Show HN: Please – local CLI that translates English –> tar (github.com)

Please is a new command-line tool that lets you control your terminal with natural-language commands while keeping all model inference local. Installable via Homebrew, an installer script, or cargo from the project's GitHub, the tool downloads a GGUF-format weight file (gpt-oss-20b-mxfp4.gguf) that you verify locally, and exposes prompts like "please summarize" or "please tar the current folder but without git-ignored files." The name is chosen so invocations read like natural phrases (e.g., "please format all the git-staged files"), and examples show it can generate commit messages, create one-off scripts from project files, resolve merge conflicts, and fix clippy diagnostics. This is significant for AI/ML practitioners interested in local-first workflows: it demonstrates an integrated developer UX for using open-source models on-device (or with host-side inference) without sending code or secrets to cloud APIs. The repo documents SSH bridging and Docker setups so you can run the CLI on a remote or container while performing inference on a local engine. Key technical implications: it relies on a ~20B open model weight (so expect nontrivial CPU/GPU/VRAM needs or quantized runtimes), uses GGUF weights and local verification (SHA256), and hooks into common developer tooling (git, Justfile, clippy).