fix(tools): scope Mastra memory cache per request

packages/tools/src/mastra/processor.ts

@@ -50,7 +50,29 @@ interface ProcessorContext {
 	addMemory: "always" | "never"
 	logger: Logger
 	promptTemplate?: PromptTemplate
-	memoryCache: MemoryCache<string>
+}
+
+const REQUEST_MEMORY_CACHE_KEY = "supermemory.memoryCache"
+
+/**
+ * Gets the cache scoped to the current Mastra request.
+ *
+ * Mastra processor instances can be reused across requests, while `state` is
+ * created for one request and shared across that request's processor calls.
+ */
+function getRequestMemoryCache(
+	state?: Record<string, unknown>,
+): MemoryCache<string> {
+	const existingCache = state?.[REQUEST_MEMORY_CACHE_KEY]
+	if (existingCache instanceof MemoryCache) {
+		return existingCache
+	}
+
+	const memoryCache = new MemoryCache<string>()
+	if (state) {
+		state[REQUEST_MEMORY_CACHE_KEY] = memoryCache
+	}
+	return memoryCache
 }
 
 /**
@@ -72,7 +94,6 @@ function createProcessorContext(
 		addMemory: options.addMemory ?? "always",
 		logger,
 		promptTemplate: options.promptTemplate,
-		memoryCache: new MemoryCache<string>(),
 	}
 }
 
@@ -136,7 +157,7 @@ export class SupermemoryInputProcessor implements Processor {
 	}
 
 	async processInput(args: ProcessInputArgs): Promise<ProcessInputResult> {
-		const { messages, messageList, requestContext } = args
+		const { messages, messageList, requestContext, state } = args
 
 		try {
 			const queryText = extractQueryText(
@@ -159,9 +180,10 @@ export class SupermemoryInputProcessor implements Processor {
 				this.ctx.mode,
 				queryText || "",
 			)
+			const memoryCache = getRequestMemoryCache(state)
 
-			if (this.ctx.memoryCache.has(turnKey)) {
-				const cachedMemories = this.ctx.memoryCache.get(turnKey) ?? ""
+			if (memoryCache.has(turnKey)) {
+				const cachedMemories = memoryCache.get(turnKey) ?? ""
 				this.ctx.logger.debug("Using cached memories", { turnKey })
 				if (cachedMemories) {
 					messageList.addSystem(cachedMemories, "supermemory")
@@ -185,7 +207,7 @@ export class SupermemoryInputProcessor implements Processor {
 				promptTemplate: this.ctx.promptTemplate,
 			})
 
-			this.ctx.memoryCache.set(turnKey, memories)
+			memoryCache.set(turnKey, memories)
 			if (memories) {
 				messageList.addSystem(memories, "supermemory")
 				this.ctx.logger.debug("Injected memories into system prompt", {

packages/tools/test/mastra/integration.test.ts

@@ -212,7 +212,7 @@ describe.skipIf(!shouldRunIntegration)(
 				fetchSpy.mockRestore()
 			})
 
-			it("should cache memories for repeated calls with same message", async () => {
+			it("should cache memories for repeated calls within a request", async () => {
 				const fetchSpy = vi.spyOn(globalThis, "fetch")
 
 				const processor = new SupermemoryInputProcessor({
@@ -226,11 +226,13 @@ describe.skipIf(!shouldRunIntegration)(
 				const messages: MastraDBMessage[] = [
 					createMessage("user", "Cache test message"),
 				]
+				const state: Record<string, unknown> = {}
 
 				const args1: ProcessInputArgs = {
 					messages,
 					systemMessages: [],
 					messageList: createIntegrationMessageList(),
+					state,
 					abort: vi.fn() as never,
 					retryCount: 0,
 				}
@@ -245,6 +247,7 @@ describe.skipIf(!shouldRunIntegration)(
 					messages,
 					systemMessages: [],
 					messageList: createIntegrationMessageList(),
+					state,
 					abort: vi.fn() as never,
 					retryCount: 0,
 				}

packages/tools/test/mastra/unit.test.ts

@@ -216,6 +216,7 @@ describe("SupermemoryInputProcessor", () => {
 			const messages: MastraDBMessage[] = [createMessage("user", "Hello")]
 			const firstMessageList = createMockMessageList()
 			const secondMessageList = createMockMessageList()
+			const state: Record<string, unknown> = {}
 
 			await processor.processInput({
 				messages,
@@ -223,7 +224,7 @@ describe("SupermemoryInputProcessor", () => {
 				messageList: firstMessageList,
 				abort: vi.fn() as never,
 				retryCount: 0,
-				state: {},
+				state,
 			})
 			await processor.processInput({
 				messages,
@@ -231,7 +232,7 @@ describe("SupermemoryInputProcessor", () => {
 				messageList: secondMessageList,
 				abort: vi.fn() as never,
 				retryCount: 0,
-				state: {},
+				state,
 			})
 
 			expect(firstMessageList.addSystem).not.toHaveBeenCalled()
@@ -239,7 +240,7 @@ describe("SupermemoryInputProcessor", () => {
 			expect(fetchMock).toHaveBeenCalledTimes(1)
 		})
 
