@@ -849,34 +1018,12 @@ function createAnthropicProvider(config) {
       authStyle: "x-api-key"
     },
     async complete(req) {
-      const { system, messages } = formatMessages(req.messages);
-      const model = req.model ?? "claude-sonnet-4-6";
-      const body = {
-        model,
-        messages,
-        max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS,
-        ...system || req.system ? { system: req.system ?? system } : {},
-        ...req.temperature !== undefined ? { temperature: req.temperature } : {},
-        ...req.topP !== undefined ? { top_p: req.topP } : {},
-        ...req.stopSequences?.length ? { stop_sequences: req.stopSequences } : {},
-        ...buildSeedMetadata(req)
-      };
-      const tools = formatTools(req.tools);
-      if (tools)
-        body.tools = tools;
+      const body = buildRequestBody(req);
       const startTime = performance.now();
-      const raw = await retry(async () => {
-        const controller = new AbortController;
-        const timer = setTimeout(() => controller.abort(), timeout);
-        try {
-          const signals = [controller.signal, req.signal].filter(Boolean);
-          const mergedSignal = signals.length > 1 ? AbortSignal.any(signals) : signals[0];
-          const resp = await request("/messages", body, mergedSignal);
-          return await resp.json();
-        } finally {
-          clearTimeout(timer);
-        }
-      }, {
+      const raw = await retry(() => executeWithTimeout(async (signal) => {
+        const resp = await request("/messages", body, signal);
+        return await resp.json();
+      }, req.signal), {
         maxRetries,
         baseDelayMs: 1000,
         shouldRetry: (e) => e instanceof ElsiumError && e.retryable
@@ -885,29 +1032,12 @@ function createAnthropicProvider(config) {
       return parseResponse(raw, latencyMs);
     },
     stream(req) {
-      const { system, messages } = formatMessages(req.messages);
-      const model = req.model ?? "claude-sonnet-4-6";
-      const body = {
-        model,
-        messages,
-        max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS,
-        stream: true,
-        ...system || req.system ? { system: req.system ?? system } : {},
-        ...req.temperature !== undefined ? { temperature: req.temperature } : {},
-        ...req.topP !== undefined ? { top_p: req.topP } : {},
-        ...req.stopSequences?.length ? { stop_sequences: req.stopSequences } : {},
-        ...buildSeedMetadata(req)
-      };
-      const tools = formatTools(req.tools);
-      if (tools)
-        body.tools = tools;
+      const body = buildRequestBody(req);
+      body.stream = true;
+      const model = body.model ?? "claude-sonnet-4-6";
       return createStream(async (emit) => {
-        const controller = new AbortController;
-        const timer = setTimeout(() => controller.abort(), timeout);
-        try {
-          const signals = [controller.signal, req.signal].filter(Boolean);
-          const mergedSignal = signals.length > 1 ? AbortSignal.any(signals) : signals[0];
-          const resp = await request("/messages", body, mergedSignal);
+        await executeWithTimeout(async (signal) => {
+          const resp = await request("/messages", body, signal);
           if (!resp.body)
             throw new ElsiumError({
               code: "STREAM_ERROR",
@@ -916,9 +1046,7 @@ function createAnthropicProvider(config) {
               retryable: false
             });
           await processAnthropicSSEStream(resp.body, model, emit);
-        } finally {
-          clearTimeout(timer);
-        }
+        }, req.signal);
       });
     },
     async listModels() {
@@ -1072,19 +1200,38 @@ function createGoogleProvider(config) {
     }
     return { role, parts };
   }
+  function convertGeminiImagePart(p) {
+    if (p.source.type === "base64") {
+      return { inlineData: { mimeType: p.source.mediaType, data: p.source.data } };
+    }
+    return { fileData: { mimeType: "image/jpeg", fileUri: p.source.url } };
+  }
+  function convertGeminiMediaPart(p) {
+    if (p.source.type === "base64") {
