Skip to content

ContextWindowStrategy

This docs was updated at: 2026-02-23

com.paragon.agents.context.ContextWindowStrategy  ยท  Interface

Strategy interface for managing conversation context length.

Context window strategies are used by Agent to ensure conversation history stays within the model's token limits. When the context exceeds the maximum token count, the strategy determines how to reduce it.

Common strategies include:

  • SlidingWindowStrategy: Removes oldest messages to fit within limit
  • SummarizationStrategy: Summarizes older messages using an LLM

Usage Example

// Create agent with sliding window strategy
Agent agent = Agent.builder()
    .name("Assistant")
    .model("openai/gpt-4o")
    .instructions("You are a helpful assistant.")
    .responder(responder)
    .contextWindow(new SlidingWindowStrategy(), 4000)
    .build();
// Or use summarization
Agent agent = Agent.builder()
    .name("Assistant")
    .model("openai/gpt-4o")
    .instructions("You are a helpful assistant.")
    .responder(responder)
    .contextWindow(SummarizationStrategy.withResponder(responder, "openai/gpt-4o-mini"), 4000)
    .build();

See Also

  • SlidingWindowStrategy
  • SummarizationStrategy
  • TokenCounter

Since: 1.0

Methods

manage

List<ResponseInputItem> manage(
      @NonNull List<ResponseInputItem> history, int maxTokens, @NonNull TokenCounter counter)

Manages the conversation history to fit within the token limit.

This method is called before each LLM request when context management is enabled. Implementations should return a modified list that fits within maxTokens.

Parameters

Name Description
history the current conversation history (may be modified or replaced)
maxTokens the maximum number of tokens allowed
counter the token counter to use for measuring content

Returns

the managed history that fits within the token limit