@mastra/memory 1.5.2 → 1.6.0-alpha.0

package/CHANGELOG.md

@@ -1,5 +1,32 @@
 # @mastra/memory
 
+## 1.6.0-alpha.0
+
+### Minor Changes
+
+- Added Observational Memory cloning when forking threads. Thread-scoped OM is cloned with remapped message IDs. Resource-scoped OM is shared when the resourceId stays the same, and cloned with remapped thread tags when the resourceId changes. Only the current OM generation is cloned (older history generations are not copied). If OM cloning fails, the already-persisted thread clone is rolled back. ([#13569](https://github.com/mastra-ai/mastra/pull/13569))
+
+### Patch Changes
+
+- Stop auto-applying `maxOutputTokens: 100_000` to observer and reflector models when users supply their own model. ([#13563](https://github.com/mastra-ai/mastra/pull/13563))
+
+  Previously, the 100k default was always injected regardless of model choice. Now it is only applied when using the built-in default model (`google/gemini-2.5-flash`). If you set a custom `observation.model` or `reflection.model`, no `maxOutputTokens` default is added — pass it explicitly in `modelSettings` if your model needs it.
+
+- Fixed duplicate user messages when observational memory async buffering is active. Sealed-for-buffering messages without observation markers are now skipped in per-step save instead of being re-saved with new IDs. ([#13130](https://github.com/mastra-ai/mastra/pull/13130))
+
+- Fixed stale continuation hints in observational memory. ([#13606](https://github.com/mastra-ai/mastra/pull/13606))
+
+  When newer outputs omit continuation hints, old hints are now cleared. This prevents outdated task and response guidance from appearing in later turns.
+
+- Fixed a bug where async buffered observations generated stale continuation hints (suggested response and current task) that would be injected into the agent's context on the next turn, causing the agent to reply to old messages or work on already-completed tasks. ([#13678](https://github.com/mastra-ai/mastra/pull/13678))
+
+- Fixed observational memory compatibility with Codex and other stream-only providers. Observer and reflector calls now use the streaming API internally, so providers that require `stream: true` in requests work correctly out of the box. ([#13563](https://github.com/mastra-ai/mastra/pull/13563))
+
+- Fixed Memory.recall() to include pagination metadata (total, page, perPage, hasMore) in its response, ensuring consistent pagination regardless of whether agentId is provided. Fixes #13277 ([#13278](https://github.com/mastra-ai/mastra/pull/13278))
+
+- Updated dependencies [[`504fc8b`](https://github.com/mastra-ai/mastra/commit/504fc8b9d0ddab717577ad3bf9c95ea4bd5377bd), [`f9c150b`](https://github.com/mastra-ai/mastra/commit/f9c150b7595ad05ad9cc9a11098e2944361e8c22), [`88de7e8`](https://github.com/mastra-ai/mastra/commit/88de7e8dfe4b7e1951a9e441bb33136e705ce24e), [`edee4b3`](https://github.com/mastra-ai/mastra/commit/edee4b37dff0af515fc7cc0e8d71ee39e6a762f0), [`3790c75`](https://github.com/mastra-ai/mastra/commit/3790c7578cc6a47d854eb12d89e6b1912867fe29), [`e7a235b`](https://github.com/mastra-ai/mastra/commit/e7a235be6472e0c870ed6c791ddb17c492dc188b), [`d51d298`](https://github.com/mastra-ai/mastra/commit/d51d298953967aab1f58ec965b644d109214f085), [`6dbeeb9`](https://github.com/mastra-ai/mastra/commit/6dbeeb94a8b1eebb727300d1a98961f882180794), [`d5f0d8d`](https://github.com/mastra-ai/mastra/commit/d5f0d8d6a03e515ddaa9b5da19b7e44b8357b07b), [`09c3b18`](https://github.com/mastra-ai/mastra/commit/09c3b1802ff14e243a8a8baea327440bc8cc2e32), [`b896379`](https://github.com/mastra-ai/mastra/commit/b8963791c6afa79484