feat(core): Add byte size limit and oldest first truncation for gen_ai messages

packages/core/src/utils/ai/messageTruncation.ts

@@ -0,0 +1,124 @@
+export const DEFAULT_GEN_AI_MESSAGES_BYTE_LIMIT = 20000;
+
+/**
+ * Calculates the UTF-8 byte size of a string.
+ */
+export function getByteSize(str: string): number {
+  return new TextEncoder().encode(str).length;
+}
+
+/**
+ * Truncates a string to fit within maxBytes using binary search.
+ */
+function truncateStringByBytes(str: string, maxBytes: number): string {
+  if (getByteSize(str) <= maxBytes) {
+    return str;
+  }
+
+  // Binary search for the longest substring that fits
+  let left = 0;
+  let right = str.length;
+  let result = '';
+
+  while (left <= right) {
+    const mid = Math.floor((left + right) / 2);
+    const candidate = str.slice(0, mid);
+    const candidateSize = getByteSize(candidate);
+
+    if (candidateSize <= maxBytes) {
+      result = candidate;
+      left = mid + 1;
+    } else {
+      right = mid - 1;
+    }
+  }
+
+  return result;
+}
+
+/**
+ * Truncates messages array using binary search to find optimal starting point.
+ * Removes oldest messages first until the array fits within maxBytes
+ * It also tries to truncate the latest message's content if it's too large.
+ *
+ */
+export function truncateMessagesByBytes(messages: unknown[], maxBytes: number): unknown[] {
+  if (!Array.isArray(messages) || messages.length === 0) {
+    return messages;
+  }
+
+  const fullSize = getByteSize(JSON.stringify(messages));
+
+  if (fullSize <= maxBytes) {
+    return messages;
+  }
+
+  // Binary search for the minimum startIndex where remaining messages fit (works for single or multiple messages)
+  let left = 0;
+  let right = messages.length - 1;
+  let bestStartIndex = messages.length;
+
+  while (left <= right) {
+    const mid = Math.floor((left + right) / 2);
+    const remainingMessages = messages.slice(mid);
+    const remainingSize = getByteSize(JSON.stringify(remainingMessages));
+
+    if (remainingSize <= maxBytes) {
+      bestStartIndex = mid;
+      right = mid - 1; // Try to keep more messages
+    } else {
+      // If we're down to a single message and it doesn't fit, break and handle content truncation
+      if (remainingMessages.length === 1) {
+        bestStartIndex = mid; // Use this single message
+        break;
+      }
+      left = mid + 1; // Need to remove more messages
+    }
+  }
+
+  const remainingMessages = messages.slice(bestStartIndex);
+
+  // SPECIAL CASE: Single message handling (either started with 1, or reduced to 1 after binary search)
+  if (remainingMessages.length === 1) {
+    const singleMessage = remainingMessages[0];
+    const singleMessageSize = getByteSize(JSON.stringify(singleMessage));
+
+    // If single message fits, return it
+    if (singleMessageSize <= maxBytes) {
+      return remainingMessages;
+    }
+
+    // Single message is too large, try to truncate its content
+    if (
+      typeof singleMessage === 'object' &&
+      singleMessage !== null &&
+      'content' in singleMessage &&
+      typeof (singleMessage as { content: unknown }).content === 'string'
+    ) {
+      const originalContent = (singleMessage as { content: string }).content;
+      const messageWithoutContent = { ...singleMessage, content: '' };
+      const otherMessagePartsSize = getByteSize(JSON.stringify(messageWithoutContent));
+      const availableContentBytes = maxBytes - otherMessagePartsSize;
+
+      if (availableContentBytes <= 0) {
+        return [];
+      }
+
+      const truncatedContent = truncateStringByBytes(originalContent, availableContentBytes);
+      return [{ ...singleMessage, content: truncatedContent }];
+    } else {
+      return [];
+    }
+  }
+
+  // Multiple messages remain and fit within limit
+  return remainingMessages;
+}
+
+/**
+ * Truncates gen_ai messages to fit within the default byte limit.
+ * This is a convenience wrapper around truncateMessagesByBytes.
+ */
+export function truncateGenAiMessages(messages: unknown[]): unknown[] {
+  return truncateMessagesByBytes(messages, DEFAULT_GEN_AI_MESSAGES_BYTE_LIMIT);
+}

packages/core/src/utils/anthropic-ai/index.ts

@@ -23,6 +23,7 @@ import {
   GEN_AI_RESPONSE_TOOL_CALLS_ATTRIBUTE,
   GEN_AI_SYSTEM_ATTRIBUTE,
 } from '../ai/gen-ai-attributes';
+import { truncateGenAiMessages } from '../ai/messageTruncation';
 import { buildMethodPath, getFinalOperationName, getSpanOperation, setTokenUsageAttributes } from '../ai/utils';
 import { handleCallbackErrors } from '../handleCallbackErrors';
 import { instrumentAsyncIterableStream, instrumentMessageStream } from './streaming';
@@ -71,16 +72,24 @@ function extractRequestAttributes(args: unknown[], methodPath: string): Record<s
   return attributes;
 }
 
