@hebo-ai/gateway 0.4.1 → 0.4.2

package/README.md

@@ -286,10 +286,9 @@ const gw = gateway({
     /**
      * Runs before any endpoint handler logic.
      * @param ctx.request Incoming request.
-     * @returns Optional RequestPatch to merge into headers / override body.
-     * Returning a Response stops execution of the endpoint.
+     * @returns Optional Response to short-circuit the request.
      */
-    onRequest: async (ctx: { request: Request }): Promise<RequestPatch | Response | void> => {
+    onRequest: async (ctx: { request: Request }): Promise<Response | void> => {
       // Example Use Cases:
       // - Verify authentication
       // - Enforce rate limits

package/dist/endpoints/chat-completions/handler.js

@@ -7,7 +7,6 @@ import { modelMiddlewareMatcher } from "../../middleware/matcher";
 import { resolveProvider } from "../../providers/registry";
 import { recordRequestDuration, recordTimePerOutputToken, recordTokenUsage, } from "../../telemetry/gen-ai";
 import { addSpanEvent, setSpanAttributes } from "../../telemetry/span";
-import { resolveRequestId } from "../../utils/headers";
 import { prepareForwardHeaders } from "../../utils/request";
 import { convertToTextCallOptions, toChatCompletions, toChatCompletionsStream } from "./converters";
 import { getChatGeneralAttributes, getChatRequestAttributes, getChatResponseAttributes, } from "./otel";
@@ -22,7 +21,6 @@ export const chatCompletions = (config) => {
         if (!ctx.request || ctx.request.method !== "POST") {
             throw new GatewayError("Method Not Allowed", 405);
         }
-        const requestId = resolveRequestId(ctx.request);
         // Parse + validate input.
         try {
             ctx.body = await ctx.request.json();
@@ -68,7 +66,7 @@ export const chatCompletions = (config) => {
         // Convert inputs to AI SDK call options.
         const textOptions = convertToTextCallOptions(inputs);
         logger.trace({
-            requestId,
+            requestId: ctx.requestId,
             options: textOptions,
         }, "[chat] AI SDK options");
         addSpanEvent("hebo.options.prepared");
@@ -127,7 +125,7 @@ export const chatCompletions = (config) => {
             },
             ...textOptions,
         });
-        logger.trace({ requestId, result }, "[chat] AI SDK result");
+        logger.trace({ requestId: ctx.requestId, result }, "[chat] AI SDK result");
         addSpanEvent("hebo.ai-sdk.completed");
         // Transform result.
         ctx.result = toChatCompletions(result, ctx.resolvedModelId);

package/dist/endpoints/embeddings/handler.js

@@ -7,7 +7,6 @@ import { modelMiddlewareMatcher } from "../../middleware/matcher";
 import { resolveProvider } from "../../providers/registry";
 import { recordRequestDuration, recordTimePerOutputToken, recordTokenUsage, } from "../../telemetry/gen-ai";
 import { addSpanEvent, setSpanAttributes } from "../../telemetry/span";
-import { resolveRequestId } from "../../utils/headers";
 import { prepareForwardHeaders } from "../../utils/request";
 import { convertToEmbedCallOptions, toEmbeddings } from "./converters";
 import { getEmbeddingsGeneralAttributes, getEmbeddingsRequestAttributes, getEmbeddingsResponseAttributes, } from "./otel";
@@ -22,7 +21,6 @@ export const embeddings = (config) => {
         if (!ctx.request || ctx.request.method !== "POST") {
             throw new GatewayError("Method Not Allowed", 405);
         }
-        const requestId = resolveRequestId(ctx.request);
         // Parse + validate input.
         try {
             ctx.body = await ctx.request.json();
@@ -67,7 +65,7 @@ export const embeddings = (config) => {
         setSpanAttributes(genAiGeneralAttrs);
         // Convert inputs to AI SDK call options.
         const embedOptions = convertToEmbedCallOptions(inputs);
-        logger.trace({ requestId, options: embedOptions }, "[embeddings] AI SDK options");
+        logger.trace({ requestId: ctx.requestId, options: embedOptions }, "[embeddings] AI SDK options");
         addSpanEvent("hebo.options.prepared");
         setSpanAttributes(getEmbeddingsRequestAttributes(inputs, genAiSignalLevel));
         // Build middleware chain (model -> forward params -> provider).
@@ -83,7 +81,7 @@ export const embeddings = (config) => {
             abortSignal: ctx.request.signal,
             ...embedOptions,
         });
-        logger.trace({ requestId, result }, "[embeddings] AI SDK result");
+        logger.trace({ requestId: ctx.requestId, result }, "[embeddings] AI SDK result");
         addSpanEvent("hebo.ai-sdk.completed");
         // Transform result.
         ctx.result = toEmbeddings(result, ctx.modelId);

