study8677/antigravity-workspace-template

What it solves

RepoBrain provides a cross-IDE repository knowledge engine that prevents AI coding agents from hallucinating or forgetting project-specific conventions. It replaces the need for agents to manually grep through thousands of lines of code or documentation, which often leads to high hallucination rates and slow response times when asking "how does X work?"

How it works

RepoBrain uses a multi-agent cluster to autonomously analyze a codebase. During a rb-refresh cycle, it groups files by import graphs and directory structure, assigning a dedicated agent to each module to generate a comprehensive Markdown knowledge document. A Map Agent then indexes these documents into a routing map. When a user asks a question via rb-ask, a Router identifies the relevant modules and feeds the corresponding agent documents to a synthesizer that produces an answer grounded in actual source code, including file paths and line numbers.

Who it’s for

Developers using AI-powered IDEs and CLI tools such as Claude Code, Cursor, Windsurf, Codex CLI, VS Code + Copilot, Cline, and Aider who want more accurate and grounded codebase Q&A.

Highlights

• Cross-IDE Compatibility: Uses a shared .repobrain/ knowledge layer that works across different AI tools and IDEs.
• Language Agnostic: Analyzes code using LLMs rather than AST parsing or regex, making it compatible with any programming language.
• ** Robotic Knowledge Generation**: Automatically generates project-level documentation like conventions.md and structure.md.
• MCP Server Support: Can be registered as a Model Context Protocol (MCP) server, allowing agents like Claude Code to call ask_project as a live tool.
• Git Integration: Includes a dedicated GitAgent to analyze git history for insights into why changes were made.