-/**
- * Add private request attributes to spans.
- * This is only recorded if recordInputs is true.
- */
 function addPrivateRequestAttributes(span: Span, params: Record<string, unknown>): void {
   if ('messages' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.messages) });
+    const messages = params.messages;
+    if (Array.isArray(messages)) {
+      const truncatedMessages = truncateGenAiMessages(messages);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedMessages) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(messages) });
+    }
   }
   if ('input' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.input) });
+    const input = params.input;
+    if (Array.isArray(input)) {
+      const truncatedInput = truncateGenAiMessages(input);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedInput) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(input) });
+    }
   }
   if ('prompt' in params) {
     span.setAttributes({ [GEN_AI_PROMPT_ATTRIBUTE]: JSON.stringify(params.prompt) });

packages/core/src/utils/google-genai/index.ts

@@ -22,6 +22,7 @@ import {
   GEN_AI_USAGE_OUTPUT_TOKENS_ATTRIBUTE,
   GEN_AI_USAGE_TOTAL_TOKENS_ATTRIBUTE,
 } from '../ai/gen-ai-attributes';
+import { truncateGenAiMessages } from '../ai/messageTruncation';
 import { buildMethodPath, getFinalOperationName, getSpanOperation } from '../ai/utils';
 import { handleCallbackErrors } from '../handleCallbackErrors';
 import { CHAT_PATH, CHATS_CREATE_METHOD, GOOGLE_GENAI_SYSTEM_NAME } from './constants';
@@ -127,26 +128,41 @@ function extractRequestAttributes(
 
   return attributes;
 }
-
 /**
  * Add private request attributes to spans.
  * This is only recorded if recordInputs is true.
  * Handles different parameter formats for different Google GenAI methods.
  */
 function addPrivateRequestAttributes(span: Span, params: Record<string, unknown>): void {
-  // For models.generateContent: ContentListUnion: Content | Content[] | PartUnion | PartUnion[]
   if ('contents' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.contents) });
+    const contents = params.contents;
+    // For models.generateContent: ContentListUnion: Content | Content[] | PartUnion | PartUnion[]
+    if (Array.isArray(contents)) {
+      const truncatedContents = truncateGenAiMessages(contents);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedContents) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(contents) });
+    }
   }
 
-  // For chat.sendMessage: message can be string or Part[]
   if ('message' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.message) });
+    const message = params.message;
+    if (Array.isArray(message)) {
+      const truncatedMessage = truncateGenAiMessages(message);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedMessage) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(message) });
+    }
   }
 
-  // For chats.create: history contains the conversation history
   if ('history' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.history) });
+    const history = params.history;
+    if (Array.isArray(history)) {
+      const truncatedHistory = truncateGenAiMessages(history);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedHistory) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(history) });
+    }
   }
 }
 

packages/core/src/utils/openai/index.ts

@@ -19,6 +19,7 @@ import {
   GEN_AI_RESPONSE_TOOL_CALLS_ATTRIBUTE,
   GEN_AI_SYSTEM_ATTRIBUTE,
 } from '../ai/gen-ai-attributes';
+import { truncateGenAiMessages } from '../ai/messageTruncation';
 import { OPENAI_INTEGRATION_NAME } from './constants';
 import { instrumentStream } from './streaming';
 import type {
@@ -188,13 +189,24 @@ function addResponseAttributes(span: Span, result: unknown, recordOutputs?: bool
   }
 }
 
-// Extract and record AI request inputs, if present. This is intentionally separate from response attributes.
 function addRequestAttributes(span: Span, params: Record<string, unknown>): void {
   if ('messages' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.messages) });
+    const messages = params.messages;
+    if (Array.isArray(messages)) {
+      const truncatedMessages = truncateGenAiMessages(messages);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedMessages) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(messages) });
+    }
   }
   if ('input' in params) {
-    span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(params.input) });
+    const input = params.input;
+    if (Array.isArray(input)) {
+      const truncatedInput = truncateGenAiMessages(input);
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(truncatedInput) });
+    } else {
+      span.setAttributes({ [GEN_AI_REQUEST_MESSAGES_ATTRIBUTE]: JSON.stringify(input) });
+    }
   }
 }
 

packages/core/src/utils/vercel-ai/index.ts

@@ -3,6 +3,7 @@ import { SEMANTIC_ATTRIBUTE_SENTRY_OP, SEMANTIC_ATTRIBUTE_SENTRY_ORIGIN } from '
 import type { Event } from '../../types-hoist/event';
 import type { Span, SpanAttributes, SpanAttributeValue, SpanJSON, SpanOrigin } from '../../types-hoist/span';
 import { spanToJSON } from '../spanUtils';
+import { truncateGenAiMessages } from '../ai/messageTruncation';
 import { toolCallSpanMap } from './constants';
 import type { TokenSummary } from './types';
 import { accumulateTokensForParent, applyAccumulatedTokens } from './utils';
@@ -190,7 +191,13 @@ function processGenerateSpan(span: Span, name: string, attributes: SpanAttribute
   }
 
   if (attributes[AI_PROMPT_ATTRIBUTE]) {
-    span.setAttribute('gen_ai.prompt', attributes[AI_PROMPT_ATTRIBUTE]);
+    const prompt = attributes[AI_PROMPT_ATTRIBUTE];
+    if (Array.isArray(prompt)) {
+      const truncatedPrompt = truncateGenAiMessages(prompt);
+      span.setAttribute('gen_ai.prompt', JSON.stringify(truncatedPrompt));
+    } else {
+      span.setAttribute('gen_ai.prompt', prompt);
+    }
   }
   if (attributes[AI_MODEL_ID_ATTRIBUTE] && !attributes[GEN_AI_RESPONSE_MODEL_ATTRIBUTE]) {
     span.setAttribute(GEN_AI_RESPONSE_MODEL_ATTRIBUTE, attributes[AI_MODEL_ID_ATTRIBUTE]);