package/dist/lifecycle.js

@@ -8,8 +8,7 @@ import { getRequestAttributes, getResponseAttributes } from "./telemetry/http";
 import { recordV8jsMemory } from "./telemetry/memory";
 import { addSpanEvent, setSpanEventsEnabled, setSpanTracer, startSpan } from "./telemetry/span";
 import { wrapStream } from "./telemetry/stream";
-import { resolveRequestId } from "./utils/headers";
-import { maybeApplyRequestPatch, prepareRequestHeaders } from "./utils/request";
+import { resolveOrCreateRequestId } from "./utils/request";
 import { prepareResponseInit, toResponse } from "./utils/response";
 export const winterCgHandler = (run, config) => {
     const parsedConfig = parseConfig(config);
@@ -24,14 +23,13 @@ export const winterCgHandler = (run, config) => {
             state: state ?? {},
             providers: parsedConfig.providers,
             models: parsedConfig.models,
+            requestId: resolveOrCreateRequestId(request),
         };
-        const headers = prepareRequestHeaders(ctx.request);
-        if (headers)
-            ctx.request = new Request(ctx.request, { headers });
         const span = startSpan(ctx.request.url);
         span.setAttributes(getBaggageAttributes(ctx.request));
         if (!span.isExisting) {
             span.setAttributes(getRequestAttributes(ctx.request, parsedConfig.telemetry?.signals?.http));
+            span.setAttributes({ "http.request.id": ctx.requestId });
         }
         const finalize = (status, reason) => {
             if (ctx.operation) {
@@ -47,8 +45,8 @@ export const winterCgHandler = (run, config) => {
             else if (status === 200 && ctx.response?.status)
                 realStatus = ctx.response.status;
             if (realStatus !== 200) {
-                (realStatus >= 500 ? logger.error : logger.warn)({
-                    requestId: resolveRequestId(ctx.request),
+                logger[realStatus >= 500 ? "error" : "warn"]({
+                    requestId: ctx.requestId,
                     err: reason ?? ctx.request.signal.reason,
                 });
                 if (realStatus >= 500)
@@ -65,16 +63,13 @@ export const winterCgHandler = (run, config) => {
                 if (onRequest instanceof Response) {
                     ctx.response = onRequest;
                 }
-                else if (onRequest) {
-                    ctx.request = maybeApplyRequestPatch(ctx.request, onRequest);
-                }
             }
             if (!ctx.response) {
                 ctx.result = (await span.runWithContext(() => run(ctx)));
                 if (ctx.result instanceof ReadableStream) {
                     ctx.result = wrapStream(ctx.result, { onDone: finalize });
                 }
-                ctx.response = toResponse(ctx.result, prepareResponseInit(ctx.request));
+                ctx.response = toResponse(ctx.result, prepareResponseInit(ctx.requestId));
             }
             if (parsedConfig.hooks?.onResponse) {
                 const onResponse = await parsedConfig.hooks.onResponse(ctx);
@@ -91,7 +86,7 @@ export const winterCgHandler = (run, config) => {
         catch (error) {
             ctx.response = toOpenAIErrorResponse(ctx.request.signal.aborted
                 ? new GatewayError(error ?? ctx.request.signal.reason, 499)
-                : error, prepareResponseInit(ctx.request));
+                : error, prepareResponseInit(ctx.requestId));
             finalize(ctx.response.status, error);
         }
         return ctx.response ?? new Response("Internal Server Error", { status: 500 });

