khy

The AI-native operating system.
A hand-written OS kernel + a Claude-Code-class agentic CLI + a 16-backend AI gateway.
One install, batteries included.

60-Second Tour · What is khy · Quickstart · The Three Pillars · Compare · 中文

🤯 What is khy

khy is a full AI-native operating system — written almost entirely by one person. It spans every layer most projects only touch one of:

• a hand-written OS kernel in C (preemptive scheduler, demand paging, copy-on-write fork, POSIX-style signals, pipes, per-process fd tables, a dual ELF + PE loader) that boots and runs under QEMU;
• a Claude-Code-class agentic CLI — streaming TUI, tool-calling loop, permission gating, sub-agents, workflows, goal mode, context compaction;
• a 16-backend AI gateway that fronts Claude, Qwen, Cursor, Kiro, Windsurf, Warp, Trae, Ollama, Codex and more behind one API, with cascade failover and circuit breaking — no vendor lock-in;
• shipped through two parallel package channels (pip install khy-os / npm i -g @khy-os/khy-os) with a self-healing, multi-language dev environment.

It's the kind of project that usually takes a team and a few years. Here it's one repo you can pip install in 30 seconds and run today.

⚡ 60-Second Tour

pip install khy-os        # or:  npm install -g @khy-os/khy-os
khy                       # launch the agentic terminal

┌─ khy ─────────────────────────────────────────────── 18% ctx ─┐
│ › refactor the gateway adapter cascade and add a test          │
│                                                                │
│ 💭 思考 · 142 字 (Ctrl+O 展开)                                 │
│ ▸ 读取 · 编辑 · 执行命令 · 4 个步骤  ✓✓✓✓                       │
│   ✓ 完成                                                        │
│ ⠹ 生成中… · 12s · ~340 tok                                      │
└────────────────────────────────────────────────────────────────┘

📽️ Demo recording: drop an asciinema cast or GIF at assets/demo.gif and it renders here. See docs/launch/DEMO.md for the exact 45-second script that shows the kernel booting and the agent coding.

🚀 Quickstart

Install (pick one channel — both ship the identical workshop)

# pip channel
python3 -m pip install -U khy-os

# npm channel
npm install -g @khy-os/khy-os

Run

khy                       # agentic terminal (the headline experience)
khy preflight             # diagnose PATH / Node / deps / claude before first run
khy ai "summarize this repo"      # one-shot AI, no REPL
khy gateway status        # see which of the 16 AI backends are live

Boot the actual OS

khy iso build --output dist/khy-os.iso     # build a bootable ISO
# then attach dist/khy-os.iso in QEMU / VMware / VirtualBox

New here? Follow the 0-to-power-user ladder: 新手成长路线 (post-install checklist → configure AI → daily use → advanced → mastery).

Full guides: 文档索引 · Windows + VMware: windows-vmware-清单

🧱 The Three Pillars

1. A real, hand-written kernel

Not a wrapper around Linux — a kernel built from boot sector up, validated under QEMU at every phase:

Hybrid micro/monolithic design — performance paths in kernel space, I/O through IPC. Roadmap: real keyboard stdin and block-device persistence.

2. A Claude-Code-class agent in your terminal

• Streaming Ink TUI with collapsible process groups, persisted thinking, real context-fill meter, and a live token/stall indicator.
• A full tool-calling loop with structured results, loop detection, and max_tokens recovery — benchmarked against Claude Code / Qwen Code / OpenCode.
• Permission gating, AskUserQuestion, sub-agents, goal mode, and a visual workflow editor with a native execution engine.

3. A 16-backend AI gateway

One endpoint in front of every model you already pay for — bring your own keys, keep your data local:

Claude · Qwen · Cursor · Kiro · Windsurf · Warp · Trae · Ollama · Codex · …

Cascade failover, circuit breakers, transient-error cooldown, and per-user multi-tenant isolation. Falls back automatically when a channel goes down.

🗺️ Architecture

            ┌──────────────────────────────────────────────┐
            │  khy CLI  ·  Ink TUI  ·  agent loop  ·  goals │   ← what you touch
            └───────────────┬──────────────────────────────┘
                            │
        ┌───────────────────┴────────────────────┐
        │   AI Gateway (16 adapters, failover)    │   ← bring your own keys
        │   Claude · Qwen · Cursor · Ollama · …   │
        └───────────────────┬─────────────────────┘
                            │
   ┌────────────────────────┴───────────────────────────┐
   │  Node backend  ·  tools  ·  workflows  ·  services  │   ← runtime
   └────────────────────────┬───────────────────────────┘
                            │
            ┌───────────────┴──────────────────┐
            │  khy kernel (C)  ·  ELF/PE loader │   ← the OS underneath
            │  scheduler · paging · IPC · fork  │
            └──────────────────────────────────┘

   Shipped as one workshop through two channels:  pip install khy-os  /  npm i -g @khy-os/khy-os

🆚 How It Compares

khy isn't trying to beat any one of these — it's the only project that is all of them at once, from the kernel to the agent.

📦 What's in the box

platform/    Python launcher + bundled runtime + shared JS    (pip entrypoint)
services/    Node backend: CLI, gateway, adapters, tools, workflows
apps/        Vue management UI + upper-layer apps
kernel/      The OS: boot, src, ELF/PE loader, MoonBit WASM, ISO build
software/    khyquant — the default built-in app (quant trading)
docs/        架构 (architecture) · 指南 (guides) · 修复记录 (changelogs)
packaging/   npm channel (@khy-os/khy-os) mirroring the pip workshop

🛣️ Roadmap

• Real keyboard stdin + block-device persistence in the kernel
• Recorded boot-to-agent demo (assets/demo.gif)
• More gateway adapters + adapter capability matrix in docs
• Hosted playground

Changelog: CHANGELOG.md

⭐ Star history

🤝 Contributing & community

• Read CONTRIBUTING.md before opening a PR.
• Be kind — see CODE_OF_CONDUCT.md.
• Found a vulnerability? See SECURITY.md (please don't open a public issue).

📜 License

Source-available. khy is free to download, run, study, and use non-commercially — see LICENSE. Copying, modifying, or redistributing the source, and commercial use, require written permission from the author (孔浩原 / Kong Haoyuan). For commercial licensing or collaboration, reach out.

If a one-person AI-native OS is the kind of thing you want to exist,
⭐ star the repo — it's the cheapest way to tell the author to keep going.