@hebo-ai/gateway 0.8.2 → 0.9.1

package/dist/index.d.ts

@@ -5,10 +5,6 @@ export * from "./errors/openai";
 export * from "./logger";
 export * from "./middleware/common";
 export * from "./middleware/matcher";
-export * from "./endpoints/chat-completions";
-export * from "./endpoints/conversations";
-export * from "./endpoints/embeddings";
-export * from "./endpoints/models";
 export * from "./models/catalog";
 export * from "./models/types";
 export * from "./providers/registry";

package/dist/index.js

@@ -4,10 +4,6 @@ export * from "./errors/openai";
 export * from "./logger";
 export * from "./middleware/common";
 export * from "./middleware/matcher";
-export * from "./endpoints/chat-completions";
-export * from "./endpoints/conversations";
-export * from "./endpoints/embeddings";
-export * from "./endpoints/models";
 export * from "./models/catalog";
 export * from "./models/types";
 export * from "./providers/registry";

package/dist/lifecycle.js

@@ -54,43 +54,60 @@ export const winterCgHandler = (run, config) => {
                 span.recordError(reason);
             }
             span.setAttributes({ "http.response.status_code_effective": realStatus });
-            if (ctx.operation === "chat" || ctx.operation === "embeddings") {
+            if (ctx.operation === "chat" ||
+                ctx.operation === "embeddings" ||
+                ctx.operation === "responses") {
                 recordRequestDuration(performance.now() - start, realStatus, ctx, parsedConfig.telemetry?.signals?.gen_ai);
             }
             span.finish();
         };
-        try {
-            if (parsedConfig.hooks?.onRequest) {
-                const onRequest = await parsedConfig.hooks.onRequest(ctx);
-                addSpanEvent("hebo.hooks.on_request.completed");
-                if (onRequest instanceof Response) {
-                    ctx.response = onRequest;
+        await span.runWithContext(async () => {
+            try {
+                if (parsedConfig.hooks?.onRequest) {
+                    const onRequest = await parsedConfig.hooks.onRequest(ctx);
+                    addSpanEvent("hebo.hooks.on_request.completed");
+                    if (onRequest instanceof Response) {
+                        ctx.response = onRequest;
+                    }
                 }
-            }
-            if (!ctx.response) {
-                ctx.result = (await span.runWithContext(() => run(ctx, parsedConfig)));
-                ctx.response = toResponse(ctx.result, prepareResponseInit(ctx.requestId), {
-                    onDone: finalize,
-                });
-            }
-            if (parsedConfig.hooks?.onResponse) {
-                const onResponse = await parsedConfig.hooks.onResponse(ctx);
-                addSpanEvent("hebo.hooks.on_response.completed");
-                if (onResponse) {
-                    ctx.response = onResponse;
+                if (!ctx.response) {
+                    ctx.result = (await run(ctx, parsedConfig));
+                    ctx.response = toResponse(ctx.result, prepareResponseInit(ctx.requestId), {
+                        onDone: finalize,
+                    });
+                }
+                if (parsedConfig.hooks?.onResponse) {
+                    const onResponse = await parsedConfig.hooks.onResponse(ctx);
+                    addSpanEvent("hebo.hooks.on_response.completed");
+                    if (onResponse) {
+                        ctx.response = onResponse;
+                    }
+                }
+                // FUTURE: this can leak if onResponse removed wrapper from response.body
+                if (!(ctx.result instanceof ReadableStream)) {
+                    finalize(ctx.response.status);
                 }
             }
-            // FUTURE: this can leak if onResponse removed wrapper from response.body
-            if (!(ctx.result instanceof ReadableStream)) {
-                finalize(ctx.response.status);
+            catch (error) {
+                if (parsedConfig.hooks?.onError) {
+                    try {
+                        ctx.error = error;
+                        const onError = await parsedConfig.hooks.onError(ctx);
+                        addSpanEvent("hebo.hooks.on_error.completed");
+                        if (onError) {
+                            ctx.response = onError;
+                        }
+                    }
+                    catch {
+                        logger.debug("[lifecycle] onError hook threw");
+                    }
+                }
+                ctx.response ??= toOpenAIErrorResponse(ctx.request.signal.aborted
+                    ? new GatewayError(error ?? ctx.request.signal.reason, 499)
+                    : error, prepareResponseInit(ctx.requestId));
+                finalize(ctx.response.status, error);
             }
-        }
-        catch (error) {
-            ctx.response = toOpenAIErrorResponse(ctx.request.signal.aborted
-                ? new GatewayError(error ?? ctx.request.signal.reason, 499)
-                : error, prepareResponseInit(ctx.requestId));
-            finalize(ctx.response.status, error);
-        }
+        });
         return ctx.response ?? new Response("Internal Server Error", { status: 500 });
     };
 };