package/dist/telemetry/http.js

@@ -1,5 +1,4 @@
 import {} from "../types";
-import { resolveRequestId } from "../utils/headers";
 const headerArr = (h, k) => (h.has(k) ? [h.get(k)] : undefined);
 export const getRequestAttributes = (request, signalLevel) => {
     if (!signalLevel || signalLevel === "off")
@@ -26,8 +25,6 @@ export const getRequestAttributes = (request, signalLevel) => {
     };
     if (signalLevel !== "required") {
         Object.assign(attrs, {
-            // FUTURE: does ElysiaJS and other frameworks attach request id?
-            "http.request.id": resolveRequestId(request),
             "user_agent.original": request.headers.get("user-agent") ?? undefined,
         });
     }

package/dist/types.d.ts

@@ -6,19 +6,6 @@ import type { Model, ModelList } from "./endpoints/models";
 import type { Logger, LoggerConfig } from "./logger";
 import type { ModelCatalog, ModelId } from "./models/types";
 import type { ProviderId, ProviderRegistry } from "./providers/types";
-/**
- * Request overrides returned from the `onRequest` hook.
- */
-export type RequestPatch = {
-    /**
-     * Headers to merge into the incoming request.
-     */
-    headers?: HeadersInit;
-    /**
-     * Body to replace on the incoming request.
-     */
-    body?: BodyInit;
-};
 /**
  * Per-request context shared across handlers and hooks.
  */
@@ -39,6 +26,10 @@ export type GatewayContext = {
      * Incoming request for the handler.
      */
     request: Request;
+    /**
+     * Resolved request ID for logging and telemetry.
+     */
+    requestId: string;
     /**
      * Parsed body from the request.
      */
@@ -80,10 +71,10 @@ export type HookContext = Omit<Readonly<GatewayContext>, "state"> & {
 };
 type RequiredHookContext<K extends keyof GatewayContext> = Omit<HookContext, K> & Required<Pick<HookContext, K>>;
 export type OnRequestHookContext = RequiredHookContext<"request">;
-export type BeforeHookContext = RequiredHookContext<"request" | "body" | "operation">;
-export type ResolveModelHookContext = RequiredHookContext<"request" | "body" | "modelId">;
-export type ResolveProviderHookContext = RequiredHookContext<"request" | "body" | "modelId" | "resolvedModelId" | "operation">;
-export type AfterHookContext = RequiredHookContext<"request" | "result" | "provider" | "resolvedModelId" | "operation">;
+export type BeforeHookContext = RequiredHookContext<"request" | "operation" | "body">;
+export type ResolveModelHookContext = RequiredHookContext<"request" | "operation" | "body" | "modelId">;
+export type ResolveProviderHookContext = RequiredHookContext<"request" | "operation" | "body" | "modelId" | "resolvedModelId">;
+export type AfterHookContext = RequiredHookContext<"request" | "operation" | "body" | "modelId" | "resolvedModelId" | "provider" | "resolvedProviderId" | "result">;
 export type OnResponseHookContext = RequiredHookContext<"request" | "response">;
 /**
  * Hooks to plugin to the gateway lifecycle.
@@ -91,10 +82,9 @@ export type OnResponseHookContext = RequiredHookContext<"request" | "response">;
 export type GatewayHooks = {
     /**
      * Runs before any endpoint handler logic.
-     * @returns Optional RequestPatch to merge into headers / override body,
-     * or Response to short-circuit the request.
+     * @returns Optional Response to short-circuit the request.
      */
-    onRequest?: (ctx: OnRequestHookContext) => void | RequestPatch | Response | Promise<void | RequestPatch | Response>;
+    onRequest?: (ctx: OnRequestHookContext) => void | Response | Promise<void | Response>;
     /**
      * Runs after request JSON is parsed and validated for chat completions / embeddings.
      * @returns Replacement parsed body, or undefined to keep original.

