@mastra/memory 1.15.0-alpha.1 → 1.15.0-alpha.3

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # @mastra/memory
 
+## 1.15.0-alpha.3
+
+### Patch Changes
+
+- Fixed reflection threshold not respecting per-record overrides set via the PATCH API. Previously, lowering the reflection threshold for a specific record had no effect on the actual reflection trigger — only the default 40k threshold was used. Now per-record overrides are correctly applied in both sync and async reflection paths. ([#15170](https://github.com/mastra-ai/mastra/pull/15170))
+
+## 1.15.0-alpha.2
+
+### Patch Changes
+
+- Fixed message history doubling when using Observational Memory with the Mastra gateway. The local ObservationalMemoryProcessor now detects when the agent's model is routed through the Mastra gateway and skips its input/output processing, since the gateway handles OM server-side. ([#15161](https://github.com/mastra-ai/mastra/pull/15161))
+
 ## 1.15.0-alpha.1
 
 ### Patch Changes

package/dist/{chunk-ZEKCVX4E.js → chunk-42AZEBIK.js}

@@ -1,6 +1,6 @@
 import { OBSERVATIONAL_MEMORY_DEFAULTS, OBSERVATION_CONTEXT_PROMPT, OBSERVATION_CONTEXT_INSTRUCTIONS, OBSERVATION_RETRIEVAL_INSTRUCTIONS, OBSERVATION_CONTINUATION_HINT } from './chunk-LSJJAJAF.js';
 import { coreFeatures } from '@mastra/core/features';
-import { resolveModelConfig } from '@mastra/core/llm';
+import { resolveModelConfig, ModelRouterLanguageModel } from '@mastra/core/llm';
 import { getThreadOMMetadata, setThreadOMMetadata, parseMemoryRequestContext } from '@mastra/core/memory';
 import { MessageHistory } from '@mastra/core/processors';
 import xxhash from 'xxhash-wasm';
@@ -4124,13 +4124,29 @@ var ReflectorRunner = class {
       model
     });
   }
-  getObservationMarkerConfig() {
+  getObservationMarkerConfig(record) {
     return {
       messageTokens: getMaxThreshold(this.observationConfig.messageTokens),
-      observationTokens: getMaxThreshold(this.reflectionConfig.observationTokens),
+      observationTokens: getMaxThreshold(
+        record ? this.getEffectiveReflectionTokens(record) : this.reflectionConfig.observationTokens
+      ),
       scope: this.scope
     };
   }
+  /**
+   * Resolve the effective reflection observationTokens for a record.
+   * Only explicit per-record overrides (stored under `_overrides`) win;
+   * the initial config snapshot is ignored so instance-level changes
+   * still take effect for existing records.
+   */
+  getEffectiveReflectionTokens(record) {
+    const overrides = record.config?._overrides;
+    const recordTokens = overrides?.reflection?.observationTokens;
+    if (recordTokens) {
+      return recordTokens;
+    }
+    return this.reflectionConfig.observationTokens;
+  }
   /**
    * Call the Reflector agent with escalating compression levels.
    */
@@ -4319,7 +4335,7 @@ var ReflectorRunner = class {
     const freshRecord = await this.storage.getObservationalMemory(record.threadId, record.resourceId);
     const currentRecord = freshRecord ?? record;
     const observationTokens = currentRecord.observationTokenCount ?? 0;
-    const reflectThreshold = getMaxThreshold(this.reflectionConfig.observationTokens);
+    const reflectThreshold = getMaxThreshold(this.getEffectiveReflectionTokens(currentRecord));
     const bufferActivation = this.reflectionConfig.bufferActivation ?? 0.5;
     const startedAt = (/* @__PURE__ */ new Date()).toISOString();
     const cycleId = `reflect-buf-${Date.now()}-${Math.random().toString(36).slice(2, 11)}`;
@@ -4348,7 +4364,7 @@ var ReflectorRunner = class {
         recordId: record.id,
         threadId: record.threadId ?? "",
         threadIds: record.threadId ? [record.threadId] : [],
-        config: this.getObservationMarkerConfig()
+        config: this.getObservationMarkerConfig(currentRecord)
       });
