ratel

I am building a multi-tool AI agent. Using ratel’s approach, design a context-engineering strategy that minimizes prompts and tool descriptions while preserving task accuracy, and explain how to reduce token usage by around 80%.

A lean context plan with prompt structure, tool exposure strategy, and token reduction methods.

My AI agent has too many tools and often chooses the wrong one. Using ratel’s method, redesign the tool-calling layer and provide a concrete plan for dynamically exposing tools, preventing tool overload, and improving selection accuracy.

A tool-governance plan covering tool filtering, layered exposure, and call optimization.

Help me design a memory system for an AI agent without a vector database or embeddings, using in-process BM25 to retrieve skills, past tasks, and document snippets, and explain suitable storage structures and retrieval flow.

A lightweight memory architecture with BM25 retrieval logic, data organization, and usage scenarios.