libmemory

What problem does it solve?

Memory management for building token-budgeted LLM context windows from conversation history and tools, enabling efficient prompts within strict model limits.

Core Features & Use Cases

• WindowBuilder constructs context windows that fit within token budgets by combining history and tools.
• MemoryIndex stores and deduplicates conversation identifiers for fast retrieval and reuse.
• Token budgeting and overhead calculations ensure reliable prompt construction across multiple models.
• Integration with Memory and Agent services to streamline end-to-end chat workflows.
• Use Case: build compact, relevant conversation history for an LLM-driven assistant while respecting model limits.

Quick Start

Create a memory window by supplying the conversation history, available tools, and a token budget, then pass the resulting messages and tools to the LLM.