Agentic Context Manager Workbench ACM 5-Part Engine

⚙️ ACM Pipeline Rules
Context Window Capacity Bar 412 / 500 Tokens Used
Active Memory
Compressed Context
Evicted / Dropped
Initial Tokens
792
Compressed Tokens
412
Token Reduction
48.0%
Retrieval Accuracy
0.92
🔍 Context Memory Probing (Fuse.js)
📜 Execution Trace Turns
📋 Active Context Management Policy Spec 

      

    

  


  

Greedy context allocation changes counts, not message text
Read the explanation
The healthcare fixture assigns token counts twenty-four, three hundred forty, one hundred twenty, two hundred eighty and twenty-eight, totaling seven hundred ninety-two. Admission combines turn index divided by five with a fixed role score, weighted equally by default. Every turn passes threshold point three. Capacity allocation then walks the original array from oldest to newest, not highest score or newest first. Token counts are assigned fixture values rather than measured tokenizer output. A five hundred token budget retains the first three turns at four hundred eighty-four. Turn four cannot fit, even compressed to one hundred twelve, and is evicted. The final user turn cannot fit its full twenty-eight, but its compressed count eleven does fit. Total is four hundred ninety-five, leaving five unused. This structural compression changes only effective token counts; original message text remains intact and searchable, with no actual summary generated. Count reduction is thirty-seven point five percent. Displayed retrieval accuracy is point eight nine because a fixed formula uses retained ratio point six two five. It is not measured by the fuzzy-search probe. That probe searches the full content of active and compressed turns; evicted turns are excluded. The access timing selection is recorded but does not change allocation logic. No real healthcare documents, model context or provider memory is queried. This preview certifies only scoped local preservation and playback, not real retrieval quality.



  


Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.