rune

Rune is a versatile, security-focused AI agent that emphasizes a zero-trust architecture. It utilizes five isolation layers—including cgroups v2, network namespaces, Seccomp BPF, Landlock filesystem restrictions, and DNS allowlisting—to safely execute tools. The project is distributed as a single binary, making it highly portable for various environments. Key capabilities include: - **AI Agent Modes**: Operates as an interactive CLI tool, a non-interactive pipe-mode utility, or a Concourse CI resource type. - **Rune Notes Hub**: A self-contained server providing a Web UI, REST API, and MCP endpoint. It treats local Markdown files as the source of truth, supporting KaTeX math, Mermaid diagrams, and SVG rendering. - **Tooling & Skills**: Features built-in tools for file manipulation, command execution, and web fetching, with a modular 'skills' system for context-specific abilities. - **Integration**: Includes a native Vim/Neovim plugin for ghost-text completion and AI-assisted editing, as well as a browser extension (Rune Chat) for contextual interaction. - **MCP Support**: Acts as both an MCP client (connecting to external servers) and an MCP server (exposing its own capabilities). - **Policy Management**: Offers granular control over AI actions via allowlists, confirmation prompts, or unrestricted modes, ensuring safe operation in sensitive environments. - **Provider Support**: Compatible with GitHub Copilot, OpenRouter, Google Gemini, and other OpenAI-compatible LLM providers.