+      return { inlineData: { mimeType: p.source.mediaType, data: p.source.data } };
+    }
+    const urlSource = p.source;
+    return { fileData: { mimeType: "application/octet-stream", fileUri: urlSource.url } };
+  }
+  function convertGeminiContentPart(p) {
+    if (p.type === "text") {
+      return { text: p.text };
+    }
+    if (p.type === "image") {
+      return convertGeminiImagePart(p);
+    }
+    if (p.type === "audio" || p.type === "document") {
+      return convertGeminiMediaPart(p);
+    }
+    return null;
+  }
   function formatGeminiMultipartContent(msg, role) {
+    const content = msg.content;
     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 } });
-        }
-      }
+    for (const p of content) {
+      const converted = convertGeminiContentPart(p);
+      if (converted)
+        parts.push(converted);
     }
     return { role, parts };
   }
@@ -1182,6 +1329,10 @@ 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) {
@@ -1447,21 +1598,48 @@ function createOpenAIProvider(config) {
     }
     return openaiMsg;
   }
+  function convertImagePart(part) {
+    if (part.source.type === "base64") {
+      const url = `data:${part.source.mediaType};base64,${part.source.data}`;
+      return { type: "image_url", image_url: { url } };
+    }
+    return { type: "image_url", image_url: { url: part.source.url } };
+  }
+  function convertAudioPart(part) {
+    if (part.source.type === "base64") {
+      const format = part.source.mediaType.split("/")[1] ?? "wav";
+      return { type: "input_audio", input_audio: { data: part.source.data, format } };
+    }
+    return { type: "text", text: "[audio: url source requires file upload]" };
+  }
+  function convertDocumentPart(part) {
+    if (part.source.type === "base64") {
+      return {
+        type: "text",
+        text: `[document: ${part.source.mediaType} content attached as base64]`
+      };
+    }
+    return { type: "text", text: `[document: ${part.source.url}]` };
+  }
+  function convertContentPart(part) {
+    if (part.type === "text")
+      return { type: "text", text: part.text };
+    if (part.type === "image")
+      return convertImagePart(part);
+    if (part.type === "audio")
+      return convertAudioPart(part);
+    if (part.type === "document")
+      return convertDocumentPart(part);
+    return null;
+  }
   function formatUserContent(msg) {
     if (typeof msg.content === "string")
       return msg.content;
     const parts = [];
     for (const part of msg.content) {
-      if (part.type === "text") {
-        parts.push({ type: "text", text: part.text });
-      } else if (part.type === "image") {
-        if (part.source.type === "base64") {
-          const url = `data:${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 } });
-        }
-      }
+      const converted = convertContentPart(part);
+      if (converted)
+        parts.push(converted);
     }
     return parts;
   }
@@ -1496,6 +1674,49 @@ function createOpenAIProvider(config) {
       }
     }));
   }
+  function buildOptionalParams(req) {
+    const params = {};
+    if (req.temperature !== undefined)
+      params.temperature = req.temperature;
+    if (req.seed !== undefined)
+      params.seed = req.seed;
+    if (req.topP !== undefined)
+      params.top_p = req.topP;
+    if (req.stopSequences?.length)
+      params.stop = req.stopSequences;
+    return params;
+  }
+  function applyResponseFormat(body, req) {
+    if (!req.schema)
+      return;
+    const jsonSchema = zodToJsonSchema(req.schema);
+    body.response_format = {
+      type: "json_schema",
+      json_schema: {
+        name: "structured_output",
+        strict: true,
+        schema: jsonSchema
+      }
+    };
+  }
+  function buildRequestBody(req) {
+    const messages = formatMessages(req.messages);
+    const model = req.model ?? "gpt-4o";
+    if (req.system) {
+      messages.unshift({ role: "system", content: req.system });
+    }
+    const body = {
+      model,
+      messages,
+      max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS2,
+      ...buildOptionalParams(req)
+    };
+    const tools = formatTools(req.tools);
