📁 Subsystem Deep Dive

services/compact/

📁 services/compact/ 📄 7 files

Tutorial: compact

The compact project serves as an intelligent memory management system for long-running AI sessions. Acting like an automated garbage collector, it monitors the context window and prevents "out of memory" errors by using Auto-Compact to trigger cleanup. The system employs Conversation Summarization to "zip" old message history into concise summaries and Micro-Compaction to prune unnecessary data, while ensuring critical information is preserved via Context Rehydration so the AI never loses track of its active files or tasks.

flowchart TD A0["Conversation Summarization (Compaction)"] A1["Automated Context Management (Auto-Compact)"] A2["Session Memory Optimization"] A3["Micro-Compaction & Pruning"] A4["Context Rehydration & Cleanup"] A5["Message Grouping & Boundaries"] A1 -->|"Triggers"| A0 A1 -->|"Attempts optimization first"| A2 A0 -->|"Restores state using"| A4 A0 -->|"Uses for safe slicing"| A5 A0 -->|"Uses token estimation from"| A3 A2 -->|"Uses formatting utils from"| A0 A4 -->|"Resets state of"| A3