AI Skill Hub 强烈推荐:rail AI技能包 是一款优质的AI工具。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的AI工具解决方案,这是一个值得深入了解的选择。
自托管系统语言,零C依赖,ARM64汇编GC
rail AI技能包 是一款基于 Haskell 开发的开源工具,专注于 haskell、aarch64、ai 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
自托管系统语言,零C依赖,ARM64汇编GC
rail AI技能包 是一款基于 Haskell 开发的开源工具,专注于 haskell、aarch64、ai 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
# 克隆仓库 git clone https://github.com/zemo-g/rail cd rail # 查看安装说明 cat README.md # 按 README 完成环境依赖安装后即可使用
# 查看帮助 rail --help # 基本运行 rail [options] <input> # 详细使用说明请查阅文档 # https://github.com/zemo-g/rail
# rail 配置说明 # 查看配置选项 rail --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export RAIL_CONFIG="/path/to/config.yml"
<p align="center"> <em>A self-hosting systems language that speaks TLS alone.</em><br> <sub>No C in the runtime. GC in ARM64 assembly. HTTPS in pure Rail.</sub> </p>
<p align="center"> <a href="#releases"><img src="https://img.shields.io/badge/v5.2.0-Stands%20alone%20(no%20as%2Fld%2Fcodesign)-ff5500?style=for-the-badge" alt="v5.2.0"></a> </p>
<p align="center"> <a href="#install"><img src="https://img.shields.io/badge/tests-178%2F178-brightgreen" alt="tests 178/178"></a> <a href="#why-rail"><img src="https://img.shields.io/badge/self--hosting-fixed%20point-blue" alt="self-hosting"></a> <a href="#what-rail-does"><img src="https://img.shields.io/badge/HTTPS-pure%20Rail-ff5500" alt="pure-Rail HTTPS"></a> <a href="#how-it-works"><img src="https://img.shields.io/badge/GC-ARM64%20assembly-purple" alt="GC in ARM64 asm"></a> <a href="#why-rail"><img src="https://img.shields.io/badge/C%20in%20seed-0-brightgreen" alt="zero C in the seed binary"></a> <a href="#releases"><img src="https://img.shields.io/badge/backends-6-orange" alt="6 backends"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-BSL%201.1-green" alt="BSL 1.1"></a> </p>
<p align="center"> <b><a href="#quick-start">Quick start</a></b> · <b><a href="#what-rail-does">What Rail does</a></b> · <b><a href="#why-rail">Why Rail</a></b> · <b><a href="CHANGELOG.md">Changelog</a></b> · <b><a href="https://github.com/zemo-g/rail/releases">Releases</a></b> </p>
---
Rail compiles itself. The compiler — ~9,200 lines of Rail — produces a ~0.9 MB ARM64 binary that compiles the compiler again and reaches a byte-identical fixed point in 2 cycles. There is no C in the runtime, no libc in the binary. The garbage collector is ARM64 assembly. The TLS 1.3 client is also Rail: import "stdlib/anthropic_client.rail" and your program talks HTTPS to api.anthropic.com with zero OpenSSL, zero curl, zero socat. As of v5.2.0, the toolchain stands entirely alone: Rail assembles, links, and code-signs its own Mach-O binaries in-process — no as, no ld, no codesign — so the self-compile is bit-reproducible (the committed seed reproduces itself byte-for-byte). It also emits its own aarch64 Linux ELF binaries and its own GPU kernels, generating Metal Shading Language from an op-DAG and JIT-compiling it at runtime (35× fused rmsnorm+QKV, 18× fused silu+hadamard). A frontier model + 1 KB Rail spec still compiles 30/30 on a held-out hard-bench — publicly reproducible.
./rail_native self && cp /tmp/rail_self ./rail_native # cycle 1
./rail_native self && cmp rail_native /tmp/rail_self # cycle 2 — byte-identical
./rail_native test # 178/178
./rail_native self -- ~9,200 lines of Rail →
-- a ~1.0 MB ARM64 binary
cp /tmp/rail_self ./rail_native -- cycle 1: install gen1
./rail_native self -- cycle 2: that binary compiles
-- the compiler again (gen2)
cmp rail_native /tmp/rail_self -- and the output is identical
-- (byte-identical fixed point)
The GC, allocator, and runtime support are ARM64 assembly embedded in the compiler itself. No gcc, no libc, no linker scripts — and since v5.2.0 no as or ld either: Rail assembles, links, and code-signs itself.
-- The self-training loop, in one flow:
-- LLM generates Rail → rail_native compiles (the oracle) →
-- passes harvested → training data feeds next round
The compiler is the fitness function. Programs that compile become training data; programs that don't are the gradient. Three independent lineages (LoRA on Gemma, Metal-GPU MLP, PCFG-REINFORCE) all use the same compiler as the binary verifier. 92 % strict pass rate on the PCFG lineage in 30 ticks.
git clone https://github.com/zemo-g/rail
cd rail
./rail_native run examples/hello.rail
Apple Silicon (ARM64 macOS) is the primary target; Linux ARM64, Linux x86_64, WebAssembly, Cortex-M4, and RISC-V rv32imc backends are supported.
./rail_native <file.rail> # compile to /tmp/rail_out
./rail_native run <file.rail> # compile + execute
./rail_native test # run the 178-test suite
./rail_native self # self-compile, fixed point at gen2
./rail_native x86 <file.rail> # cross-compile to Linux x86_64
./rail_native linux <file.rail> # cross-compile to Linux ARM64
./rail_native wasm <file.rail> # compile to WebAssembly
./rail_native cortexm <file.rail># compile to Cortex-M4 (Thumb-2)
./rail_native riscv32 <file.rail># compile to RISC-V rv32imc
创新性的自托管系统语言
该工具使用 NOASSERTION 协议,商用场景请仔细阅读协议条款,必要时咨询法律意见。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
📄 NOASSERTION — 请查阅原始协议条款了解具体使用限制。
总体来看,rail AI技能包 是一款质量优秀的AI工具,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | rail |
| 原始描述 | 开源AI工具:A self-hosting systems language. Zero C dependencies. GC in ARM64 assembly. Pure。⭐9 · Haskell |
| Topics | haskellaarch64aiarm64 |
| GitHub | https://github.com/zemo-g/rail |
| License | NOASSERTION |
| 语言 | Haskell |
收录时间:2026-05-16 · 更新时间:2026-05-30 · License:NOASSERTION · AI Skill Hub 不对第三方内容的准确性作法律背书。