+    if (tools)
+      body.tools = tools;
+    applyResponseFormat(body, req);
+    return body;
+  }
   function parseResponse(raw, latencyMs) {
     const traceId = generateTraceId();
     const choice = raw.choices[0];
@@ -1540,23 +1761,7 @@ function createOpenAIProvider(config) {
       authStyle: "bearer"
     },
     async complete(req) {
-      const messages = formatMessages(req.messages);
-      const model = req.model ?? "gpt-4o";
-      if (req.system) {
-        messages.unshift({ role: "system", content: req.system });
-      }
-      const body = {
-        model,
-        messages,
-        max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS2,
-        ...req.temperature !== undefined ? { temperature: req.temperature } : {},
-        ...req.seed !== undefined ? { seed: req.seed } : {},
-        ...req.topP !== undefined ? { top_p: req.topP } : {},
-        ...req.stopSequences?.length ? { stop: req.stopSequences } : {}
-      };
-      const tools = formatTools(req.tools);
-      if (tools)
-        body.tools = tools;
+      const body = buildRequestBody(req);
       const startTime = performance.now();
       const raw = await retry(async () => {
         const controller = new AbortController;
@@ -1578,25 +1783,10 @@ function createOpenAIProvider(config) {
       return parseResponse(raw, latencyMs);
     },
     stream(req) {
-      const messages = formatMessages(req.messages);
-      const model = req.model ?? "gpt-4o";
-      if (req.system) {
-        messages.unshift({ role: "system", content: req.system });
-      }
-      const body = {
-        model,
-        messages,
-        max_tokens: req.maxTokens ?? DEFAULT_MAX_TOKENS2,
-        stream: true,
-        stream_options: { include_usage: true },
-        ...req.temperature !== undefined ? { temperature: req.temperature } : {},
-        ...req.seed !== undefined ? { seed: req.seed } : {},
-        ...req.topP !== undefined ? { top_p: req.topP } : {},
-        ...req.stopSequences?.length ? { stop: req.stopSequences } : {}
-      };
-      const tools = formatTools(req.tools);
-      if (tools)
-        body.tools = tools;
+      const body = buildRequestBody(req);
+      body.stream = true;
+      body.stream_options = { include_usage: true };
+      const model = body.model ?? "gpt-4o";
       return createStream(async (emit) => {
         const controller = new AbortController;
         const timer = setTimeout(() => controller.abort(), timeout);
@@ -1716,12 +1906,29 @@ var PROVIDER_FACTORIES = {
   openai: createOpenAIProvider,
   google: createGoogleProvider
 };
+registerProviderMetadata("anthropic", {
+  baseUrl: "https://api.anthropic.com/v1/messages",
+  capabilities: ["tools", "vision", "streaming", "system"],
+  authStyle: "x-api-key"
+});
+registerProviderMetadata("openai", {
+  baseUrl: "https://api.openai.com/v1/chat/completions",
+  capabilities: ["tools", "vision", "streaming", "system", "json_mode"],
+  authStyle: "bearer"
+});
+registerProviderMetadata("google", {
+  baseUrl: "https://generativelanguage.googleapis.com/v1beta/models",
+  capabilities: ["tools", "vision", "streaming", "system"],
+  authStyle: "bearer"
+});
 function validateGatewayConfig(config) {
-  const factory = PROVIDER_FACTORIES[config.provider];
+  const factory = PROVIDER_FACTORIES[config.provider] ?? getProviderFactory(config.provider);
   if (!factory) {
+    const available = [...Object.keys(PROVIDER_FACTORIES), ...listProviders()];
+    const unique = [...new Set(available)];
     throw new ElsiumError({
       code: "CONFIG_ERROR",
-      message: `Unknown provider: ${config.provider}. Available: ${Object.keys(PROVIDER_FACTORIES).join(", ")}`,
+      message: `Unknown provider: ${config.provider}. Available: ${unique.join(", ")}`,
       retryable: false
     });
   }
@@ -1799,6 +2006,24 @@ async function accumulateStreamEvents(stream, emit) {
   }
   return { textContent, usage, stopReason, id };
 }
+function extractFromToolCalls(response) {
+  if (response.stopReason !== "tool_use" || !response.message.toolCalls?.length) {
+    return;
+  }
