fix(translator): honor Chat targets for Responses clients (#9161)

Honor explicit Chat targets for Responses-shaped clients while preserving native Responses providers and selecting token fields from the outbound protocol.

Includes focused regression coverage and the required changelog fragment.

changelog.d/fixes/9161-codex-responses-chat-targets.md

@@ -0,0 +1 @@
+- **fix(translator):** Honor configured Chat targets for Responses-shaped clients while preserving native Responses providers and outbound token fields ([#9161](https://github.com/diegosouzapw/OmniRoute/pull/9161)) — thanks @Zartharas

open-sse/handlers/chatCore/sanitization.ts

@@ -6,8 +6,8 @@ export function sanitizeChatRequestBody(
   sourceFormat: string,
   targetFormat: string
 ): Record<string, unknown> {
-  const prefersResponsesTokenField =
-    sourceFormat === FORMATS.OPENAI_RESPONSES || targetFormat === FORMATS.OPENAI_RESPONSES;
+  void sourceFormat;
+  const prefersResponsesTokenField = targetFormat === FORMATS.OPENAI_RESPONSES;
 
   if (prefersResponsesTokenField) {
     if (body.max_output_tokens === undefined) {

open-sse/handlers/chatCore/targetFormat.ts

@@ -2,13 +2,12 @@
  * chatCore wire target-format resolver (Quality Gate v2 / Fase 9 — chatCore god-file
  * decomposition, #3501).
  *
- * Pure resolution of the provider alias + the upstream target format used to translate the request:
- * apiFormat==="responses" forces OpenAI Responses; otherwise the model's registry target format, then
- * the per-model custom override (#2905), then AgentRouter's matching inbound protocol when the
- * connection has no explicit override, then the provider default. Returns both `alias` (reused by
- * the handler when stripping the `alias/` prefix off the upstream model id) and `targetFormat`.
- * Side-effect-free; sits alongside the other request-setup resolvers
- * (resolveChatCoreRequestSetup / resolveChatCoreRequestFormat).
+ * Pure resolution of the provider alias + the upstream target format used to translate the request.
+ * Model/custom overrides win first. A Responses-shaped inbound request normally keeps the Responses
+ * wire format, except for custom OpenAI-compatible connections explicitly configured for Chat.
+ * AgentRouter may inherit the inbound protocol when no explicit connection override exists.
+ * Returns both `alias` (reused by the handler when stripping the `alias/` prefix off the upstream
+ * model id) and `targetFormat`.
  */
 
 import { PROVIDER_ID_TO_ALIAS, getModelTargetFormat } from "../../config/providerModels.ts";
@@ -46,15 +45,19 @@ export function resolveChatCoreTargetFormat(opts: {
       sourceFormat === FORMATS.CLAUDE)
       ? sourceFormat
       : undefined;
+  const providerTargetFormat = getTargetFormat(provider, providerSpecificData);
+  const customOpenAICompatible = provider.startsWith("openai-compatible-");
   // #8994: model-level targetFormat overrides (from registry or custom-model DB override)
   // take precedence over apiFormat="responses" — otherwise Vertex Claude models with
   // targetFormat="claude" get wrongly routed to OpenAI Responses format.
+  // #9161: a custom OpenAI-compatible Chat connection must likewise keep its configured
+  // outbound protocol when a Responses-shaped client (for example Codex) calls /responses.
   let targetFormat =
     modelTargetFormat ||
     customModelTargetFormat ||
-    (apiFormat === "responses"
+    (apiFormat === "responses" && !customOpenAICompatible
       ? FORMATS.OPENAI_RESPONSES
-      : inferredAgentRouterTargetFormat || getTargetFormat(provider, providerSpecificData));
+      : inferredAgentRouterTargetFormat || providerTargetFormat);
   if (nativeXaiResponsesPassthrough) targetFormat = FORMATS.OPENAI_RESPONSES;
   return { alias, targetFormat };
 }

tests/unit/codex-responses-to-chat-9161.test.ts

@@ -0,0 +1,67 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { resolveChatCoreTargetFormat } from "../../open-sse/handlers/chatCore/targetFormat.ts";
+import { sanitizeChatRequestBody } from "../../open-sse/handlers/chatCore/sanitization.ts";
+import { FORMATS } from "../../open-sse/translator/formats.ts";
+
+test("#9161 Responses client honors a custom OpenAI-compatible Chat target", () => {
+  const result = resolveChatCoreTargetFormat({
+    provider: "openai-compatible-chat-regression",
+    resolvedModel: "custom-chat-model",
+    apiFormat: "responses",
+    sourceFormat: FORMATS.OPENAI_RESPONSES,
+    customModelTargetFormat: undefined,
+    providerSpecificData: { apiType: "chat" },
+  });
+
+  assert.equal(result.targetFormat, FORMATS.OPENAI);
+});
+
+test("#9161 Responses client preserves a custom OpenAI-compatible Responses target", () => {
+  const result = resolveChatCoreTargetFormat({
+    provider: "openai-compatible-responses-regression",
+    resolvedModel: "custom-responses-model",
+    apiFormat: "responses",
+    sourceFormat: FORMATS.OPENAI_RESPONSES,
+    customModelTargetFormat: undefined,
+    providerSpecificData: { apiType: "responses" },
+  });
+
+  assert.equal(result.targetFormat, FORMATS.OPENAI_RESPONSES);
+});
+
+test("#9161 official OpenAI keeps Responses routing for the Responses API", () => {
+  const result = resolveChatCoreTargetFormat({
+    provider: "openai",
+    resolvedModel: "gpt-4o",
+    apiFormat: "responses",
+    sourceFormat: FORMATS.OPENAI_RESPONSES,
+    customModelTargetFormat: undefined,
+    providerSpecificData: undefined,
+  });
+
+  assert.equal(result.targetFormat, FORMATS.OPENAI_RESPONSES);
+});
+
+test("#9161 Responses source targeting Chat converts max_output_tokens to max_tokens", () => {
+  const body = sanitizeChatRequestBody(
+    { max_output_tokens: 256 },
+    FORMATS.OPENAI_RESPONSES,
+    FORMATS.OPENAI
+  );
+
+  assert.equal(body.max_tokens, 256);
+  assert.equal(body.max_output_tokens, undefined);
+});
+
+test("#9161 Responses target keeps the Responses token field", () => {
+  const body = sanitizeChatRequestBody(
+    { max_tokens: 128 },
+    FORMATS.OPENAI,
+    FORMATS.OPENAI_RESPONSES
+  );
+
+  assert.equal(body.max_output_tokens, 128);
+  assert.equal(body.max_tokens, undefined);
+});