ShellClaw

Local LLM autocomplete for your terminal

Type a Zsh command. See a useful suffix in gray. Accept it with Ctrl+Space. The model and your command memory stay on your Mac.

zsh Local

~ shellclaw status

shellclaw: running

Ctrl+Space accepts the gray suffix

0.5BFine-tuned Qwen2.5-Coder model
Localllama.cpp inference with Metal
<1 msMeasured 2,000-entry memory fast path
30 secIdle window before model unload

From install to first suggestion

Homebrew installs the binary and hook, downloads the local model, and starts the daemon. Open a new terminal when setup finishes.

Official repository and source code: GitHub - Edwardd02/Shell-Claw.

  1. InstallRun three Homebrew commands.
  2. TypeStart a command you actually use.
  3. AcceptPress Ctrl+Space when the suffix is useful.
Install with Homebrew
brew tap edwardd02/shellclawbrew trust edwardd02/shellclawbrew install shellclaw

A short path from keystroke to ghost text

Memory handles familiar commands quickly. The local model takes over when recall has no valid match.

01

Zsh hook

Captures the current line, cursor, directory, and response deadline.

02

Local daemon

Checks SQLite memory first, then uses Qwen2.5-Coder through llama.cpp.

03

POSTDISPLAY

Renders only the validated suffix and ignores stale responses.

Personal without a hosted profile

Executed commands are stored in a local SQLite database so future suggestions can match your habits and current working directory.

  • InferenceRuns on your Mac with no hosted API.
  • MemoryLives in ~/.shellclaw/memory.db.
  • TelemetryNone. Persistent file logs are off by default.
  • ControlDelete ~/.shellclaw to erase local data.

Built for Apple Silicon and Zsh

Bash support is experimental. Linux and Intel macOS are not released yet.

Share feedback