+  const structuredCall = response.message.toolCalls.find((tc) => tc.name === "_structured_output");
+  return structuredCall?.arguments;
+}
+function extractJsonFromText(response) {
+  let text = typeof response.message.content === "string" ? response.message.content : "";
+  text = text.replace(/^```(?:json)?\s*\n?([\s\S]*?)\n?\s*```$/gm, "$1").trim();
+  const jsonMatch = text.match(/(\{[\s\S]*\}|\[[\s\S]*\])/);
+  if (!jsonMatch) {
+    throw ElsiumError.validation("LLM response did not contain valid JSON", {
+      response: text
+    });
+  }
+  return JSON.parse(jsonMatch[0]);
+}
 function gateway(config) {
   const factory = validateGatewayConfig(config);
   const provider = factory({
@@ -1820,6 +2045,7 @@ function gateway(config) {
     allMiddleware.push(xm);
   }
   const composedMiddleware = allMiddleware.length ? composeMiddleware(allMiddleware) : null;
+  const composedStreamMiddleware = config.streamMiddleware?.length ? composeStreamMiddleware(config.streamMiddleware) : null;
   async function executeWithMiddleware(request) {
     const req = { ...request, model: request.model ?? defaultModel };
     if (!composedMiddleware) {
@@ -1844,11 +2070,11 @@ function gateway(config) {
       validateRequestLimits(request, maxMessages, maxInputTokens);
       const req = { ...request, model: request.model ?? defaultModel };
       if (composedMiddleware) {
-        const ctx = buildMiddlewareContext(req, provider.name, defaultModel, request.metadata ?? {});
+        const ctx2 = buildMiddlewareContext(req, provider.name, defaultModel, request.metadata ?? {});
         return createStream(async (emit) => {
-          await composedMiddleware(ctx, async (c) => {
+          await composedMiddleware(ctx2, async (c) => {
             const result = await accumulateStreamEvents(provider.stream(c.request), emit);
-            const latencyMs = Math.round(performance.now() - ctx.startTime);
+            const latencyMs = Math.round(performance.now() - ctx2.startTime);
             return {
               id: result.id,
               message: { role: "assistant", content: result.textContent },
@@ -1858,116 +2084,54 @@ function gateway(config) {
               provider: provider.name,
               stopReason: result.stopReason,
               latencyMs,
-              traceId: ctx.traceId
+              traceId: ctx2.traceId
             };
           });
         });
       }
-      return provider.stream(req);
+      const rawStream = provider.stream(req);
+      if (!composedStreamMiddleware)
+        return rawStream;
+      const ctx = buildMiddlewareContext(req, provider.name, defaultModel, request.metadata ?? {});
+      return createStream(async (emit) => {
+        const processed = composedStreamMiddleware(ctx, rawStream, (_c, s) => s);
+        for await (const event of processed) {
+          emit(event);
+        }
+      });
     },
     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
-        });
-      }
-      const parsed = JSON.parse(jsonMatch[0]);
+      const parsed = extractFromToolCalls(response) ?? extractJsonFromText(response);
       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/cache.ts
+var log4 = createLogger();
+// ../gateway/src/output-guardrails.ts
+var log5 = createLogger();
+// ../gateway/src/batch.ts
+var log6 = createLogger();
 // ../observe/src/span.ts
 function createSpan(name, options = {}) {
   const id = generateId("spn");
@@ -2038,7 +2202,7 @@ function createSpan(name, options = {}) {
 }
 // ../observe/src/tracer.ts
 import { writeFileSync } from "node:fs";
-var log2 = createLogger();
+var log7 = createLogger();
 function observe(config = {}) {
   const {
     output = ["console"],
@@ -2061,7 +2225,7 @@ function observe(config = {}) {
           try {
             writeFileSync(filename, JSON.stringify(spansToExport, null, 2));
           } catch (err2) {
-            log2.error("Failed to write trace file", {
+            log7.error("Failed to write trace file", {
               error: err2 instanceof Error ? err2.message : String(err2)
             });
           }
@@ -2136,7 +2300,7 @@ 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]" : "[...]";
