Skip to content

A self-improving RLM agent for coding workflows and long-running autonomous tasks.

Rust

Stop Azure OpenAI from storing responses server-side (#2948)

Sebastian Müller4e20e9f81781
diff --git a/packages/ai/.changes/azure-responses-store-false.md b/packages/ai/.changes/azure-responses-store-false.md
new file mode 100644
index 0000000000..3f960dfb22
--- /dev/null
+++ b/packages/ai/.changes/azure-responses-store-false.md
@@ -0,0 +1 @@
+- Fixed Azure OpenAI Responses requests to send `store: false`, so responses are no longer stored server-side, and to omit `prompt_cache_key` when `cacheRetention` is `"none"`, matching the OpenAI Responses provider.
diff --git a/packages/ai/src/providers/azure-openai-responses.ts b/packages/ai/src/providers/azure-openai-responses.ts
index b52632086d..452cee0c66 100644
--- a/packages/ai/src/providers/azure-openai-responses.ts
+++ b/packages/ai/src/providers/azure-openai-responses.ts
@@ -246,7 +246,8 @@ function buildParams(
model: deploymentName,
input: messages,
stream: true,
- prompt_cache_key: options?.sessionId,
+ prompt_cache_key: options?.cacheRetention === "none" ? undefined : options?.sessionId,
+ store: false,
};
if (options?.maxTokens) {
diff --git a/packages/ai/test/azure-openai-base-url.test.ts b/packages/ai/test/azure-openai-base-url.test.ts
index fcc358210f..19d5042ced 100644
--- a/packages/ai/test/azure-openai-base-url.test.ts
+++ b/packages/ai/test/azure-openai-base-url.test.ts
@@ -1,6 +1,6 @@
import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
import { streamAzureOpenAIResponses } from "../src/providers/azure-openai-responses.js";
-import type { Context } from "../src/types.js";
+import type { Context, StreamOptions } from "../src/types.js";
import { getFixtureModel } from "./fixture-models.js";
interface CapturedAzureClientOptions {
@@ -13,12 +13,14 @@ interface CapturedAzureClientOptions {
const azureMock = vi.hoisted(() => ({
constructorCalls: [] as CapturedAzureClientOptions[],
+ createCalls: [] as unknown[],
}));
vi.mock("openai", () => {
class AzureOpenAI {
responses = {
- create: () => {
+ create: (params: unknown) => {
+ azureMock.createCalls.push(params);
throw new Error("mock create");
},
};
@@ -45,6 +47,7 @@ const originalEnv = Object.fromEntries(azureEnvVars.map((name) => [name, process
beforeEach(() => {
azureMock.constructorCalls.length = 0;
+ azureMock.createCalls.length = 0;
for (const name of azureEnvVars) delete process.env[name];
});
@@ -56,9 +59,10 @@ afterEach(() => {
}
});
-async function streamOnce() {
+async function streamOnce(options: StreamOptions = {}) {
return streamAzureOpenAIResponses(getFixtureModel("azure-openai-responses", "gpt-4o-mini"), context, {
apiKey: "test-api-key",
+ ...options,
}).result();
}
@@ -117,3 +121,12 @@ describe("azure-openai-responses base URL normalization", () => {
expect(result.errorMessage).toContain("Invalid Azure OpenAI base URL");
});
});
+
+it.each([
+ [undefined, "session-1"],
+ ["none", undefined],
+] as const)("sends store: false and honors cacheRetention=%s", async (cacheRetention, key) => {
+ process.env.AZURE_OPENAI_RESOURCE_NAME = "res";
+ await streamOnce({ sessionId: "session-1", cacheRetention });
+ expect(azureMock.createCalls).toMatchObject([{ store: false, prompt_cache_key: key }]);
+});