       void writer.custom(startMarker).catch(() => {
       });
@@ -4460,7 +4476,7 @@ ${unreflectedContent}` : freshRecord.bufferedReflection;
         threadId: freshRecord.threadId ?? "",
         generationCount: afterRecord?.generationCount ?? freshRecord.generationCount ?? 0,
         observations: afterRecord?.activeObservations,
-        config: this.getObservationMarkerConfig()
+        config: this.getObservationMarkerConfig(freshRecord)
       });
       void writer.custom(activationMarker).catch(() => {
       });
@@ -4491,7 +4507,7 @@ ${unreflectedContent}` : freshRecord.bufferedReflection;
       observabilityContext
     } = opts;
     const lockKey = this.buffering.getLockKey(record.threadId, record.resourceId);
-    const reflectThreshold = getMaxThreshold(this.reflectionConfig.observationTokens);
+    const reflectThreshold = getMaxThreshold(this.getEffectiveReflectionTokens(record));
     if (this.buffering.isAsyncReflectionEnabled() && observationTokens < reflectThreshold) {
       const shouldTrigger = (() => {
         if (!this.buffering.isAsyncReflectionEnabled()) return false;
@@ -4570,7 +4586,7 @@ ${unreflectedContent}` : freshRecord.bufferedReflection;
         recordId: record.id,
         threadId,
         threadIds: [threadId],
-        config: this.getObservationMarkerConfig()
+        config: this.getObservationMarkerConfig(record)
       });
       await writer.custom(startMarker).catch(() => {
       });
@@ -8606,6 +8622,10 @@ function getOmObservabilityContext(args) {
     metrics: args.metrics
   };
 }
+var GATEWAY_STATE_KEY = "__isGatewayModel";
+function isMastraGatewayModel(model) {
+  return model instanceof ModelRouterLanguageModel && model.gatewayId === "mastra";
+}
 var ObservationalMemoryProcessor = class {
   id = "observational-memory";
   name = "Observational Memory";
@@ -8641,6 +8661,11 @@ var ObservationalMemoryProcessor = class {
       omDebug(`[OM:processInputStep:NO-CONTEXT] getThreadContext returned null \u2014 returning early`);
       return messageList;
     }
+    if (isMastraGatewayModel(model)) {
+      state[GATEWAY_STATE_KEY] = true;
+      omDebug(`[OM:processInputStep:GATEWAY] gateway handles OM \u2014 skipping local processing`);
+      return messageList;
+    }
     const { threadId, resourceId } = context;
     const memoryContext = parseMemoryRequestContext(requestContext);
     const readOnly = memoryContext?.memoryConfig?.readOnly;
@@ -8757,6 +8782,7 @@ var ObservationalMemoryProcessor = class {
     const state = _state ?? {};
     const context = this.engine.getThreadContext(requestContext, messageList);
     if (!context) return messageList;
+    if (state[GATEWAY_STATE_KEY]) return messageList;
     const observabilityContext = getOmObservabilityContext(args);
     state.__omObservabilityContext = observabilityContext;
     return this.engine.getTokenCounter().runWithModelContext(state.__omActorModelContext, async () => {
@@ -8808,5 +8834,5 @@ function getObservationsAsOf(activeObservations, asOf) {
 }
 
 export { ModelByInputTokens, OBSERVER_SYSTEM_PROMPT, ObservationalMemory, ObservationalMemoryProcessor, TokenCounter, buildObserverPrompt, buildObserverSystemPrompt, combineObservationGroupRanges, deriveObservationGroupProvenance, extractCurrentTask, formatMessagesForObserver, formatToolResultForObserver, getObservationsAsOf, hasCurrentTaskSection, injectAnchorIds, optimizeObservationsForContext, parseAnchorId, parseObservationGroups, parseObserverOutput, reconcileObservationGroupsFromReflection, renderObservationGroupsForReflection, resolveToolResultValue, stripEphemeralAnchorIds, stripObservationGroups, truncateStringByTokens, wrapInObservationGroup };
-//# sourceMappingURL=chunk-ZEKCVX4E.js.map
-//# sourceMappingURL=chunk-ZEKCVX4E.js.map
+//# sourceMappingURL=chunk-42AZEBIK.js.map
+//# sourceMappingURL=chunk-42AZEBIK.js.map