-  log2.info("span", {
+  log7.info("span", {
     trace: span.traceId,
     span: span.name,
     kind: span.kind,
@@ -2173,9 +2337,11 @@ function createNoopSpan(name, kind) {
     }
   };
 }
+// ../observe/src/experiment.ts
+var log8 = createLogger();
 // ../observe/src/otel.ts
-var log3 = createLogger();
-// ../../node_modules/.bun/@hono+node-server@1.19.9/node_modules/@hono/node-server/dist/index.mjs
+var log9 = createLogger();
+// ../../node_modules/.bun/@hono+node-server@1.19.10/node_modules/@hono/node-server/dist/index.mjs
 import { createServer as createServerHTTP } from "http";
 import { Http2ServerRequest as Http2ServerRequest2 } from "http2";
 import { Http2ServerRequest } from "http2";
@@ -2709,7 +2875,7 @@ var serve = (options, listeningListener) => {
   return server;
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/compose.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/compose.js
 var compose = (middleware, onError, onNotFound) => {
   return (context, next) => {
     let index = -1;
@@ -2753,10 +2919,10 @@ var compose = (middleware, onError, onNotFound) => {
   };
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/request/constants.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/request/constants.js
 var GET_MATCH_RESULT = /* @__PURE__ */ Symbol();
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/utils/body.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/utils/body.js
 var parseBody = async (request, options = /* @__PURE__ */ Object.create(null)) => {
   const { all = false, dot = false } = options;
   const headers = request instanceof HonoRequest ? request.raw.headers : request.headers;
@@ -2824,7 +2990,7 @@ var handleParsingNestedValues = (form, key, value) => {
   });
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/utils/url.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/utils/url.js
 var splitPath = (path) => {
   const paths = path.split("/");
   if (paths[0] === "") {
@@ -3024,7 +3190,7 @@ var getQueryParams = (url, key) => {
 };
 var decodeURIComponent_ = decodeURIComponent;
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/request.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/request.js
 var tryDecodeURIComponent = (str) => tryDecode(str, decodeURIComponent_);
 var HonoRequest = class {
   raw;
@@ -3135,7 +3301,7 @@ var HonoRequest = class {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/utils/html.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/utils/html.js
 var HtmlEscapedCallbackPhase = {
   Stringify: 1,
   BeforeStream: 2,
@@ -3173,7 +3339,7 @@ var resolveCallback = async (str, phase, preserveCallbacks, context, buffer) =>
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/context.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/context.js
 var TEXT_PLAIN = "text/plain; charset=UTF-8";
 var setDefaultContentType = (contentType, headers) => {
   return {
@@ -3340,7 +3506,7 @@ var Context = class {
   };
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router.js
 var METHOD_NAME_ALL = "ALL";
 var METHOD_NAME_ALL_LOWERCASE = "all";
 var METHODS = ["get", "post", "put", "delete", "options", "patch"];
@@ -3348,10 +3514,10 @@ var MESSAGE_MATCHER_IS_ALREADY_BUILT = "Can not add a route since the matcher is
 var UnsupportedPathError = class extends Error {
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/utils/constants.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/utils/constants.js
 var COMPOSED_HANDLER = "__COMPOSED_HANDLER";
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/hono-base.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/hono-base.js
 var notFoundHandler = (c) => {
   return c.text("404 Not Found", 404);
 };
@@ -3570,7 +3736,7 @@ var Hono = class _Hono {
   };
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/reg-exp-router/matcher.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/reg-exp-router/matcher.js
 var emptyParam = [];
 function match(method, path) {
   const matchers = this.buildAllMatchers();
@@ -3591,7 +3757,7 @@ function match(method, path) {
   return match2(method, path);
 }
