@librechat/agents 3.1.57 → 3.1.60

package/src/summarization/__tests__/trigger.test.ts

@@ -0,0 +1,50 @@
+import { shouldTriggerSummarization } from '@/summarization';
+
+describe('shouldTriggerSummarization', () => {
+  it('uses pre-prune pressure for token_ratio triggers when messages were pruned', () => {
+    const result = shouldTriggerSummarization({
+      trigger: { type: 'token_ratio', value: 0.8 },
+      maxContextTokens: 2500,
+      prePruneContextTokens: 3200,
+      remainingContextTokens: 1200,
+      messagesToRefineCount: 4,
+    });
+
+    expect(result).toBe(true);
+  });
+
+  it('uses pre-prune remaining tokens for remaining_tokens triggers when available', () => {
+    const result = shouldTriggerSummarization({
+      trigger: { type: 'remaining_tokens', value: 500 },
+      maxContextTokens: 2500,
+      prePruneContextTokens: 2300,
+      remainingContextTokens: 1400,
+      messagesToRefineCount: 2,
+    });
+
+    expect(result).toBe(true);
+  });
+
+  it('falls back to post-prune remaining tokens when pre-prune totals are unavailable', () => {
+    const result = shouldTriggerSummarization({
+      trigger: { type: 'token_ratio', value: 0.6 },
+      maxContextTokens: 2500,
+      remainingContextTokens: 1200,
+      messagesToRefineCount: 2,
+    });
+
+    expect(result).toBe(false);
+  });
+
+  it('does not trigger when there is nothing to refine', () => {
+    const result = shouldTriggerSummarization({
+      trigger: { type: 'token_ratio', value: 0.1 },
+      maxContextTokens: 2500,
+      prePruneContextTokens: 2400,
+      remainingContextTokens: 100,
+      messagesToRefineCount: 0,
+    });
+
+    expect(result).toBe(false);
+  });
+});

package/src/summarization/index.ts

@@ -0,0 +1,102 @@
+import type { SummarizationTrigger } from '@/types';
+
+/**
+ * Determines whether summarization should be triggered based on the configured trigger
+ * and current context state.
+ *
+ * Default behavior (no trigger configured): returns `true` whenever messages were pruned.
+ * This is intentional — when an admin enables summarization without specifying a trigger,
+ * summarization fires on any context overflow that causes pruning.
+ *
+ * When a trigger IS configured but required runtime data is missing (e.g., maxContextTokens
+ * unavailable for a token_ratio trigger), returns `false` — we cannot evaluate the condition,
+ * so we do not fire.
+ */
+export function shouldTriggerSummarization(params: {
+  trigger?: SummarizationTrigger;
+  maxContextTokens?: number;
+  prePruneContextTokens?: number;
+  remainingContextTokens?: number;
+  messagesToRefineCount: number;
+}): boolean {
+  const {
+    trigger,
+    maxContextTokens,
+    prePruneContextTokens,
+    remainingContextTokens,
+    messagesToRefineCount,
+  } = params;
+  if (messagesToRefineCount <= 0) {
+    return false;
+  }
+
+  // No trigger configured: default to always summarize when pruning occurs.
+  if (!trigger || typeof trigger.type !== 'string') {
+    return true;
+  }
+
+  const triggerValue =
+    typeof trigger.value === 'number' && Number.isFinite(trigger.value)
+      ? trigger.value
+      : undefined;
+
+  // Trigger configured but value is invalid: cannot evaluate, do not fire.
+  if (triggerValue == null) {
+    return false;
+  }
+
+  if (trigger.type === 'token_ratio') {
+    const prePruneRemainingContextTokens =
+      maxContextTokens != null &&
+      Number.isFinite(maxContextTokens) &&
+      maxContextTokens > 0 &&
+      prePruneContextTokens != null &&
+      Number.isFinite(prePruneContextTokens)
+        ? maxContextTokens - prePruneContextTokens
+        : undefined;
+    const effectiveRemainingContextTokens =
+      prePruneRemainingContextTokens ?? remainingContextTokens;
+
+    // Required runtime data missing: cannot evaluate token_ratio, do not fire.
+    if (
+      maxContextTokens == null ||
+      !Number.isFinite(maxContextTokens) ||
+      maxContextTokens <= 0 ||
+      effectiveRemainingContextTokens == null ||
+      !Number.isFinite(effectiveRemainingContextTokens)
+    ) {
+      return false;
+    }
+    const usedRatio = 1 - effectiveRemainingContextTokens / maxContextTokens;
+    return usedRatio >= triggerValue;
+  }
+
+  if (trigger.type === 'remaining_tokens') {
+    const prePruneRemainingContextTokens =
+      maxContextTokens != null &&
+      Number.isFinite(maxContextTokens) &&
+      maxContextTokens > 0 &&
+      prePruneContextTokens != null &&
+      Number.isFinite(prePruneContextTokens)
+        ? maxContextTokens - prePruneContextTokens
+        : undefined;
+    const effectiveRemainingContextTokens =
+      prePruneRemainingContextTokens ?? remainingContextTokens;
+
+    // Required runtime data missing: cannot evaluate remaining_tokens, do not fire.
+    if (
+      effectiveRemainingContextTokens == null ||
+      !Number.isFinite(effectiveRemainingContextTokens)
+    ) {
+      return false;
+    }
+    return effectiveRemainingContextTokens <= triggerValue;
+  }
+
+  if (trigger.type === 'messages_to_refine') {
+    return messagesToRefineCount >= triggerValue;
+  }
+
+  // Unrecognized trigger type: cannot evaluate, do not fire.
+  return false;
+}