The agentic runtime
for building software.
XR turns your terminal, editor, and workflow into an AI-native development environment. Plan, execute, and ship — with 12,000+ skills, any model, and security built in from the core.
npm i -g @rrrtx/xr && xrTrusted by teams building the future
One runtime. Every surface.
From the terminal you live in to the editor you ship from — XR is a coherent system, not a Frankenstein of plugins.
Shell-native
A real terminal. Real pipes. Real tools. Sub-10ms cold start, with streaming agents that respond like a teammate.
Marketplace
12,000+ verified skills and extensions. Type-safe, versioned, and signed. Compose them in seconds.
Secure by default
Capability-based security, human-in-the-loop confirmations, signed skills, and end-to-end audit.
- Sandboxed execution per skill
- Granular capability tokens
- Signed, verifiable skills
- SOC 2 Type II & HIPAA ready
Built for how developers actually work.
XR is a unified system — not a wrapper around a chatbot. Every part is designed to be composed, extended, and trusted.
Agentic Runtime
A performant, sandboxed runtime that plans, executes, and iterates — with deterministic replays and full audit trails.
Native Shell
XR lives inside your terminal. Native commands, pipes, and a sub-10ms cold start. No browser required.
Editor Agnostic
First-class integrations with VS Code, Neovim, JetBrains, Zed, and Cursor. Your agent, every surface.
Skill Graph
Compose 12,000+ verified skills into pipelines. Each skill is typed, versioned, and permission-scoped.
Any Model
Open-weight to frontier — GPT-5, Claude Opus, Gemini, DeepSeek, Qwen, Llama, Groq. Route by cost, latency, or task.
Security by Default
Capability-based security, human-in-the-loop confirmations, signed skills, and end-to-end audit logging.
Reproducible Workflows
Every run is a replayable trace. Share sessions, diff executions, and roll agents back like code.
MCP Native
Use any Model Context Protocol server out of the box. Build your own MCP tools with a single file.
Blazing Fast
Rust core. Streaming by default. Sub-second tool calls on local models. Designed for daily driving.
Designed like a modern kernel.
XR has a Rust core, a typed skill layer, and a model router — with a permission boundary between every component.
From zero to shipping in minutes.
No containers. No API gymnastics. A real runtime you can use today.
Start building with XR today.
Free for individuals. Pro for builders who ship. Enterprise for teams that need control.
Answers to common questions.
Is XR open-source?+
Yes. The XR runtime, CLI, and most official skills are MIT-licensed. Some cloud-only features ship under a commercial license.
Which editors are supported?+
VS Code, Neovim, JetBrains IDEs, Zed, Cursor, and Windsurf all have first-party extensions. The CLI works anywhere.
Can I run XR locally?+
Absolutely. The runtime is designed for local-first operation with Ollama, Llama.cpp, vLLM, and any OpenAI-compatible endpoint.
What is a skill?+
A skill is a typed, versioned, capability-scoped unit of work — think of it as a better tool. Skills compose into agents.
How is XR different from coding agents?+
Coding agents are one feature of XR. XR is a full runtime: shell, editor, skills, models, security, and deployment in one coherent system.
Do you support MCP?+
Yes. XR is MCP-native — you can use any MCP server without an adapter. We also extend MCP with typed streams and state.