fix: use azure-openai-responses for Azure custom providers (#50851) (thanks @kunalk16)

* Add azure-openai-responses * Unit tests update for updated API * Add entry for PR #50851 * Add comma to address PR comment Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Address PR comment on sanitization of output * Address review comment * Revert commits * Revert commit * Update changelog stating Azure OpenAI only Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> * Add references * Address PR comment on sanitization of output * Address review comment * Revert commits * Revert commit * Address PR comment on sanitization of output * Address review comment * Revert commits * Revert commit * Fix generated file * Add azure openai responses to OPENAI_RESPONSES_APIS * Add azure openai responses to createParallelToolCallsWrapper * Adding azure openai responses to attempt.ts * Add azure openai responses to google.ts * Address PR comment on sanitization of output * Revert commit * Address PR comment on sanitization of output * Revert commit * Address PR comment on sanitization of output * Revert commit * Fix changelog * Fix linting * fix: cover azure responses wrapper path (#50851) (thanks @kunalk16) --------- Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com> Co-authored-by: Ayaan Zaidi <hi@obviy.us>
2026-04-25 23:47:20 +00:00 · 2026-03-30 16:17:03 +05:30
parent 2fbd5e3f5f
commit 34b0a19a16
13 changed files with 75 additions and 23 deletions
--- a/src/agents/pi-embedded-runner-extraparams.test.ts
+++ b/src/agents/pi-embedded-runner-extraparams.test.ts
@@ -364,6 +364,7 @@ describe("applyExtraParamsToAgent", () => {
    applyModelId: string;
    model:
      | Model<"openai-responses">
+      | Model<"azure-openai-responses">
      | Model<"openai-codex-responses">
      | Model<"openai-completions">
      | Model<"anthropic-messages">;
@@ -418,7 +419,11 @@ describe("applyExtraParamsToAgent", () => {
  function runParallelToolCallsPayloadMutationCase(params: {
    applyProvider: string;
    applyModelId: string;
-    model: Model<"openai-completions"> | Model<"openai-responses"> | Model<"anthropic-messages">;
+    model:
+      | Model<"openai-completions">
+      | Model<"openai-responses">
+      | Model<"azure-openai-responses">
+      | Model<"anthropic-messages">;
    cfg?: Record<string, unknown>;
    extraParamsOverride?: Record<string, unknown>;
    payload?: Record<string, unknown>;
@@ -656,6 +661,34 @@ describe("applyExtraParamsToAgent", () => {
    expect(payload.parallel_tool_calls).toBe(true);
  });

+  it("injects parallel_tool_calls for azure-openai-responses payloads when configured", () => {
+    const payload = runParallelToolCallsPayloadMutationCase({
+      applyProvider: "azure-openai-responses",
+      applyModelId: "gpt-5",
+      cfg: {
+        agents: {
+          defaults: {
+            models: {
+              "azure-openai-responses/gpt-5": {
+                params: {
+                  parallelToolCalls: true,
+                },
+              },
+            },
+          },
+        },
+      },
+      model: {
+        api: "azure-openai-responses",
+        provider: "azure-openai-responses",
+        id: "gpt-5",
+        baseUrl: "https://example.openai.azure.com/openai/v1",
+      } as unknown as Model<"azure-openai-responses">,
+    });
+
+    expect(payload.parallel_tool_calls).toBe(true);
+  });
+
  it("does not inject parallel_tool_calls for unsupported APIs", () => {
    const payload = runParallelToolCallsPayloadMutationCase({
      applyProvider: "anthropic",
@@ -2664,11 +2697,11 @@ describe("applyExtraParamsToAgent", () => {
        },
      },
      model: {
-        api: "openai-responses",
+        api: "azure-openai-responses",
        provider: "azure-openai-responses",
        id: "gpt-5.4",
        baseUrl: "https://example.openai.azure.com/openai/v1",
-      } as unknown as Model<"openai-responses">,
+      } as unknown as Model<"azure-openai-responses">,
    });
    expect(payload).not.toHaveProperty("service_tier");
  });
@@ -2829,7 +2862,7 @@ describe("applyExtraParamsToAgent", () => {
      applyProvider: "azure-openai-responses",
      applyModelId: "gpt-4o",
      model: {
-        api: "openai-responses",
+        api: "azure-openai-responses",
        provider: "azure-openai-responses",
        id: "gpt-4o",
        name: "gpt-4o",
@@ -2840,7 +2873,7 @@ describe("applyExtraParamsToAgent", () => {
        contextWindow: 128_000,
        maxTokens: 16_384,
        compat: { supportsStore: false },
-      } as unknown as Model<"openai-responses">,
+      } as unknown as Model<"azure-openai-responses">,
    });
    expect(payload).not.toHaveProperty("store");
  });
@@ -2917,11 +2950,11 @@ describe("applyExtraParamsToAgent", () => {
      applyProvider: "azure-openai-responses",
      applyModelId: "gpt-4o",
      model: {
-        api: "openai-responses",
+        api: "azure-openai-responses",
        provider: "azure-openai-responses",
        id: "gpt-4o",
        baseUrl: "https://example.openai.azure.com/openai/v1",
-      } as unknown as Model<"openai-responses">,
+      } as unknown as Model<"azure-openai-responses">,
    });
    expect(payload).not.toHaveProperty("context_management");
  });
@@ -2945,11 +2978,11 @@ describe("applyExtraParamsToAgent", () => {
        },
      },
      model: {
-        api: "openai-responses",
+        api: "azure-openai-responses",
        provider: "azure-openai-responses",
        id: "gpt-4o",
        baseUrl: "https://example.openai.azure.com/openai/v1",
-      } as unknown as Model<"openai-responses">,
+      } as unknown as Model<"azure-openai-responses">,
    });
    expect(payload.context_management).toEqual([
      {
@@ -3081,16 +3114,16 @@ describe("applyExtraParamsToAgent", () => {
    expect(payload.prompt_cache_retention).toBe("24h");
  });