package/dist/models/anthropic/middleware.d.ts

@@ -1,5 +1,5 @@
 import type { LanguageModelMiddleware } from "ai";
 import type { ChatCompletionsReasoningEffort } from "../../endpoints/chat-completions/schema";
-export declare function mapClaudeReasoningEffort(effort: ChatCompletionsReasoningEffort, modelId: string): "low" | "high" | "medium" | "max";
+export declare function mapClaudeReasoningEffort(effort: ChatCompletionsReasoningEffort, modelId: string): "low" | "medium" | "high" | "max";
 export declare const claudeReasoningMiddleware: LanguageModelMiddleware;
 export declare const claudePromptCachingMiddleware: LanguageModelMiddleware;

package/dist/models/anthropic/presets.js

@@ -6,7 +6,12 @@ const CLAUDE_BASE = {
     },
     capabilities: ["attachments", "tool_call", "structured_output", "temperature"],
     context: 200000,
-    providers: ["anthropic", "bedrock", "vertex", "azure"],
+    providers: [
+        "anthropic",
+        "bedrock",
+        "vertex",
+        "azure",
+    ],
 };
 const CLAUDE_PDF_MODALITIES = {
     modalities: {

package/dist/models/google/middleware.d.ts

@@ -1,7 +1,7 @@
 import type { EmbeddingModelMiddleware, LanguageModelMiddleware } from "ai";
 import type { ChatCompletionsReasoningEffort } from "../../endpoints/chat-completions/schema";
 export declare const geminiDimensionsMiddleware: EmbeddingModelMiddleware;
-export declare function mapGeminiReasoningEffort(effort: ChatCompletionsReasoningEffort, modelId: string): "low" | "high" | "minimal" | "medium";
+export declare function mapGeminiReasoningEffort(effort: ChatCompletionsReasoningEffort, modelId: string): "minimal" | "low" | "medium" | "high";
 export declare const GEMINI_DEFAULT_MAX_OUTPUT_TOKENS = 65536;
 export declare const GEMINI_2_5_PRO_MIN_THINKING_BUDGET = 128;
 export declare const geminiReasoningMiddleware: LanguageModelMiddleware;

package/dist/models/google/middleware.js

@@ -71,9 +71,15 @@ export const geminiReasoningMiddleware = {
             };
         }
         else if (modelId.includes("gemini-3") && reasoning.effort) {
-            target.thinkingConfig = {
-                thinkingLevel: mapGeminiReasoningEffort(reasoning.effort, modelId),
-            };
+            if (reasoning.effort === "none") {
+                // thinkingBudget: 0 fully disables thinking (thinkingLevel: "minimal" still allows some)
+                target.thinkingConfig = { thinkingBudget: 0 };
+            }
+            else {
+                target.thinkingConfig = {
+                    thinkingLevel: mapGeminiReasoningEffort(reasoning.effort, modelId),
+                };
+            }
             // FUTURE: warn if model is gemini-3 and max_tokens (unsupported) was ignored
         }
         const thinkingConfig = (target.thinkingConfig ??= {});

