diff --git a/packages/tools/src/openai/middleware.test.ts b/packages/tools/src/openai/middleware.test.ts
new file mode 100644
index 000000000..6efbd8524
--- /dev/null
+++ b/packages/tools/src/openai/middleware.test.ts
@@ -0,0 +1,121 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest"
+
+// Mock the Supermemory SDK and the conversations client so the middleware's
+// memory-save routing can be asserted without network access.
+const { clientAddMock, addConversationMock } = vi.hoisted(() => ({
+	clientAddMock: vi.fn(),
+	addConversationMock: vi.fn(),
+}))
+
+vi.mock("supermemory", () => ({
+	default: class MockSupermemory {
+		add = clientAddMock
+	},
+}))
+
+vi.mock("../conversations-client", () => ({
+	addConversation: addConversationMock,
+}))
+
+import { createOpenAIMiddleware } from "./middleware"
+
+const EMPTY_PROFILE = {
+	profile: { static: [], dynamic: [] },
+	searchResults: { results: [] },
+}
+
+function makeFakeOpenAI() {
+	return {
+		chat: {
+			completions: {
+				create: vi.fn().mockResolvedValue({ id: "chatcmpl-1" }),
+			},
+		},
+		responses: {
+			create: vi.fn().mockResolvedValue({ id: "resp-1" }),
+		},
+	}
+}
+
+describe("createOpenAIMiddleware Responses API memory save", () => {
+	beforeEach(() => {
+		clientAddMock.mockReset().mockResolvedValue({ id: "mem-1" })
+		addConversationMock.mockReset().mockResolvedValue({ id: "conv-doc-1" })
+		process.env.SUPERMEMORY_API_KEY = "sm_test_key"
+		vi.stubGlobal(
+			"fetch",
+			vi.fn(async () => ({
+				ok: true,
+				json: async () => EMPTY_PROFILE,
+			})),
+		)
+	})
+
+	afterEach(() => {
+		vi.unstubAllGlobals()
+	})
+
+	it("routes a customId save through /v4/conversations like the chat path", async () => {
+		const openai = makeFakeOpenAI()
+		// biome-ignore lint/suspicious/noExplicitAny: minimal OpenAI stub for the wrapped surface
+		createOpenAIMiddleware(openai as any, "user-1", {
+			containerTag: "user-1",
+			customId: "conv-1",
+			addMemory: "always",
+			mode: "query",
+		})
+
+		await openai.responses.create({ input: "I love hiking", instructions: "" })
+
+		expect(addConversationMock).toHaveBeenCalledTimes(1)
+		const payload = addConversationMock.mock.calls[0]?.[0]
+		expect(payload.conversationId).toBe("conv-1")
+		expect(payload.messages).toEqual([
+			{ role: "user", content: "I love hiking" },
+		])
+		// The raw client.add fallback (which would have stored the internal
+		// "conversation:" prefix as the customId) must not fire.
+		expect(clientAddMock).not.toHaveBeenCalled()
+	})
+
+	it("never stores the internal conversation: prefix in the add fallback", async () => {
+		const openai = makeFakeOpenAI()
+		// biome-ignore lint/suspicious/noExplicitAny: minimal OpenAI stub for the wrapped surface
+		createOpenAIMiddleware(openai as any, "user-1", {
+			containerTag: "user-1",
+			customId: "conv-1",
+			addMemory: "always",
+			mode: "query",
+		})
+
+		// Without an API key the conversation branch cannot run, so the save
+		// falls through to client.add — the stored customId must be the
+		// user's configured id, not the prefixed routing marker.
+		delete process.env.SUPERMEMORY_API_KEY
+
+		await openai.responses.create({ input: "I love hiking", instructions: "" })
+
+		expect(addConversationMock).not.toHaveBeenCalled()
+		expect(clientAddMock).toHaveBeenCalledTimes(1)
+		expect(clientAddMock.mock.calls[0]?.[0]?.customId).toBe("conv-1")
+	})
+
+	it("saves without a customId via plain client.add", async () => {
+		const openai = makeFakeOpenAI()
+		// biome-ignore lint/suspicious/noExplicitAny: minimal OpenAI stub for the wrapped surface
+		createOpenAIMiddleware(openai as any, "user-1", {
+			containerTag: "user-1",
+			addMemory: "always",
+			mode: "query",
+			// biome-ignore lint/suspicious/noExplicitAny: exercising the optional-customId path
+		} as any)
+
+		await openai.responses.create({ input: "I love hiking", instructions: "" })
+
+		expect(addConversationMock).not.toHaveBeenCalled()
+		expect(clientAddMock).toHaveBeenCalledTimes(1)
+		const arg = clientAddMock.mock.calls[0]?.[0]
+		expect(arg.content).toBe("I love hiking")
+		expect(arg.customId).toBeUndefined()
+	})
+})
diff --git a/packages/tools/src/openai/middleware.ts b/packages/tools/src/openai/middleware.ts
index 7fc4267e7..78df52da4 100644
--- a/packages/tools/src/openai/middleware.ts
+++ b/packages/tools/src/openai/middleware.ts
@@ -367,11 +367,13 @@ const addMemoryTool = async (
 			return
 		}
 
-		// Fallback to old behavior for non-conversation memories
+		// Fallback to old behavior for non-conversation memories. The
+		// "conversation:" prefix is an internal routing marker — never let
+		// it reach the stored customId (colons fail customId validation).
 		const response = await client.add({
 			content,
 			containerTags: [containerTag],
-			customId,
+			customId: customId?.replace(/^conversation:/, ""),
 		})
 
 		logger.info("Memory saved successfully", {
@@ -548,9 +550,26 @@ export function createOpenAIMiddleware(
 		if (addMemory === "always" && input?.trim()) {
 			const content = customId ? `Input: ${input}` : input
 			const memoryCustomId = customId ? `conversation:${customId}` : undefined
+			// Mirror the chat path: hand the input over as messages together
+			// with the credentials so a configured customId routes through
+			// /v4/conversations. Without these, addMemoryTool fell through to
+			// client.add() with the internal "conversation:" prefix baked
+			// into the stored customId — which the API rejects, so Responses
+			// API memory saves silently failed whenever customId was set.
+			const inputAsMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] =
+				[{ role: "user", content: input }]
 
 			operations.push(
-				addMemoryTool(client, containerTag, content, memoryCustomId, logger),
+				addMemoryTool(
+					client,
+					containerTag,
+					content,
+					memoryCustomId,
+					logger,
+					customId ? inputAsMessages : undefined,
+					process.env.SUPERMEMORY_API_KEY,
+					baseUrl,
+				),
 			)
 		}