package/dist/utils/request.d.ts

@@ -1,4 +1,2 @@
-import type { RequestPatch } from "../types";
-export declare const prepareRequestHeaders: (request: Request) => Headers | undefined;
+export declare const resolveOrCreateRequestId: (request: Request) => string;
 export declare const prepareForwardHeaders: (request: Request) => Record<string, string>;
-export declare const maybeApplyRequestPatch: (request: Request, patch: RequestPatch) => Request;

package/dist/utils/request.js

@@ -1,15 +1,8 @@
 import pkg from "../../package.json" with { type: "json" };
-import { REQUEST_ID_HEADER } from "./headers";
+import { resolveRequestId } from "./headers";
 const GATEWAY_VERSION = pkg.version;
-export const prepareRequestHeaders = (request) => {
-    const existingRequestId = request.headers.get(REQUEST_ID_HEADER);
-    if (existingRequestId)
-        return;
-    const requestId = "req_" + crypto.getRandomValues(new Uint32Array(2)).reduce((s, n) => s + n.toString(36), "");
-    const headers = new Headers(request.headers);
-    headers.set(REQUEST_ID_HEADER, requestId);
-    return headers;
-};
+const createRequestId = () => "req_" + crypto.getRandomValues(new Uint32Array(2)).reduce((s, n) => s + n.toString(36), "");
+export const resolveOrCreateRequestId = (request) => resolveRequestId(request) ?? createRequestId();
 export const prepareForwardHeaders = (request) => {
     const userAgent = request.headers.get("user-agent");
     const appendedUserAgent = userAgent
@@ -19,19 +12,3 @@ export const prepareForwardHeaders = (request) => {
         "user-agent": appendedUserAgent,
     };
 };
-export const maybeApplyRequestPatch = (request, patch) => {
-    if (!patch.headers && patch.body === undefined)
-        return request;
-    if (!patch.headers) {
-        // eslint-disable-next-line no-invalid-fetch-options
-        return new Request(request, { body: patch.body });
-    }
-    const headers = new Headers(request.headers);
-    for (const [key, value] of new Headers(patch.headers)) {
-        headers.set(key, value);
-    }
-    const init = { headers };
-    if (patch.body !== undefined)
-        init.body = patch.body;
-    return new Request(request, init);
-};

package/dist/utils/response.d.ts

@@ -1,3 +1,3 @@
-export declare const prepareResponseInit: (request: Request) => ResponseInit;
+export declare const prepareResponseInit: (requestId: string) => ResponseInit;
 export declare const mergeResponseInit: (defaultHeaders: HeadersInit, responseInit?: ResponseInit) => ResponseInit;
 export declare const toResponse: (result: ReadableStream | Uint8Array<ArrayBuffer> | object | string, responseInit?: ResponseInit) => Response;

package/dist/utils/response.js

@@ -1,4 +1,4 @@
-import { REQUEST_ID_HEADER, resolveRequestId } from "./headers";
+import { REQUEST_ID_HEADER } from "./headers";
 const TEXT_ENCODER = new TextEncoder();
 class JsonToSseTransformStream extends TransformStream {
     constructor() {
@@ -12,8 +12,8 @@ class JsonToSseTransformStream extends TransformStream {
         });
     }
 }
-export const prepareResponseInit = (request) => ({
-    headers: { [REQUEST_ID_HEADER]: resolveRequestId(request) },
+export const prepareResponseInit = (requestId) => ({
+    headers: { [REQUEST_ID_HEADER]: requestId },
 });
 export const mergeResponseInit = (defaultHeaders, responseInit) => {
     const headers = new Headers(defaultHeaders);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hebo-ai/gateway",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "description": "AI gateway as a framework. For full control over models, routing & lifecycle. OpenAI-compatible /chat/completions, /embeddings & /models.",
   "keywords": [
     "ai",

package/src/endpoints/chat-completions/handler.ts

@@ -29,7 +29,6 @@ import {
   recordTokenUsage,
 } from "../../telemetry/gen-ai";
 import { addSpanEvent, setSpanAttributes } from "../../telemetry/span";
-import { resolveRequestId } from "../../utils/headers";
 import { prepareForwardHeaders } from "../../utils/request";
 import { convertToTextCallOptions, toChatCompletions, toChatCompletionsStream } from "./converters";
 import {
@@ -52,8 +51,6 @@ export const chatCompletions = (config: GatewayConfig): Endpoint => {
       throw new GatewayError("Method Not Allowed", 405);
     }
 
