经 AI Skill Hub 精选评估,Pixi LLM 配方 获评「推荐使用」。这款AI工具在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 7.5 分,适合有一定技术背景的用户使用。
Pixi LLM 配方 是一款基于 Shell 开发的开源工具,专注于 AI、LLM、Shell 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。
Pixi LLM 配方 是一款基于 Shell 开发的开源工具,专注于 AI、LLM、Shell 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/crusaderky/pixi-llm-recipes cd pixi-llm-recipes # 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 pixi-llm-recipes --help # 基本运行 pixi-llm-recipes [options] <input> # 详细使用说明请查阅文档 # https://github.com/crusaderky/pixi-llm-recipes
# pixi-llm-recipes 配置说明 # 查看配置选项 pixi-llm-recipes --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 PIXI_LLM_RECIPES_CONFIG="/path/to/config.yml"
My personal setup for running LLMs locally and using them with the pi coding agent. Everything lives in pixi environments, so there's no Docker and nothing to install by hand: clone the repo, pixi install, and you're done. Everything is pinned and 100% reproducible.
Only linux-64, linux-aarch64, and win-64 are configured. There are no macOS or Windows CUDA builds simply because I have no hardware to test them on. There are no Windows source builds because the OS defeated me. PRs welcome if you can validate them.
The pi-extensions conda package installs a pinned selection of pi plugins, so the agent setup is versioned and reproducible:
| Extension | Purpose |
|---|---|
| [pi-autoresearch](https://pi.dev/packages/pi-autoresearch) | autonomous experiment loops for optimization |
| [pi-btw](https://pi.dev/packages/pi-btw) | build-time workspace tooling |
| [pi-llama-cpp](https://pi.dev/packages/pi-llama-cpp) | zero-config llama.cpp integration |
| [pi-ollama-cloud](https://pi.dev/packages/pi-ollama-cloud) | Ollama cloud model provider + web search / web fetch |
| [rpiv-advisor](https://pi.dev/packages/@juicesharp/rpiv-advisor) | your local model can ask a larger datacenter model when in trouble |
| [rpiv-ask-user-question](https://pi.dev/packages/@juicesharp/rpiv-ask-user-question) | stop and ask the user when in doubt |
| [pi-token-speed](https://pi.dev/packages/pi-token-speed) | token throughput monitoring |
| [pi-usage-extension](https://pi.dev/packages/@tmustier/pi-usage-extension) | tokens usage tracking |
| [caveman](https://github.com/JuliusBrussee/caveman) | drastically reduce output tokens consumed |
| [rtk](https://github.com/rtk-ai/rtk) | drastically reduce input tokens consumed |
By default pi runs inside a bubblewrap container: read-only root filesystem and no access to /home beyond the workspace directory you point it at. This is the recommended way to run it (Linux only).
pixi r install-apparmor # one-off: install AppArmor profile for BubbleWrap
pixi r pi /path/to/workspace # sandboxed
pixi r pi # sandboxed in a temporary directory (just for chatting)
pixi r pi /path/to/workspace -- -p "Hello" # Pass arbitrary parameters
pixi r pi - -- -p "Hello" # In a temporary directory; pass arbitrary parameters
If you need full host access for development or debugging, or if you are on Windows, there's an escape hatch:
pixi r pi-unsafe /path/to/workspace
pixi r pi-unsafe # In a temporary directory
# (useful to run with no AGENTS.md)
pixi r pi-unsafe /path/to/workspace -- -p "Hello" # Pass arbitrary parameters
pixi r pi-unsafe - -- -p "Hello" # In a temporary directory; pass arbitrary parameters
高质量的开源AI工具,易于安装和使用
该工具未明确声明开源协议,商业使用前请联系原作者确认授权范围,避免侵权风险。
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。
AI Skill Hub 点评:Pixi LLM 配方 的核心功能完整,质量良好。对于AI 技术爱好者来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | pixi-llm-recipes |
| 原始描述 | 开源AI工具:pixi recipes for local LLM stack。⭐8 · Shell |
| Topics | AILLMShell |
| GitHub | https://github.com/crusaderky/pixi-llm-recipes |
| 语言 | Shell |
收录时间:2026-06-11 · 更新时间:2026-06-11 · License:未公布 · AI Skill Hub 不对第三方内容的准确性作法律背书。