Skip to content

v0.3.0 — Bundled binary, one-command install on every platform

Latest
Latest

Choose a tag to compare

View all tags
@gautamkishore gautamkishore released this 02 Jun 07:59
· 2 commits to main since this release
v0.3.0
ff1ab15

What's New

shellwise now ships with a bundled llama-server binary for macOS (arm64 + x64), Linux (x64), and Windows (x64). No more Visual Studio, no more CMake, no more separate Ollama install.

pip install shellwise   # that's it

Install

pip install shellwise
sw                      # first run downloads the model (~400 MB, one time)

Highlights

One-command install on every platform

  • Windows: no Visual Studio Build Tools, no CMake, no Ollama .exe
  • macOS: Apple Silicon and Intel both work out of the box
  • Linux: glibc-based x86_64 distributions (Ubuntu, Debian, Fedora, etc.)

Bundled llama-server

  • Ships the llama-server binary + all shared libraries in the wheel
  • Started on first query, kept alive for the session, killed cleanly on exit
  • OpenAI-compatible HTTP API (/v1/chat/completions) with /health polling
  • Graceful shutdown via SIGTERM → process group → SIGKILL fallback

Priority backend selection

shellwise picks the first available:

  1. Bundled llama-server (default — no extra install)
  2. llama-cpp-python (opt-in [cpu] or [gpu] extra)
  3. Ollama (last resort, no Python deps)

Robustness fixes

  • Model download now supports HTTP Range resume (Ctrl+C-safe, restartable)
  • SHA256 verification on every download
  • Multiple mirror URLs (Hugging Face primary, with fallbacks)
  • Streaming progress bar with byte-accurate reporting

Bug fixes

  • Ubuntu/Linux fix: subprocess.run(['command', '-v', x]) raised FileNotFoundError because command is a shell builtin on Linux, not a binary. Replaced with portable shutil.which().
  • Inline regex regression test added to prevent recurrence.

Compatibility

  • requires-python = ">=3.8" (down from 3.9)
  • All source files audited and updated with from __future__ import annotations for forward-compat
  • 182/182 tests passing

Alternative backends

If you'd rather use a different inference stack:

# GPU (CUDA)
CMAKE_ARGS="-DGGML_CUDA=on" pip install "shellwise[gpu]"

# Or use Ollama
ollama pull qwen2:0.5b
pip install shellwise   # without [cpu]

Links

By eulogik

Assets 2
Loading