LLM Context Window Decay & Semantic Retrieval Workbench

Fuse.js Hybrid Benchmark
Direct Context Window Stuffing Attention Drift Unfiltered
450
Total Window Tokens
38.4%
Fact Recall Loss
5
Multi-turn Length
$0.027
Est. Cost / 1k Queries
Context Memory Degradation Curve
Sequential Turn Log Stream
Pinned Local Index (Fuse.js Hybrid) Zero Token Waste
92
Retrieved Context Tokens
100.0%
Recall Precision
79.6%
Token Savings
$0.005
Est. Cost / 1k Queries
Semantic Match Score Comparison
Fuse.js Match Results (Top Relevance)
Context Architecture & Cost Audit Summary

Direct context window stuffing consumes 450 tokens with an estimated recall decay loss of 38.4% across 5 turns. Switching to local hybrid semantic retrieval with Fuse.js isolates the query context down to 92 tokens, yielding a 79.6% token reduction and eliminating memory drift.

Enjoy this tool? Build your own with Super