Nabu Nisaba
Enhances LLM agent capabilities with semantic code search, structural code views, and an intelligent workspace TUI for efficient context management.
概要
Nabu Nisaba is a research prototype that provides LLM agents, specifically designed for Claude Code, with advanced code intelligence and a dynamic workspace environment. It combines `nabu`, a codebase search engine offering semantic and full-text search, structural code views, and outline-like features across multiple programming languages, with `nisaba`, an MCP server that emulates a workspace TUI (Terminal User Interface). This innovative approach enables agents to manage token usage efficiently, navigate codebases with enhanced architectural understanding, and dynamically incorporate 'augments' – self-managed cognitive resources for knowledge compilation and procedural memory, all while maintaining a streamlined interaction flow through system prompt manipulation rather than verbose message exchanges.
主な機能
- Provides semantic and full-text search (FTS) for codebases across Python, Java, C++, and Perl.
- Exposes structural code views and semantic relationship mapping (packages, classes, methods) to LLM agents.
- Emulates a workspace Terminal User Interface (TUI) for agents to manage context and workspace organization.
- 1 GitHub stars
- Utilizes a man-in-the-middle proxy to dynamically inject and update system prompts with workspace state.
- Implements dynamic 'augments' for LLM agents to autonomously manage knowledge, skills, and procedural memory.
ユースケース
- Improving LLM agent's understanding of complex software architectures during development.
- Reducing token usage and increasing context stability for LLM agents in long-running development sessions.
- Enabling LLM agents to manage their own context and workspace organization for efficient operations.
- Researching and developing advanced cognitive patterns for LLM agents in code-related tasks.