645fcec596a201f936b9a2), [`85c84eb`](https://github.com/mastra-ai/mastra/commit/85c84ebb78aebfcba9d209c8e152b16d7a00cb71), [`a89272a`](https://github.com/mastra-ai/mastra/commit/a89272a5d71939b9fcd284e6a6dc1dd091a6bdcf), [`ee9c8df`](https://github.com/mastra-ai/mastra/commit/ee9c8df644f19d055af5f496bf4942705f5a47b7), [`77b4a25`](https://github.com/mastra-ai/mastra/commit/77b4a254e51907f8ff3a3ba95596a18e93ae4b35), [`276246e`](https://github.com/mastra-ai/mastra/commit/276246e0b9066a1ea48bbc70df84dbe528daaf99), [`08ecfdb`](https://github.com/mastra-ai/mastra/commit/08ecfdbdad6fb8285deef86a034bdf4a6047cfca), [`d5f628c`](https://github.com/mastra-ai/mastra/commit/d5f628ca86c6f6f3ff1035d52f635df32dd81cab), [`524c0f3`](https://github.com/mastra-ai/mastra/commit/524c0f3c434c3d9d18f66338dcef383d6161b59c), [`c18a0e9`](https://github.com/mastra-ai/mastra/commit/c18a0e9cef1e4ca004b2963d35e4cfc031971eac), [`4bd21ea`](https://github.com/mastra-ai/mastra/commit/4bd21ea43d44d0a0427414fc047577f9f0aa3bec), [`115a7a4`](https://github.com/mastra-ai/mastra/commit/115a7a47db5e9896fec12ae6507501adb9ec89bf), [`22a48ae`](https://github.com/mastra-ai/mastra/commit/22a48ae2513eb54d8d79dad361fddbca97a155e8), [`3c6ef79`](https://github.com/mastra-ai/mastra/commit/3c6ef798481e00d6d22563be2de98818fd4dd5e0), [`9311c17`](https://github.com/mastra-ai/mastra/commit/9311c17d7a0640d9c4da2e71b814dc67c57c6369), [`7edf78f`](https://github.com/mastra-ai/mastra/commit/7edf78f80422c43e84585f08ba11df0d4d0b73c5), [`1c4221c`](https://github.com/mastra-ai/mastra/commit/1c4221cf6032ec98d0e094d4ee11da3e48490d96), [`d25b9ea`](https://github.com/mastra-ai/mastra/commit/d25b9eabd400167255a97b690ffbc4ee4097ded5), [`fe1ce5c`](https://github.com/mastra-ai/mastra/commit/fe1ce5c9211c03d561606fda95cbfe7df1d9a9b5), [`b03c0e0`](https://github.com/mastra-ai/mastra/commit/b03c0e0389a799523929a458b0509c9e4244d562), [`0a8366b`](https://github.com/mastra-ai/mastra/commit/0a8366b0a692fcdde56c4d526e4cf03c502ae4ac), [`85664e9`](https://github.com/mastra-ai/mastra/commit/85664e9fd857320fbc245e301f764f45f66f32a3), [`bc79650`](https://github.com/mastra-ai/mastra/commit/bc796500c6e0334faa158a96077e3fb332274869), [`9257d01`](https://github.com/mastra-ai/mastra/commit/9257d01d1366d81f84c582fe02b5e200cf9621f4), [`3a3a59e`](https://github.com/mastra-ai/mastra/commit/3a3a59e8ffaa6a985fe3d9a126a3f5ade11a6724), [`3108d4e`](https://github.com/mastra-ai/mastra/commit/3108d4e649c9fddbf03253a6feeb388a5fa9fa5a), [`0c33b2c`](https://github.com/mastra-ai/mastra/commit/0c33b2c9db537f815e1c59e2c898ffce2e395a79), [`191e5bd`](https://github.com/mastra-ai/mastra/commit/191e5bd29b82f5bda35243945790da7bc7b695c2), [`f77cd94`](https://github.com/mastra-ai/mastra/commit/f77cd94c44eabed490384e7d19232a865e13214c), [`e8135c7`](https://github.com/mastra-ai/mastra/commit/e8135c7e300dac5040670eec7eab896ac6092e30), [`daca48f`](https://github.com/mastra-ai/mastra/commit/daca48f0fb17b7ae0b62a2ac40cf0e491b2fd0b7), [`257d14f`](https://github.com/mastra-ai/mastra/commit/257d14faca5931f2e4186fc165b6f0b1f915deee), [`352f25d`](https://github.com/mastra-ai/mastra/commit/352f25da316b24cdd5b410fd8dddf6a8b763da2a), [`93477d0`](https://github.com/mastra-ai/mastra/commit/93477d0769b8a13ea5ed73d508d967fb23eaeed9), [`31c78b3`](https://github.com/mastra-ai/mastra/commit/31c78b3eb28f58a8017f1dcc795c33214d87feac), [`0bc0720`](https://github.com/mastra-ai/mastra/commit/0bc07201095791858087cc56f353fcd65e87ab54), [`36516ac`](https://github.com/mastra-ai/mastra/commit/36516aca1021cbeb42e74751b46a2614101f37c8), [`e947652`](https://github.com/mastra-ai/mastra/commit/e9476527fdecb4449e54570e80dfaf8466901254), [`3c6ef79`](https://github.com/mastra-ai/mastra/commit/3c6ef798481e00d6d22563be2de98818fd4dd5e0), [`9257d01`](https://github.com/mastra-ai/mastra/commit/9257d01d1366d81f84c582fe02b5e200cf9621f4), [`ec248f6`](https://github.com/mastra-ai/mastra/commit/ec248f6b56e8a037c066c49b2178e2507471d988)]:
+  - @mastra/core@1.9.0-alpha.0
+
 ## 1.5.2
 
 ### Patch Changes

package/LICENSE.md

@@ -1,3 +1,18 @@
+Portions of this software are licensed as follows:
+
+- All content that resides under any directory named "ee/" within this
+  repository, including but not limited to:
+  - `packages/core/src/auth/ee/`
+  - `packages/server/src/server/auth/ee/`
+    is licensed under the license defined in `ee/LICENSE`.
+
+- All third-party components incorporated into the Mastra Software are
+  licensed under the original license provided by the owner of the
+  applicable component.
+
+- Content outside of the above-mentioned directories or restrictions is
+  available under the "Apache License 2.0" as defined below.
+
 # Apache License 2.0
 
 Copyright (c) 2025 Kepler Software, Inc.

package/dist/{chunk-PVFLHAZX.cjs → chunk-5UYAHJVJ.cjs}

@@ -1622,6 +1622,11 @@ See https://mastra.ai/docs/memory/observational-memory#models for model recommen
     const messageTokens = config.observation?.messageTokens ?? OBSERVATIONAL_MEMORY_DEFAULTS.observation.messageTokens;
     const observationTokens = config.reflection?.observationTokens ?? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.observationTokens;
     const isSharedBudget = config.shareTokenBudget ?? false;
+    const isDefaultModelSelection = (model) => model === void 0 || model === "default";
+    const observationSelectedModel = config.model ?? config.observation?.model ?? config.reflection?.model;
+    const reflectionSelectedModel = config.model ?? config.reflection?.model ?? config.observation?.model;
+    const observationDefaultMaxOutputTokens = config.observation?.modelSettings?.maxOutputTokens ?? (isDefaultModelSelection(observationSelectedModel) ? OBSERVATIONAL_MEMORY_DEFAULTS.observation.modelSettings.maxOutputTokens : void 0);
+    const reflectionDefaultMaxOutputTokens = config.reflection?.modelSettings?.maxOutputTokens ?? (isDefaultModelSelection(reflectionSelectedModel) ? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.modelSettings.maxOutputTokens : void 0);
     const totalBudget = messageTokens + observationTokens;
     const userExplicitlyConfiguredAsync = config.observation?.bufferTokens !== void 0 || config.observation?.bufferActivation !== void 0 || config.reflection?.bufferActivation !== void 0;
     const asyncBufferingDisabled = config.observation?.bufferTokens === false || config.scope === "resource" && !userExplicitlyConfiguredAsync;
@@ -1653,7 +1658,7 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
       shareTokenBudget: isSharedBudget,
       modelSettings: {
         temperature: config.observation?.modelSettings?.temperature ?? OBSERVATIONAL_MEMORY_DEFAULTS.observation.modelSettings.temperature,
-        maxOutputTokens: config.observation?.modelSettings?.maxOutputTokens ?? OBSERVATIONAL_MEMORY_DEFAULTS.observation.modelSettings.maxOutputTokens
+        ...observationDefaultMaxOutputTokens !== void 0 ? { maxOutputTokens: observationDefaultMaxOutputTokens } : {}
       },
       providerOptions: config.observation?.providerOptions ?? OBSERVATIONAL_MEMORY_DEFAULTS.observation.providerOptions,
       maxTokensPerBatch: config.observation?.maxTokensPerBatch ?? OBSERVATIONAL_MEMORY_DEFAULTS.observation.maxTokensPerBatch,
@@ -1674,7 +1679,7 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