-    const requestId = resolveRequestId(ctx.request);
-
     // Parse + validate input.
     try {
       ctx.body = await ctx.request.json();
@@ -107,7 +104,7 @@ export const chatCompletions = (config: GatewayConfig): Endpoint => {
     const textOptions = convertToTextCallOptions(inputs);
     logger.trace(
       {
-        requestId,
+        requestId: ctx.requestId,
         options: textOptions,
       },
       "[chat] AI SDK options",
@@ -178,7 +175,7 @@ export const chatCompletions = (config: GatewayConfig): Endpoint => {
       },
       ...textOptions,
     });
-    logger.trace({ requestId, result }, "[chat] AI SDK result");
+    logger.trace({ requestId: ctx.requestId, result }, "[chat] AI SDK result");
     addSpanEvent("hebo.ai-sdk.completed");
 
     // Transform result.

package/src/endpoints/embeddings/handler.ts

@@ -22,7 +22,6 @@ import {
   recordTokenUsage,
 } from "../../telemetry/gen-ai";
 import { addSpanEvent, setSpanAttributes } from "../../telemetry/span";
-import { resolveRequestId } from "../../utils/headers";
 import { prepareForwardHeaders } from "../../utils/request";
 import { convertToEmbedCallOptions, toEmbeddings } from "./converters";
 import {
@@ -45,8 +44,6 @@ export const embeddings = (config: GatewayConfig): Endpoint => {
       throw new GatewayError("Method Not Allowed", 405);
     }
 
-    const requestId = resolveRequestId(ctx.request);
-
     // Parse + validate input.
     try {
       ctx.body = await ctx.request.json();
@@ -98,7 +95,10 @@ export const embeddings = (config: GatewayConfig): Endpoint => {
 
     // Convert inputs to AI SDK call options.
     const embedOptions = convertToEmbedCallOptions(inputs);
-    logger.trace({ requestId, options: embedOptions }, "[embeddings] AI SDK options");
+    logger.trace(
+      { requestId: ctx.requestId, options: embedOptions },
+      "[embeddings] AI SDK options",
+    );
     addSpanEvent("hebo.options.prepared");
     setSpanAttributes(getEmbeddingsRequestAttributes(inputs, genAiSignalLevel));
 
@@ -116,7 +116,7 @@ export const embeddings = (config: GatewayConfig): Endpoint => {
       abortSignal: ctx.request.signal,
       ...embedOptions,
     });
-    logger.trace({ requestId, result }, "[embeddings] AI SDK result");
+    logger.trace({ requestId: ctx.requestId, result }, "[embeddings] AI SDK result");
     addSpanEvent("hebo.ai-sdk.completed");
 
     // Transform result.

package/src/lifecycle.ts

@@ -15,8 +15,7 @@ import { getRequestAttributes, getResponseAttributes } from "./telemetry/http";
 import { recordV8jsMemory } from "./telemetry/memory";
 import { addSpanEvent, setSpanEventsEnabled, setSpanTracer, startSpan } from "./telemetry/span";
 import { wrapStream } from "./telemetry/stream";
-import { resolveRequestId } from "./utils/headers";
-import { maybeApplyRequestPatch, prepareRequestHeaders } from "./utils/request";
+import { resolveOrCreateRequestId } from "./utils/request";
 import { prepareResponseInit, toResponse } from "./utils/response";
 
 export const winterCgHandler = (
@@ -37,15 +36,14 @@ export const winterCgHandler = (
       state: state ?? {},
       providers: parsedConfig.providers,
       models: parsedConfig.models,
+      requestId: resolveOrCreateRequestId(request),
     };
 