-  it("keeps prompt cache fields for direct Azure OpenAI openai-responses endpoints", () => {
+  it("keeps prompt cache fields for direct Azure OpenAI azure-openai-responses endpoints", () => {
    const payload = runResponsesPayloadMutationCase({
      applyProvider: "azure-openai-responses",
      applyModelId: "gpt-4o",
      model: {
-        api: "openai-responses",
+        api: "azure-openai-responses",
        provider: "azure-openai-responses",
        id: "gpt-4o",
        baseUrl: "https://example.openai.azure.com/openai/v1",
-      } as unknown as Model<"openai-responses">,
+      } as unknown as Model<"azure-openai-responses">,
      payload: {
        store: false,
        prompt_cache_key: "session-azure",
--- a/src/agents/pi-embedded-runner/extra-params.test-support.ts
+++ b/src/agents/pi-embedded-runner/extra-params.test-support.ts
@@ -22,7 +22,7 @@ function createMockStream(): ReturnType<StreamFn> {
 }

 type RunExtraParamsCaseParams<
-  TApi extends "openai-completions" | "openai-responses",
+  TApi extends "openai-completions" | "openai-responses" | "azure-openai-responses",
  TPayload extends Record<string, unknown>,
 > = {
  applyModelId?: string;
@@ -36,7 +36,7 @@ type RunExtraParamsCaseParams<
 };

 export function runExtraParamsCase<
-  TApi extends "openai-completions" | "openai-responses",
+  TApi extends "openai-completions" | "openai-responses" | "azure-openai-responses",
  TPayload extends Record<string, unknown>,
 >(params: RunExtraParamsCaseParams<TApi, TPayload>): ExtraParamsCapture<TPayload> {
  const captured: ExtraParamsCapture<TPayload> = {
--- a/src/agents/pi-embedded-runner/extra-params.ts
+++ b/src/agents/pi-embedded-runner/extra-params.ts
@@ -278,7 +278,11 @@ function createParallelToolCallsWrapper(
 ): StreamFn {
  const underlying = baseStreamFn ?? streamSimple;
  return (model, context, options) => {
-    if (model.api !== "openai-completions" && model.api !== "openai-responses") {
+    if (
+      model.api !== "openai-completions" &&
+      model.api !== "openai-responses" &&
+      model.api !== "azure-openai-responses"
+    ) {
      return underlying(model, context, options);
    }
    log.debug(
@@ -439,7 +443,10 @@ function applyPostPluginStreamWrappers(
        log.debug(
          `applying OpenAI text verbosity=${openAITextVerbosity} for ${ctx.provider}/${ctx.modelId}`,
        );
-        ctx.agent.streamFn = createOpenAITextVerbosityWrapper(ctx.agent.streamFn, openAITextVerbosity);
+        ctx.agent.streamFn = createOpenAITextVerbosityWrapper(
+          ctx.agent.streamFn,
+          openAITextVerbosity,
+        );
      }
    }
  }
--- a/src/agents/pi-embedded-runner/google.ts
+++ b/src/agents/pi-embedded-runner/google.ts
@@ -627,7 +627,9 @@ export async function sanitizeSessionHistory(params: {
  );

  const isOpenAIResponsesApi =
-    params.modelApi === "openai-responses" || params.modelApi === "openai-codex-responses";
+    params.modelApi === "openai-responses" ||
+    params.modelApi === "openai-codex-responses" ||
+    params.modelApi === "azure-openai-responses";
  const hasSnapshot = Boolean(params.provider || params.modelApi || params.modelId);
  const priorSnapshot = hasSnapshot ? readLastModelSnapshot(params.sessionManager) : null;
  const modelChanged = priorSnapshot
--- a/src/agents/pi-embedded-runner/model.ts
+++ b/src/agents/pi-embedded-runner/model.ts
@@ -99,6 +99,7 @@ function normalizeResolvedTransportApi(api: unknown): ModelDefinitionConfig["api
    case "openai-codex-responses":
    case "openai-completions":
    case "openai-responses":
+    case "azure-openai-responses":
      return api;
    default:
      return undefined;
--- a/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/openai-stream-wrappers.ts
@@ -8,7 +8,7 @@ import { streamWithPayloadPatch } from "./stream-payload-utils.js";
 type OpenAIServiceTier = "auto" | "default" | "flex" | "priority";
 type OpenAITextVerbosity = "low" | "medium" | "high";