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/reg-exp-router/node.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/reg-exp-router/node.js
 var LABEL_REG_EXP_STR = "[^/]+";
 var ONLY_WILDCARD_REG_EXP_STR = ".*";
 var TAIL_WILDCARD_REG_EXP_STR = "(?:|/.*)";
@@ -3695,7 +3861,7 @@ var Node = class _Node {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/reg-exp-router/trie.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/reg-exp-router/trie.js
 var Trie = class {
   #context = { varIndex: 0 };
   #root = new Node;
@@ -3751,7 +3917,7 @@ var Trie = class {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/reg-exp-router/router.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/reg-exp-router/router.js
 var nullMatcher = [/^$/, [], /* @__PURE__ */ Object.create(null)];
 var wildcardRegExpCache = /* @__PURE__ */ Object.create(null);
 function buildWildcardRegExp(path) {
@@ -3916,7 +4082,7 @@ var RegExpRouter = class {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/reg-exp-router/prepared-router.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/reg-exp-router/prepared-router.js
 var PreparedRegExpRouter = class {
   name = "PreparedRegExpRouter";
   #matchers;
@@ -3988,7 +4154,7 @@ var PreparedRegExpRouter = class {
   match = match;
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/smart-router/router.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/smart-router/router.js
 var SmartRouter = class {
   name = "SmartRouter";
   #routers = [];
@@ -4043,7 +4209,7 @@ var SmartRouter = class {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/trie-router/node.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/trie-router/node.js
 var emptyParams = /* @__PURE__ */ Object.create(null);
 var hasChildren = (children) => {
   for (const _ in children) {
@@ -4212,7 +4378,7 @@ var Node2 = class _Node2 {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/router/trie-router/router.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/router/trie-router/router.js
 var TrieRouter = class {
   name = "TrieRouter";
   #node;
@@ -4234,7 +4400,7 @@ var TrieRouter = class {
   }
 };
 
-// ../../node_modules/.bun/hono@4.12.3/node_modules/hono/dist/hono.js
+// ../../node_modules/.bun/hono@4.12.5/node_modules/hono/dist/hono.js
 var Hono2 = class extends Hono {
   constructor(options = {}) {
     super(options);
@@ -4334,12 +4500,12 @@ function requestIdMiddleware() {
   };
 }
 function requestLoggerMiddleware(logger) {
-  const log4 = logger ?? createLogger();
+  const log10 = logger ?? createLogger();
   return async (c, next) => {
     const start = Date.now();
     await next();
     const duration = Date.now() - start;
-    log4.info(`${c.req.method} ${c.req.path}`, {
+    log10.info(`${c.req.method} ${c.req.path}`, {
       method: c.req.method,
       path: c.req.path,
       status: c.res.status,
@@ -4349,6 +4515,158 @@ function requestLoggerMiddleware(logger) {
   };
 }
 
+// ../../node_modules/.bun/hono@4.12.5/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@4.12.5/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@4.12.5/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);
+};
+
+// 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: ${json}
+
+`;
+}
+function streamResponse(c, source) {
+  const headers = sseHeaders();
+  for (const [key, value] of Object.entries(headers)) {
+    c.header(key, value);
+  }
+  return stream(c, async (s) => {
+    try {
+      for await (const event of source) {
+        const sseData = formatSSE("message", event);
+        await s.write(sseData);
+      }
+    } catch (err2) {
+      const errorEvent = {
+        type: "error",
+        error: err2 instanceof Error ? err2 : new Error(String(err2))
+      };
+      const sseData = formatSSE("error", {
+        type: "error",
+        message: errorEvent.error.message
+      });
+      await s.write(sseData);
+    }
+  });
+}
+
 // src/routes.ts
 function parseJsonBody(raw2) {
   try {
@@ -4369,6 +4687,31 @@ function resolveAgent(name, agents, defaultAgent) {
     return { agent };
   return { error: name ? `Agent "${name}" not found` : "No default agent configured" };
