@mastra/memory 1.18.1 → 1.18.2-alpha.0

package/CHANGELOG.md

@@ -1,5 +1,42 @@
 # @mastra/memory
 
+## 1.18.2-alpha.0
+
+### Patch Changes
+
+- Fixed Observational Memory missing the observation threshold for messages with large file parts. Previously `TokenCounter`'s local/sync counting path only stringified the file descriptor (`type`, `mimeType`, `filename`) for non-image files, so a 100KB PDF looked like ~8 tokens to OM and the conversation kept replaying the full unobserved history past every reasonable threshold. ([#16562](https://github.com/mastra-ai/mastra/pull/16562))
+
+  `TokenCounter` now estimates non-image file part tokens locally from the attachment's byte size and mime type using a per-provider heuristic, mirroring the existing local image-token estimator:
+  - Anthropic PDFs ≈ `bytes / 3` (floor 1500)
+  - Google PDFs ≈ `bytes / 20` (floor 258)
+  - OpenAI / unknown PDFs ≈ `bytes / 4` (floor 500)
+  - Text-ish mime types (`text/*`, JSON, XML, YAML) ≈ `bytes / 4`
+  - Other binary ≈ `bytes / 4`
+
+  URL-only file parts (no body to size) keep the previous descriptor-only local estimate. `countMessagesAsync()` continues to prefer provider token-count endpoints for supported providers; this change only improves the local fallback used when no provider endpoint is available.
+
+  ```ts
+  // Before: this PDF counted as ~8 tokens locally regardless of size, so OM never triggered.
+  const part = {
+    type: 'file',
+    data: largePdfBase64,
+    mimeType: 'application/pdf',
+    filename: 'report.pdf',
+  };
+  // counter.countMessage(message) ≈ 8
+
+  // After: estimated locally from byte size on the active provider.
+  // counter.countMessage(message) ≈ tens of thousands of tokens
+  //   → OM threshold trips as expected.
+  ```
+
+  The internal token-estimate cache version was bumped, which invalidates persisted estimates from older `@mastra/memory` releases on the next read; entries are recomputed automatically.
+
+  Fixes https://github.com/mastra-ai/mastra/issues/16522
+
+- Updated dependencies [[`bad08e9`](https://github.com/mastra-ai/mastra/commit/bad08e99c5291884c3ac76743c78c74f53a302c2)]:
+  - @mastra/core@1.35.0-alpha.1
+
 ## 1.18.1
 
 ### Patch Changes

package/dist/{chunk-LCRYFBV3.cjs → chunk-4AQHFADP.cjs}

@@ -5539,7 +5539,7 @@ var IMAGE_FILE_EXTENSIONS = /* @__PURE__ */ new Set([
   "heif",
   "avif"
 ]);
-var TOKEN_ESTIMATE_CACHE_VERSION = 6;
+var TOKEN_ESTIMATE_CACHE_VERSION = 7;
 var CLIENT_TOKEN_ESTIMATE_SOURCE = "client";
 var DEFAULT_IMAGE_ESTIMATOR = {
   baseTokens: 85,
@@ -6050,6 +6050,42 @@ function estimateAnthropicImageTokens(dimensions, sourceStats) {
   }
   return 1600;
 }
+function estimateFileTokensFromBytes(provider, mimeType, sizeBytes) {
+  const normalizedMime = (mimeType ?? "").toLowerCase().split(";", 1)[0].trim();
+  const isPdf = normalizedMime === "application/pdf";
+  const isTextish = normalizedMime.startsWith("text/") || ["application/json", "application/xml", "application/x-yaml", "application/yaml"].includes(normalizedMime);
+  if (isPdf) {
+    if (provider === "google") return Math.max(258, Math.ceil(sizeBytes / 20));
+    if (provider === "anthropic") return Math.max(1500, Math.ceil(sizeBytes / 3));
+    return Math.max(500, Math.ceil(sizeBytes / 4));
+  }
+  if (isTextish) return Math.max(1, Math.ceil(sizeBytes / 4));
+  return Math.max(1, Math.ceil(sizeBytes / 4));
+}
+function estimateNonImageFileTokens(modelContext, part) {
+  const sourceStats = resolveImageSourceStats(getObjectValue(part, "data"));
+  if (sourceStats.sizeBytes === void 0) {
+    return void 0;
+  }
+  const provider = resolveProviderId(modelContext);
+  const modelId = modelContext?.modelId ?? null;
+  const mimeType = getAttachmentMimeType(part, "application/octet-stream");
+  const filename = getAttachmentFilename(part) ?? null;
+  const tokens = estimateFileTokensFromBytes(provider, mimeType, sourceStats.sizeBytes);
+  return {
+    tokens,
+    cachePayload: JSON.stringify({
+      kind: "non-image-file",
+      provider: provider ?? "fallback",
+      modelId,
+      estimator: "bytes",
+      source: sourceStats.source,
+      sizeBytes: sourceStats.sizeBytes,
+      mimeType,
+      filename
+    })
+  };
+}
 function estimateGoogleImageTokens(modelContext, part, dimensions) {
   if (isGoogleGemini3Model(modelContext)) {
     const mediaResolution = resolveGoogleMediaResolution(part);
@@ -6477,6 +6513,15 @@ var TokenCounter = class _TokenCounter {
       return this.readOrPersistFixedPartEstimate(part, "image-like-file", estimate.cachePayload, estimate.tokens);
     }
     if (part.type === "file") {
+      const byteEstimate = estimateNonImageFileTokens(this.getModelContext(), part);
+      if (byteEstimate) {
+        return this.readOrPersistFixedPartEstimate(
+          part,
+          "non-image-file",
+          byteEstimate.cachePayload,
+          byteEstimate.tokens
+        );
+      }
       return this.readOrPersistPartEstimate(part, "file-descriptor", serializeNonImageFilePartForTokenCounting(part));
     }
     return void 0;
@@ -9692,5 +9737,5 @@ exports.stripEphemeralAnchorIds = stripEphemeralAnchorIds;
 exports.stripObservationGroups = stripObservationGroups;
 exports.truncateStringByTokens = truncateStringByTokens;
 exports.wrapInObservationGroup = wrapInObservationGroup;
-//# sourceMappingURL=chunk-LCRYFBV3.cjs.map
-//# sourceMappingURL=chunk-LCRYFBV3.cjs.map
+//# sourceMappingURL=chunk-4AQHFADP.cjs.map
+//# sourceMappingURL=chunk-4AQHFADP.cjs.map