-const OPENAI_RESPONSES_APIS = new Set(["openai-responses"]);
+const OPENAI_RESPONSES_APIS = new Set(["openai-responses", "azure-openai-responses"]);
 const OPENAI_RESPONSES_PROVIDERS = new Set(["openai", "azure-openai", "azure-openai-responses"]);

 function isDirectOpenAIBaseUrl(baseUrl: unknown): boolean {
@@ -358,7 +358,9 @@ export function createOpenAIFastModeWrapper(baseStreamFn: StreamFn | undefined):
  const underlying = baseStreamFn ?? streamSimple;
  return (model, context, options) => {
    if (
-      (model.api !== "openai-responses" && model.api !== "openai-codex-responses") ||
+      (model.api !== "openai-responses" &&
+        model.api !== "openai-codex-responses" &&
+        model.api !== "azure-openai-responses") ||
      (model.provider !== "openai" && model.provider !== "openai-codex")
    ) {
      return underlying(model, context, options);
--- a/src/agents/pi-embedded-runner/run/attempt.ts
+++ b/src/agents/pi-embedded-runner/run/attempt.ts
@@ -992,6 +992,7 @@ export async function runEmbeddedAttempt(

      if (
        params.model.api === "openai-responses" ||
+        params.model.api === "azure-openai-responses" ||
        params.model.api === "openai-codex-responses"
      ) {
        const inner = activeSession.agent.streamFn;
--- a/src/agents/transcript-policy.ts
+++ b/src/agents/transcript-policy.ts
@@ -81,7 +81,9 @@ export function resolveTranscriptPolicy(params: {
  const requiresOpenAiCompatibleToolIdSanitization =
    params.modelApi === "openai-completions" ||
    (!isOpenAi &&
-      (params.modelApi === "openai-responses" || params.modelApi === "openai-codex-responses"));
+      (params.modelApi === "openai-responses" ||
+        params.modelApi === "openai-codex-responses" ||
+        params.modelApi === "azure-openai-responses"));

  // Anthropic Claude endpoints can reject replayed `thinking` blocks unless the
  // original signatures are preserved byte-for-byte. Drop them at send-time to
--- a/src/commands/onboard-custom.test.ts
+++ b/src/commands/onboard-custom.test.ts
@@ -482,7 +482,7 @@ describe("applyCustomApiConfig", () => {
    const provider = result.config.models?.providers?.[providerId];

    expect(provider?.baseUrl).toBe("https://user123-resource.openai.azure.com/openai/v1");
-    expect(provider?.api).toBe("openai-responses");
+    expect(provider?.api).toBe("azure-openai-responses");
    expect(provider?.authHeader).toBe(false);
    expect(provider?.headers).toEqual({ "api-key": "abcd1234" });

@@ -568,7 +568,7 @@ describe("applyCustomApiConfig", () => {
    expect(result.providerIdRenamedFrom).toBeUndefined();
    const provider = result.config.models?.providers?.[oldProviderId];
    expect(provider?.baseUrl).toBe("https://my-resource.openai.azure.com/openai/v1");
-    expect(provider?.api).toBe("openai-responses");
+    expect(provider?.api).toBe("azure-openai-responses");
    expect(provider?.authHeader).toBe(false);
    expect(provider?.headers).toEqual({ "api-key": "key789" });
  });
--- a/src/commands/onboard-custom.ts
+++ b/src/commands/onboard-custom.ts
@@ -687,7 +687,7 @@ export function applyCustomApiConfig(params: ApplyCustomApiConfigParams): Custom
    normalizeOptionalProviderApiKey(existingApiKey);

  const providerApi = isAzureOpenAi
-    ? ("openai-responses" as const)
+    ? ("azure-openai-responses" as const)
    : resolveProviderApi(params.compatibility);
  const azureHeaders = isAzure && normalizedApiKey ? { "api-key": normalizedApiKey } : undefined;

--- a/src/config/schema.base.generated.ts
+++ b/src/config/schema.base.generated.ts
@@ -1038,6 +1038,7 @@ export const GENERATED_BASE_CONFIG_SCHEMA = {
                    "github-copilot",
                    "bedrock-converse-stream",
                    "ollama",
+                    "azure-openai-responses",
                  ],
                },
                injectNumCtxForOpenAICompat: {
@@ -1142,6 +1143,7 @@ export const GENERATED_BASE_CONFIG_SCHEMA = {
                          "github-copilot",
                          "bedrock-converse-stream",
                          "ollama",
+                          "azure-openai-responses",
                        ],
                      },
                      reasoning: {
--- a/src/config/types.models.ts
+++ b/src/config/types.models.ts
@@ -10,6 +10,7 @@ export const MODEL_APIS = [
  "github-copilot",
  "bedrock-converse-stream",
  "ollama",
+  "azure-openai-responses",
 ] as const;

 export type ModelApi = (typeof MODEL_APIS)[number];