package/dist/models/meta/presets.js

@@ -6,7 +6,12 @@ const LLAMA_3_BASE = {
     },
     capabilities: ["attachments", "tool_call", "temperature"],
     context: 128000,
-    providers: ["groq", "bedrock", "vertex", "azure"],
+    providers: [
+        "groq",
+        "bedrock",
+        "vertex",
+        "azure",
+    ],
 };
 export const llama31_8b = presetFor()("meta/llama-3.1-8b", {
     ...LLAMA_3_BASE,
@@ -63,7 +68,12 @@ const LLAMA_4_BASE = {
     },
     capabilities: ["attachments", "tool_call", "temperature"],
     context: 1000000,
-    providers: ["groq", "bedrock", "vertex", "azure"],
+    providers: [
+        "groq",
+        "bedrock",
+        "vertex",
+        "azure",
+    ],
 };
 export const llama4Scout = presetFor()("meta/llama-4-scout", {
     ...LLAMA_4_BASE,

package/dist/providers/registry.d.ts

@@ -5,7 +5,7 @@ export declare const resolveProvider: (args: {
     providers: ProviderRegistry;
     models: ModelCatalog;
     modelId: ModelId;
-    operation: "chat" | "embeddings";
+    operation: "chat" | "embeddings" | "responses";
 }) => ProviderV3;
 export type CanonicalIdsOptions = {
     mapping?: Partial<Record<ModelId, string>>;

package/dist/types.d.ts

@@ -3,10 +3,12 @@ import type { Tracer } from "@opentelemetry/api";
 import type { ChatCompletions, ChatCompletionsBody, ChatCompletionsStream } from "./endpoints/chat-completions/schema";
 import type { ConversationStorage } from "./endpoints/conversations/storage/types";
 import type { Embeddings, EmbeddingsBody } from "./endpoints/embeddings/schema";
+import type { Responses, ResponsesBody, ResponsesStream } from "./endpoints/responses/schema";
 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";
+export type GatewayOperation = "chat" | "embeddings" | "responses" | "models" | "conversations";
 /**
  * Per-request context shared across handlers and hooks.
  */
@@ -34,7 +36,7 @@ export type GatewayContext = {
     /**
      * Parsed body from the request.
      */
-    body?: ChatCompletionsBody | EmbeddingsBody;
+    body?: ChatCompletionsBody | EmbeddingsBody | ResponsesBody;
     /**
      * Incoming model ID.
      */
@@ -46,7 +48,7 @@ export type GatewayContext = {
     /**
      * Operation type.
      */
-    operation?: "chat" | "embeddings" | "models" | "conversations";
+    operation?: GatewayOperation;
     /**
      * Resolved provider instance.
      */
@@ -58,11 +60,15 @@ export type GatewayContext = {
     /**
      * Result returned by the handler (pre-response).
      */
-    result?: ChatCompletions | ChatCompletionsStream | Embeddings | Model | ModelList;
+    result?: ChatCompletions | ChatCompletionsStream | Embeddings | Model | ModelList | Responses | ResponsesStream;
     /**
      * Response object returned by the handler.
      */
     response?: Response;
+    /**
+     * Error thrown during execution.
+     */
+    error?: unknown;
 };
 /**
  * Hook context: all fields readonly except `state`.
@@ -77,6 +83,7 @@ export type ResolveModelHookContext = RequiredHookContext<"request" | "operation
 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">;
+export type OnErrorHookContext = RequiredHookContext<"error">;
 /**
  * Hooks to plugin to the gateway lifecycle.
  */
@@ -87,10 +94,10 @@ export type GatewayHooks = {
      */
     onRequest?: (ctx: OnRequestHookContext) => void | Response | Promise<void | Response>;
     /**
-     * Runs after request JSON is parsed and validated for chat completions / embeddings.
+     * Runs after request JSON is parsed and validated for chat completions / embeddings / responses.
      * @returns Replacement parsed body, or undefined to keep original.
      */
-    before?: (ctx: BeforeHookContext) => void | ChatCompletionsBody | EmbeddingsBody | Promise<void | ChatCompletionsBody | EmbeddingsBody>;
+    before?: (ctx: BeforeHookContext) => void | ChatCompletionsBody | EmbeddingsBody | ResponsesBody | Promise<void | ChatCompletionsBody | EmbeddingsBody | ResponsesBody>;
     /**
      * Maps a user-provided model ID or alias to a canonical ID.
      * @returns Canonical model ID or undefined to keep original.
@@ -105,12 +112,17 @@ export type GatewayHooks = {
      * Runs after the endpoint handler.
      * @returns Result to replace, or undefined to keep original.
      */
-    after?: (ctx: AfterHookContext) => void | ChatCompletions | ChatCompletionsStream | Embeddings | Promise<void | ChatCompletions | ChatCompletionsStream | Embeddings>;
+    after?: (ctx: AfterHookContext) => void | ChatCompletions | ChatCompletionsStream | Embeddings | Model | ModelList | Responses | ResponsesStream | Promise<void | ChatCompletions | ChatCompletionsStream | Embeddings | Model | ModelList | Responses | ResponsesStream>;
     /**
      * Runs after the lifecycle has produced the final Response.
      * @returns Replacement Response, or undefined to keep original.
      */
     onResponse?: (ctx: OnResponseHookContext) => void | Response | Promise<void | Response>;
+    /**
+     * Runs when the lifecycle catches an error.
+     * @returns Optional Response to replace the default error response.
+     */
+    onError?: (ctx: OnErrorHookContext) => void | Response | Promise<void | Response>;
 };
 export type TelemetrySignalLevel = "off" | "required" | "recommended" | "full";
 export declare const DEFAULT_CHAT_TIMEOUT_MS: number;

