Package com.tnsai.context.compaction

package com.tnsai.context.compaction

Context compaction for long-running agent sessions.

This package provides mechanisms to automatically manage context size by summarizing older messages when approaching token limits.

Compaction Strategies

• TruncatingCompactor (Phase 1) — truncates large tool-call arguments and tool-result content in older messages. Fast, no LLM call.
• LLMContextCompactor (Phase 2) — uses an LLM to summarize older messages into a concise summary.
• TwoPhaseCompactor — combines both: truncation first, then LLM summarization if still over threshold.

Quick Start

 CompactionConfig config = CompactionConfig.builder()
     .thresholdRatio(0.85)
     .preserveLastN(5)
     .build();

 // Two-phase compaction (recommended)
 TwoPhaseCompactor compactor = new TwoPhaseCompactor(
     new TruncatingCompactor(200),
     new LLMContextCompactor(llm)
 );
 CompactionResult result = compactor.compact(messages, config);
 

Since:

0.2.3

Author:

TnsAI Team

Class	Description
CompactionConfig	Configuration for context compaction.
CompactionConfig.Builder	Builder for CompactionConfig.
CompactionResult	Result of a context compaction operation.
ContextCompactor	Interface for context compaction strategies.
LLMContextCompactor	LLM-based context compactor that summarizes old messages.
TruncatingCompactor	Phase-1 compactor that truncates large tool-call arguments in older messages.
TwoPhaseCompactor	Two-phase context compactor combining truncation and LLM summarization.