-    const headers = prepareRequestHeaders(ctx.request);
-    if (headers) ctx.request = new Request(ctx.request, { headers });
-
     const span = startSpan(ctx.request.url);
     span.setAttributes(getBaggageAttributes(ctx.request));
     if (!span.isExisting) {
       span.setAttributes(getRequestAttributes(ctx.request, parsedConfig.telemetry?.signals?.http));
+      span.setAttributes({ "http.request.id": ctx.requestId });
     }
 
     const finalize = (status: number, reason?: unknown) => {
@@ -65,8 +63,8 @@ export const winterCgHandler = (
       else if (status === 200 && ctx.response?.status) realStatus = ctx.response.status;
 
       if (realStatus !== 200) {
-        (realStatus >= 500 ? logger.error : logger.warn)({
-          requestId: resolveRequestId(ctx.request),
+        logger[realStatus >= 500 ? "error" : "warn"]({
+          requestId: ctx.requestId,
           err: reason ?? ctx.request.signal.reason,
         });
 
@@ -86,8 +84,6 @@ export const winterCgHandler = (
 
         if (onRequest instanceof Response) {
           ctx.response = onRequest;
-        } else if (onRequest) {
-          ctx.request = maybeApplyRequestPatch(ctx.request, onRequest);
         }
       }
 
@@ -98,7 +94,7 @@ export const winterCgHandler = (
           ctx.result = wrapStream(ctx.result, { onDone: finalize });
         }
 
-        ctx.response = toResponse(ctx.result!, prepareResponseInit(ctx.request));
+        ctx.response = toResponse(ctx.result!, prepareResponseInit(ctx.requestId));
       }
 
       if (parsedConfig.hooks?.onResponse) {
@@ -118,7 +114,7 @@ export const winterCgHandler = (
         ctx.request.signal.aborted
           ? new GatewayError(error ?? ctx.request.signal.reason, 499)
           : error,
-        prepareResponseInit(ctx.request),
+        prepareResponseInit(ctx.requestId),
       );
       finalize(ctx.response.status, error);
     }

package/src/telemetry/http.ts

@@ -1,5 +1,4 @@
 import { type TelemetrySignalLevel } from "../types";
-import { resolveRequestId } from "../utils/headers";
 
 const headerArr = (h: Headers, k: string) => (h.has(k) ? [h.get(k)!] : undefined);
 
@@ -29,8 +28,6 @@ export const getRequestAttributes = (request: Request, signalLevel?: TelemetrySi
 
   if (signalLevel !== "required") {
     Object.assign(attrs, {
-      // FUTURE: does ElysiaJS and other frameworks attach request id?
-      "http.request.id": resolveRequestId(request),
       "user_agent.original": request.headers.get("user-agent") ?? undefined,
     });
   }

package/src/types.ts

@@ -12,20 +12,6 @@ import type { Logger, LoggerConfig } from "./logger";
 import type { ModelCatalog, ModelId } from "./models/types";
 import type { ProviderId, ProviderRegistry } from "./providers/types";
 
-/**
- * Request overrides returned from the `onRequest` hook.
- */
-export type RequestPatch = {
-  /**
-   * Headers to merge into the incoming request.
-   */
-  headers?: HeadersInit;
-  /**
-   * Body to replace on the incoming request.
-   */
-  body?: BodyInit;
-};
-
 /**
  * Per-request context shared across handlers and hooks.
  */
@@ -46,6 +32,10 @@ export type GatewayContext = {
    * Incoming request for the handler.
    */
   request: Request;
+  /**
+   * Resolved request ID for logging and telemetry.
+   */
+  requestId: string;
   /**
    * Parsed body from the request.
    */
@@ -95,13 +85,22 @@ export type HookContext = Omit<Readonly<GatewayContext>, "state"> & {
 type RequiredHookContext<K extends keyof GatewayContext> = Omit<HookContext, K> &
   Required<Pick<HookContext, K>>;
 export type OnRequestHookContext = RequiredHookContext<"request">;
-export type BeforeHookContext = RequiredHookContext<"request" | "body" | "operation">;
-export type ResolveModelHookContext = RequiredHookContext<"request" | "body" | "modelId">;
+export type BeforeHookContext = RequiredHookContext<"request" | "operation" | "body">;
+export type ResolveModelHookContext = RequiredHookContext<
+  "request" | "operation" | "body" | "modelId"
+>;
 export type ResolveProviderHookContext = RequiredHookContext<
-  "request" | "body" | "modelId" | "resolvedModelId" | "operation"
+  "request" | "operation" | "body" | "modelId" | "resolvedModelId"
 >;
 export type AfterHookContext = RequiredHookContext<
-  "request" | "result" | "provider" | "resolvedModelId" | "operation"
+  | "request"
+  | "operation"
+  | "body"
+  | "modelId"
+  | "resolvedModelId"
+  | "provider"
+  | "resolvedProviderId"
+  | "result"
 >;
 export type OnResponseHookContext = RequiredHookContext<"request" | "response">;
 