 }
+var MAX_BODY_SIZE = 1048576;
+function parseRequestBody(c, rawText) {
+  if (rawText.length > MAX_BODY_SIZE) {
+    return { ok: false, response: c.json({ error: "Request body too large (max 1MB)" }, 413) };
+  }
+  const parsed = parseJsonBody(rawText);
+  if (!parsed.ok) {
+    return { ok: false, response: c.json({ error: "Invalid JSON in request body" }, 400) };
+  }
+  return { ok: true, data: parsed.data };
+}
+function buildChatResponse(result, model) {
+  const content = typeof result.message.content === "string" ? result.message.content : "";
+  return {
+    message: content,
+    usage: {
+      inputTokens: result.usage.totalInputTokens,
+      outputTokens: result.usage.totalOutputTokens,
+      totalTokens: result.usage.totalTokens,
+      cost: result.usage.totalCost
+    },
+    model: model ?? "default",
+    traceId: result.traceId
+  };
+}
 function createRoutes(deps) {
   const app = new Hono2;
   let totalRequests = 0;
@@ -4404,15 +4747,10 @@ function createRoutes(deps) {
   });
   app.post("/chat", async (c) => {
     totalRequests++;
-    const MAX_BODY_SIZE = 1048576;
     const rawText = await c.req.text();
-    if (rawText.length > MAX_BODY_SIZE) {
-      return c.json({ error: "Request body too large (max 1MB)" }, 413);
-    }
-    const parsed = parseJsonBody(rawText);
-    if (!parsed.ok) {
-      return c.json({ error: "Invalid JSON in request body" }, 400);
-    }
+    const parsed = parseRequestBody(c, rawText);
+    if (!parsed.ok)
+      return parsed.response;
     const body = parsed.data;
     if (!body.message) {
       return c.json({ error: "message is required" }, 400);
@@ -4421,6 +4759,14 @@ function createRoutes(deps) {
     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);
@@ -4428,37 +4774,20 @@ function createRoutes(deps) {
       return elsiumErrorResponse(c, err2, "Agent execution failed");
     }
     deps.tracer?.trackLLMCall({
-      model: "unknown",
+      model: resolved.agent.config.model ?? "unknown",
       inputTokens: result.usage.totalInputTokens,
       outputTokens: result.usage.totalOutputTokens,
       cost: result.usage.totalCost,
       latencyMs: 0
     });
-    const content = typeof result.message.content === "string" ? result.message.content : "";
-    const response = {
-      message: content,
-      usage: {
-        inputTokens: result.usage.totalInputTokens,
-        outputTokens: result.usage.totalOutputTokens,
-        totalTokens: result.usage.totalTokens,
-        cost: result.usage.totalCost
-      },
-      model: resolved.agent.config.model ?? "default",
-      traceId: result.traceId
-    };
-    return c.json(response);
+    return c.json(buildChatResponse(result, resolved.agent.config.model));
   });
   app.post("/complete", async (c) => {
     totalRequests++;
-    const MAX_BODY_SIZE = 1048576;
     const rawText = await c.req.text();
-    if (rawText.length > MAX_BODY_SIZE) {
-      return c.json({ error: "Request body too large (max 1MB)" }, 413);
-    }
-    const parsed = parseJsonBody(rawText);
-    if (!parsed.ok) {
-      return c.json({ error: "Invalid JSON in request body" }, 400);
-    }
+    const parsed = parseRequestBody(c, rawText);
+    if (!parsed.ok)
+      return parsed.response;
     const body = parsed.data;
     if (!body.messages?.length) {
       return c.json({ error: "messages array is required" }, 400);
@@ -4467,6 +4796,16 @@ function createRoutes(deps) {
       role: m.role,
       content: m.content
     }));
+    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({
@@ -4507,13 +4846,13 @@ function createRoutes(deps) {
 }
 
 // src/app.ts
-var log4 = createLogger();
+var log10 = 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";
-    log4.error("Unhandled error", { error: err2.message, code, path: c.req.path });
+    log10.error("Unhandled error", { error: err2.message, code, path: c.req.path });
     return c.json({ error: err2.message, code }, statusCode);
   });
   app.notFound((c) => {
@@ -4534,7 +4873,7 @@ function createApp(config) {
   });
   const serverConfig = config.server ?? {};