       shareTokenBudget: isSharedBudget,
       modelSettings: {
         temperature: config.reflection?.modelSettings?.temperature ?? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.modelSettings.temperature,
-        maxOutputTokens: config.reflection?.modelSettings?.maxOutputTokens ?? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.modelSettings.maxOutputTokens
+        ...reflectionDefaultMaxOutputTokens !== void 0 ? { maxOutputTokens: reflectionDefaultMaxOutputTokens } : {}
       },
       providerOptions: config.reflection?.providerOptions ?? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.providerOptions,
       bufferActivation: asyncBufferingDisabled ? void 0 : config?.reflection?.bufferActivation ?? OBSERVATIONAL_MEMORY_DEFAULTS.reflection.bufferActivation,
@@ -2416,18 +2421,17 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
     const agent = this.getObserverAgent();
     const prompt = buildObserverPrompt(existingObservations, messagesToObserve, options);
     const doGenerate = async () => {
-      const result2 = await this.withAbortCheck(
-        () => agent.generate(prompt, {
+      return this.withAbortCheck(async () => {
+        const streamResult = await agent.stream(prompt, {
           modelSettings: {
             ...this.observationConfig.modelSettings
           },
           providerOptions: this.observationConfig.providerOptions,
           ...abortSignal ? { abortSignal } : {},
           ...options?.requestContext ? { requestContext: options.requestContext } : {}
-        }),
-        abortSignal
-      );
-      return result2;
+        });
+        return streamResult.getFullOutput();
+      }, abortSignal);
     };
     let result = await doGenerate();
     let parsed = parseObserverOutput(result.text);
@@ -2474,17 +2478,17 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
       this.observedMessageIds.add(msg.id);
     }
     const doGenerate = async () => {
-      return this.withAbortCheck(
-        () => agent$1.generate(prompt, {
+      return this.withAbortCheck(async () => {
+        const streamResult = await agent$1.stream(prompt, {
           modelSettings: {
             ...this.observationConfig.modelSettings
           },
           providerOptions: this.observationConfig.providerOptions,
           ...abortSignal ? { abortSignal } : {},
           ...requestContext ? { requestContext } : {}
-        }),
-        abortSignal
-      );
+        });
+        return streamResult.getFullOutput();
+      }, abortSignal);
     };
     let result = await doGenerate();
     let parsed = parseMultiThreadObserverOutput(result.text);
@@ -2542,8 +2546,8 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
         `[OM:callReflector] ${isRetry ? `retry #${attemptNumber - 1}` : "first attempt"}: level=${currentLevel}, originalTokens=${originalTokens}, targetThreshold=${targetThreshold}, promptLen=${prompt.length}, skipContinuationHints=${skipContinuationHints}`
       );
       let chunkCount = 0;
-      const result = await this.withAbortCheck(
-        () => agent.generate(prompt, {
+      const result = await this.withAbortCheck(async () => {
+        const streamResult = await agent.stream(prompt, {
           modelSettings: {
             ...this.reflectionConfig.modelSettings
           },
@@ -2570,9 +2574,9 @@ Async buffering is enabled by default \u2014 this opt-out is only needed when us
               omError(`[OM:callReflector] onError after ${chunkCount} chunks`, error);
             }
           } : {}
-        }),
-        abortSignal
-      );
+        });
+        return streamResult.getFullOutput();
+      }, abortSignal);
