agenticforge-context

You manage LLM token budgets by configuring maxTokens, systemTokenBudget, and historyTokenBudget to construct context that fits within model limits. This approach tracks totalTokens during assembly and indicates whether history truncation occurred to prevent overflow.

Token-aware context building assembles system instructions, conversation history, and user queries while respecting configurable token limits. It tracks totalTokens during assembly and flags whether truncation occurred to prevent exceeding the model window.

Yes, you can configure separate token budgets for system prompts and conversation history. The context builder accepts configurable maxTokens, systemTokenBudget, and historyTokenBudget parameters, allowing you to tailor context size per model and agent workflow.

You fit long conversation history into model limits by allocating a specific historyTokenBudget during context assembly. The builder tracks totalTokens, truncates history when necessary to prioritize the user query, and indicates when truncation occurred.

Yes, the context builder supports optional augmentation with retrieved or computed context similar to RAG. It integrates external context into assembled messages while strictly respecting configured token budgets to improve relevance without overflow.

Your agent workflow loses earlier messages because token-aware context assembly applies history truncation when conversation exceeds the historyTokenBudget. The builder indicates whether truncation occurred to maintain compliance with the model's maxTokens limit.