sage-context-management

To manage LLM context limits, you can estimate tokens, prune older non-critical messages, and auto-compact history. This preserves recent context and system prompts while keeping extended coding sessions concise within model limits.

Token estimation accurately calculates tokens for messages, including system, user, and assistant roles. This provider-aware calculation helps deterministically control context size before hitting the model's hard limits.

Auto-compaction summarizes long histories to fit within context budgets using optional LLM-assisted summaries. It preserves important context information while dropping older non-system messages to maintain dialogue conciseness.

Yes, pruning drops older non-system messages while retaining recent context and system prompts. This ensures critical information is preserved during extended planning tasks or code review workflows.

You need context pruning when messages approach the model's context limit during long-running assistant sessions. It helps control context size deterministically by dropping older non-critical messages while preserving important information.

Yes, you invoke the ContextManager with provider and model configuration. This enables accurate token estimation and auto-compaction tailored to the specific context limits of your chosen LLM provider.