       omDebug(
         `[OM:callReflector] attempt #${attemptNumber} returned: textLen=${result.text?.length}, textPreview="${result.text?.slice(0, 120)}...", inputTokens=${result.usage?.inputTokens ?? result.totalUsage?.inputTokens}, outputTokens=${result.usage?.outputTokens ?? result.totalUsage?.outputTokens}`
       );
@@ -2911,19 +2915,17 @@ ${suggestedResponse}
           omDebug(
             `[OM:threshold] activation succeeded, obsTokens=${updatedRecord.observationTokenCount}, activeObsLen=${updatedRecord.activeObservations?.length}`
           );
-          if (activationResult.suggestedContinuation || activationResult.currentTask) {
-            const thread = await this.storage.getThreadById({ threadId });
-            if (thread) {
-              const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
-                suggestedResponse: activationResult.suggestedContinuation,
-                currentTask: activationResult.currentTask
-              });
-              await this.storage.updateThread({
-                id: threadId,
-                title: thread.title ?? "",
-                metadata: newMetadata
-              });
-            }
+          const thread = await this.storage.getThreadById({ threadId });
+          if (thread) {
+            const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
+              suggestedResponse: activationResult.suggestedContinuation,
+              currentTask: activationResult.currentTask
+            });
+            await this.storage.updateThread({
+              id: threadId,
+              title: thread.title ?? "",
+              metadata: newMetadata
+            });
           }
           await this.maybeAsyncReflect(
             updatedRecord,
@@ -3279,19 +3281,17 @@ ${suggestedResponse}
             _ObservationalMemory.lastBufferedBoundary.set(bufKey, 0);
             this.storage.setBufferingObservationFlag(record.id, false, 0).catch(() => {
             });
-            if (activationResult.suggestedContinuation || activationResult.currentTask) {
-              const thread = await this.storage.getThreadById({ threadId });
-              if (thread) {
-                const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
-                  suggestedResponse: activationResult.suggestedContinuation,
-                  currentTask: activationResult.currentTask
-                });
-                await this.storage.updateThread({
-                  id: threadId,
-                  title: thread.title ?? "",
-                  metadata: newMetadata
-                });
-              }
+            const thread = await this.storage.getThreadById({ threadId });
+            if (thread) {
+              const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
+                suggestedResponse: activationResult.suggestedContinuation,
+                currentTask: activationResult.currentTask
+              });
+              await this.storage.updateThread({
+                id: threadId,
+                title: thread.title ?? "",
+                metadata: newMetadata
+              });
             }
             await this.maybeReflect({
               record,
@@ -3510,24 +3510,30 @@ ${suggestedResponse}
     return messageList;
   }
   /**
-   * Save messages to storage, regenerating IDs for any messages that were
-   * previously saved with observation markers (sealed).
+   * Save messages to storage while preventing duplicate inserts for sealed messages.
    *
-   * After saving, tracks which messages now have observation markers
-   * so their IDs won't be reused in future save cycles.