   app.use("*", requestIdMiddleware());
-  app.use("*", requestLoggerMiddleware(log4));
+  app.use("*", requestLoggerMiddleware(log10));
   if (serverConfig.cors) {
     app.use("*", corsMiddleware(serverConfig.cors));
   }
@@ -4578,11 +4917,11 @@ function createApp(config) {
         const drainTimeoutMs = typeof serverConfig.gracefulShutdown === "object" ? serverConfig.gracefulShutdown.drainTimeoutMs : undefined;
         shutdownManager = createShutdownManager({
           drainTimeoutMs,
-          onDrainStart: () => log4.info("Draining connections..."),
-          onDrainComplete: () => log4.info("Drain complete")
+          onDrainStart: () => log10.info("Draining connections..."),
+          onDrainComplete: () => log10.info("Drain complete")
         });
       }
-      log4.info("ElsiumAI server started", {
+      log10.info("ElsiumAI server started", {
         url: `http://${hostname}:${listenPort}`,
         routes: ["POST /chat", "POST /complete", "GET /health", "GET /metrics", "GET /agents"]
       });
@@ -4599,7 +4938,7 @@ function createApp(config) {
   };
 }
 // src/rbac.ts
-var log5 = createLogger();
+var log11 = createLogger();
 var BUILT_IN_ROLES = [
   {
     name: "admin",
@@ -4637,7 +4976,7 @@ function matchPermission(granted, required) {
 }
 function createRBAC(config) {
   if (config.trustRoleHeader === true) {
-    log5.warn("RBAC: trustRoleHeader is enabled — any client can self-assign roles via the X-Role header. Only use this in development or behind a trusted reverse proxy.");
+    log11.warn("RBAC: trustRoleHeader is enabled — any client can self-assign roles via the X-Role header. Only use this in development or behind a trusted reverse proxy.");
   }
   const roleMap = new Map;
   for (const role of BUILT_IN_ROLES) {
@@ -4693,10 +5032,113 @@ function createRBAC(config) {
     }
   };
 }
+// src/tenant.ts
+var log12 = createLogger();
+var tenantUsage = new Map;
+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);
+        log12.debug("Using default tenant", { tenantId: defaultTenant.tenantId });
+      } else {
+        return c.json({ error: "Tenant identification required" }, 401);
+      }
+    } else {
+      c.set("tenant", tenant);
+      log12.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();
+  };
+}
+function getOrCreateUsage(tenantId) {
+  let usage = tenantUsage.get(tenantId);
+  if (!usage) {
+    const now = Date.now();
+    usage = {
+      minute: { tokens: 0, cost: 0, windowStart: now },
+      day: { tokens: 0, cost: 0, windowStart: now }
+    };
+    tenantUsage.set(tenantId, usage);
+  }
+  return usage;
+}
+function resetWindowIfExpired(window, durationMs) {
+  const now = Date.now();
+  if (now - window.windowStart > durationMs) {
+    window.tokens = 0;
+    window.cost = 0;
+    window.windowStart = now;
+  }
+}
+function tenantBudgetMiddleware() {
+  return async (c, next) => {
+    const tenant = c.get("tenant");
+    if (!tenant?.limits) {
+      await next();
+      return;
+    }
+    const usage = getOrCreateUsage(tenant.tenantId);
+    resetWindowIfExpired(usage.minute, 60000);
+    resetWindowIfExpired(usage.day, 86400000);
+    if (tenant.limits.maxTokensPerMinute && usage.minute.tokens >= tenant.limits.maxTokensPerMinute) {
+      return c.json({ error: "Token rate limit exceeded", retryAfterMs: 60000 }, 429);
+    }
+    if (tenant.limits.maxCostPerDay && usage.day.cost >= tenant.limits.maxCostPerDay) {
+      return c.json({ error: "Daily cost limit exceeded" }, 429);
+    }
+    await next();
+    const tokenCount = Number(c.res.headers.get("x-token-count")) || 0;
+    const cost = Number(c.res.headers.get("x-cost")) || 0;
+    if (tokenCount > 0) {
+      usage.minute.tokens += tokenCount;
+      usage.day.tokens += tokenCount;
+    }
+    if (cost > 0) {
+      usage.minute.cost += cost;
+      usage.day.cost += cost;
+    }
+  };
+}
 export {
+  tenantRateLimitMiddleware,
+  tenantMiddleware,
+  tenantBudgetMiddleware,
+  streamResponse,
+  sseHeaders,
   requestLoggerMiddleware,
   requestIdMiddleware,
   rateLimitMiddleware,
+  formatSSE,
   createRoutes,
   createRBAC,
   createApp,