package/dist/utils/env.js

@@ -2,6 +2,6 @@ const NODE_ENV = typeof process === "undefined"
     ? // oxlint-disable-next-line no-unsafe-member-access
         (globalThis.NODE_ENV ?? globalThis.ENV?.NODE_ENV)
     : // oxlint-disable-next-line no-unsafe-assignment
-        process.env?.NODE_ENV;
+        process.env?.["NODE_ENV"];
 export const isProduction = () => NODE_ENV === "production";
 export const isTest = () => NODE_ENV === "test";

package/dist/utils/preset.js

@@ -1,7 +1,6 @@
 function isPlainObject(v) {
     if (!v || typeof v !== "object" || Array.isArray(v))
         return false;
-    // oxlint-disable-next-line no-unsafe-assignment
     const proto = Object.getPrototypeOf(v);
     return proto === Object.prototype || proto === null;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hebo-ai/gateway",
-  "version": "0.8.2",
+  "version": "0.9.1",
   "description": "AI gateway as a framework. For full control over models, routing & lifecycle. OpenAI-compatible /chat/completions, /embeddings & /models.",
   "keywords": [
     "ai",
@@ -66,6 +66,10 @@
       "types": "./dist/endpoints/models/index.d.ts",
       "import": "./dist/endpoints/models/index.js"
     },
+    "./endpoints/responses": {
+      "types": "./dist/endpoints/responses/index.d.ts",
+      "import": "./dist/endpoints/responses/index.js"
+    },
     "./storage/memory": {
       "types": "./dist/endpoints/conversations/storage/memory.d.ts",
       "import": "./dist/endpoints/conversations/storage/memory.js",
@@ -156,7 +160,7 @@
     "typecheck": "oxlint --type-check",
     "test": "bun test",
     "check": "bun lint && bun typecheck",
-    "fix": "bun lint:staged && bun fmt:staged"
+    "fix": "bun lint:staged && bun format:staged"
   },
   "dependencies": {
     "@ai-sdk/provider": "^3.0.8",
@@ -199,7 +203,7 @@
     "pg": "^8.19.0",
     "pino": "^10.3.1",
     "postgres": "^3.4.8",
-    "typescript": "^5.9.3",
+    "typescript": "^6.0.2",
     "vite": "^7.3.1",
     "vite-tsconfig-paths": "^6.1.1",
     "voyage-ai-provider": "^3.0.0"
@@ -218,7 +222,7 @@
     "mysql2": "^3.11.0",
     "pg": "^8.13.0",
     "postgres": "^3.4.0",
-    "typescript": "^5.9.3",
+    "typescript": ">=5.9.3",
     "voyage-ai-provider": "^3.0.0"
   },
   "peerDependenciesMeta": {