+   * Sealed messages that do not yet contain a completed observation boundary are
+   * skipped because async buffering already persisted them.
    */
   async saveMessagesWithSealedIdTracking(messagesToSave, sealedIds, threadId, resourceId, state) {
+    const filteredMessages = [];
     for (const msg of messagesToSave) {
       if (sealedIds.has(msg.id)) {
-        msg.id = crypto.randomUUID();
+        if (this.findLastCompletedObservationBoundary(msg) !== -1) {
+          filteredMessages.push(msg);
+        }
+      } else {
+        filteredMessages.push(msg);
       }
     }
-    await this.messageHistory.persistMessages({
-      messages: messagesToSave,
-      threadId,
-      resourceId
-    });
-    for (const msg of messagesToSave) {
+    if (filteredMessages.length > 0) {
+      await this.messageHistory.persistMessages({
+        messages: filteredMessages,
+        threadId,
+        resourceId
+      });
+    }
+    for (const msg of filteredMessages) {
       if (this.findLastCompletedObservationBoundary(msg) !== -1) {
         sealedIds.add(msg.id);
       }
@@ -3810,19 +3816,17 @@ ${result.observations}` : result.observations;
       const newMessageIds = messagesToObserve.map((m) => m.id);
       const existingIds = freshRecord?.observedMessageIds ?? record.observedMessageIds ?? [];
       const allObservedIds = [.../* @__PURE__ */ new Set([...Array.isArray(existingIds) ? existingIds : [], ...newMessageIds])];
-      if (result.suggestedContinuation || result.currentTask) {
-        const thread = await this.storage.getThreadById({ threadId });
-        if (thread) {
-          const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
-            suggestedResponse: result.suggestedContinuation,
-            currentTask: result.currentTask
-          });
-          await this.storage.updateThread({
-            id: threadId,
-            title: thread.title ?? "",
-            metadata: newMetadata
-          });
-        }
+      const thread = await this.storage.getThreadById({ threadId });
+      if (thread) {
+        const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
+          suggestedResponse: result.suggestedContinuation,
+          currentTask: result.currentTask
+        });
+        await this.storage.updateThread({
+          id: threadId,
+          title: thread.title ?? "",
+          metadata: newMetadata
+        });
       }
       await this.storage.updateActiveObservations({
         id: record.id,
@@ -4058,7 +4062,7 @@ ${result.observations}` : result.observations;
       messagesToBuffer,
       void 0,
       // No abort signal for background ops
-      { requestContext }
+      { skipContinuationHints: true, requestContext }
     );
     if (!result.observations) {
       omDebug(`[OM:doAsyncBufferedObservation] empty observations returned, skipping buffer storage`);
@@ -4676,8 +4680,8 @@ ${unreflectedContent}` : freshRecord.bufferedReflection;
         if (thread) {
           const newMetadata = memory.setThreadOMMetadata(thread.metadata, {
             lastObservedAt: threadLastObservedAt.toISOString(),
-            ...result.suggestedContinuation && { suggestedResponse: result.suggestedContinuation },
-            ...result.currentTask && { currentTask: result.currentTask }
+            suggestedResponse: result.suggestedContinuation,
+            currentTask: result.currentTask
           });
           await this.storage.updateThread({
             id: threadId,
@@ -5128,5 +5132,5 @@ exports.formatMessagesForObserver = formatMessagesForObserver;
 exports.hasCurrentTaskSection = hasCurrentTaskSection;
 exports.optimizeObservationsForContext = optimizeObservationsForContext;
 exports.parseObserverOutput = parseObserverOutput;
-//# sourceMappingURL=chunk-PVFLHAZX.cjs.map
-//# sourceMappingURL=chunk-PVFLHAZX.cjs.map
+//# sourceMappingURL=chunk-5UYAHJVJ.cjs.map
+//# sourceMappingURL=chunk-5UYAHJVJ.cjs.map