@@ -111,12 +110,9 @@ export type OnResponseHookContext = RequiredHookContext<"request" | "response">;
 export type GatewayHooks = {
   /**
    * Runs before any endpoint handler logic.
-   * @returns Optional RequestPatch to merge into headers / override body,
-   * or Response to short-circuit the request.
+   * @returns Optional Response to short-circuit the request.
    */
-  onRequest?: (
-    ctx: OnRequestHookContext,
-  ) => void | RequestPatch | Response | Promise<void | RequestPatch | Response>;
+  onRequest?: (ctx: OnRequestHookContext) => void | Response | Promise<void | Response>;
   /**
    * Runs after request JSON is parsed and validated for chat completions / embeddings.
    * @returns Replacement parsed body, or undefined to keep original.

package/src/utils/request.ts

@@ -1,22 +1,13 @@
-import type { RequestPatch } from "../types";
-
 import pkg from "../../package.json" with { type: "json" };
-import { REQUEST_ID_HEADER } from "./headers";
+import { resolveRequestId } from "./headers";
 
 const GATEWAY_VERSION = pkg.version;
 
-export const prepareRequestHeaders = (request: Request) => {
-  const existingRequestId = request.headers.get(REQUEST_ID_HEADER);
-  if (existingRequestId) return;
-
-  const requestId =
-    "req_" + crypto.getRandomValues(new Uint32Array(2)).reduce((s, n) => s + n.toString(36), "");
+const createRequestId = () =>
+  "req_" + crypto.getRandomValues(new Uint32Array(2)).reduce((s, n) => s + n.toString(36), "");
 
-  const headers = new Headers(request.headers);
-  headers.set(REQUEST_ID_HEADER, requestId);
-
-  return headers;
-};
+export const resolveOrCreateRequestId = (request: Request) =>
+  resolveRequestId(request) ?? createRequestId();
 
 export const prepareForwardHeaders = (request: Request): Record<string, string> => {
   const userAgent = request.headers.get("user-agent");
@@ -28,22 +19,3 @@ export const prepareForwardHeaders = (request: Request): Record<string, string>
     "user-agent": appendedUserAgent,
   };
 };
-
-export const maybeApplyRequestPatch = (request: Request, patch: RequestPatch) => {
-  if (!patch.headers && patch.body === undefined) return request;
-
-  if (!patch.headers) {
-    // eslint-disable-next-line no-invalid-fetch-options
-    return new Request(request, { body: patch.body });
-  }
-
-  const headers = new Headers(request.headers);
-  for (const [key, value] of new Headers(patch.headers)) {
-    headers.set(key, value);
-  }
-
-  const init: RequestInit = { headers };
-  if (patch.body !== undefined) init.body = patch.body;
-
-  return new Request(request, init);
-};

package/src/utils/response.ts

@@ -1,4 +1,4 @@
-import { REQUEST_ID_HEADER, resolveRequestId } from "./headers";
+import { REQUEST_ID_HEADER } from "./headers";
 
 const TEXT_ENCODER = new TextEncoder();
 
@@ -15,8 +15,8 @@ class JsonToSseTransformStream extends TransformStream<unknown, string> {
   }
 }
 
-export const prepareResponseInit = (request: Request): ResponseInit => ({
-  headers: { [REQUEST_ID_HEADER]: resolveRequestId(request)! },
+export const prepareResponseInit = (requestId: string): ResponseInit => ({
+  headers: { [REQUEST_ID_HEADER]: requestId },
 });
 
 export const mergeResponseInit = (