-		it("should use cached memories on second call with same message", async () => {
+		it("should use cached memories within the same request", async () => {
 			fetchMock.mockResolvedValue({
 				ok: true,
 				json: () =>
@@ -254,11 +255,13 @@ describe("SupermemoryInputProcessor", () => {
 			})
 
 			const messages: MastraDBMessage[] = [createMessage("user", "Hello")]
+			const state: Record<string, unknown> = {}
 
 			const args1: ProcessInputArgs = {
 				messages,
 				systemMessages: [],
 				messageList: createMockMessageList(),
+				state,
 				abort: vi.fn() as never,
 				retryCount: 0,
 			}
@@ -270,6 +273,7 @@ describe("SupermemoryInputProcessor", () => {
 				messages,
 				systemMessages: [],
 				messageList: createMockMessageList(),
+				state,
 				abort: vi.fn() as never,
 				retryCount: 0,
 			}
@@ -278,6 +282,104 @@ describe("SupermemoryInputProcessor", () => {
 			expect(fetchMock).toHaveBeenCalledTimes(1)
 		})
 
+		it("should not reuse an empty profile result across requests", async () => {
+			let callCount = 0
+			fetchMock.mockImplementation(() => {
+				const currentCall = ++callCount
+				return Promise.resolve({
+					ok: true,
+					json: () =>
+						Promise.resolve(
+							currentCall === 1
+								? createMockProfileResponse()
+								: createMockProfileResponse(["Profile from second request"]),
+						),
+				})
+			})
+
+			const processor = new SupermemoryInputProcessor({
+				containerTag: TEST_CONFIG.containerTag,
+				customId: TEST_CONFIG.customId,
+				apiKey: TEST_CONFIG.apiKey,
+				mode: "profile",
+			})
+			const messages: MastraDBMessage[] = [createMessage("user", "Hello")]
+			const firstMessageList = createMockMessageList()
+			const secondMessageList = createMockMessageList()
+
+			await processor.processInput({
+				messages,
+				systemMessages: [],
+				messageList: firstMessageList,
+				state: {},
+				abort: vi.fn() as never,
+				retryCount: 0,
+			})
+			await processor.processInput({
+				messages,
+				systemMessages: [],
+				messageList: secondMessageList,
+				state: {},
+				abort: vi.fn() as never,
+				retryCount: 0,
+			})
+
+			expect(fetchMock).toHaveBeenCalledTimes(2)
+			expect(firstMessageList.addSystem).not.toHaveBeenCalled()
+			expect(secondMessageList.calls[0]?.args[0]).toContain(
+				"Profile from second request",
+			)
+		})
+
+		it("should not share a cache when request state is unavailable", async () => {
+			let callCount = 0
+			fetchMock.mockImplementation(() => {
+				const currentCall = ++callCount
+				return Promise.resolve({
+					ok: true,
+					json: () =>
+						Promise.resolve(
+							createMockProfileResponse([`Profile from call ${currentCall}`]),
+						),
+				})
+			})
+
+			const processor = new SupermemoryInputProcessor({
+				containerTag: TEST_CONFIG.containerTag,
+				customId: TEST_CONFIG.customId,
+				apiKey: TEST_CONFIG.apiKey,
+				mode: "profile",
+			})
+			const messages: MastraDBMessage[] = [createMessage("user", "Hello")]
+			const firstMessageList = createMockMessageList()
+			const secondMessageList = createMockMessageList()
+
+			await processor.processInput({
+				messages,
+				systemMessages: [],
+				messageList: firstMessageList,
+				state: undefined as never,
+				abort: vi.fn() as never,
+				retryCount: 0,
+			})
+			await processor.processInput({
+				messages,
+				systemMessages: [],
+				messageList: secondMessageList,
+				state: undefined as never,
+				abort: vi.fn() as never,
+				retryCount: 0,
+			})
+
+			expect(fetchMock).toHaveBeenCalledTimes(2)
+			expect(firstMessageList.calls[0]?.args[0]).toContain(
+				"Profile from call 1",
+			)
+			expect(secondMessageList.calls[0]?.args[0]).toContain(
+				"Profile from call 2",
+			)
+		})
+
 		it("should refetch memories for different user message", async () => {
 			let callCount = 0
 			fetchMock.mockImplementation(() => {
@@ -297,11 +399,13 @@ describe("SupermemoryInputProcessor", () => {
 				apiKey: TEST_CONFIG.apiKey,
 				mode: "query",
 			})
+			const state: Record<string, unknown> = {}
 
 			const args1: ProcessInputArgs = {
 				messages: [createMessage("user", "First message")],
 				systemMessages: [],
 				messageList: createMockMessageList(),
+				state,
 				abort: vi.fn() as never,
 				retryCount: 0,
 			}
@@ -313,6 +417,7 @@ describe("SupermemoryInputProcessor", () => {
 				messages: [createMessage("user", "Different message")],
 				systemMessages: [],
 				messageList: createMockMessageList(),
+				state,
 				abort: vi.fn() as never,
 				retryCount: 0,
 			}