shtym 0.1.3

AI-powered summary filter that distills any command's output.

shtym

AI-powered summary filter that distills any command's output.

Overview

Shtym is a command wrapper designed to reduce context size for both human users and AI coding agents. It wraps command execution and, when an LLM is available, summarizes the output; otherwise it passes output through unchanged.

Installation

pip install shtym

# with Ollama support (requires a running Ollama instance)
pip install "shtym[ollama]"

Configuration

When using Ollama, you can configure the behavior using environment variables:

• SHTYM_LLM_SETTINGS__BASE_URL: Ollama server URL (defaults to http://localhost:11434)
• SHTYM_LLM_SETTINGS__MODEL: Model to use (defaults to gpt-oss:20b)

Example:

export SHTYM_LLM_SETTINGS__BASE_URL=http://localhost:11434
export SHTYM_LLM_SETTINGS__MODEL=llama2
stym run pytest tests/

Usage

Wrap any command with stym run:

# Run tests
stym run pytest tests/

# Run linter
stym run ruff check .

# Build project
stym run npm run build

# Any command with options
stym run ls -la

# Pipe output to other commands
stym run pytest tests/ | grep FAILED

Key Features

• Exit code inheritance: Shtym preserves the wrapped command's exit code, making it CI/CD friendly
• Clean stdout: Output contains only command results, no progress indicators or metadata
• Transparent wrapper: Works seamlessly with existing workflows and scripts
• Optional LLM summaries: If Ollama is available, output is summarized by the configured model; otherwise passthrough is used automatically

Design Philosophy

Shtym follows Unix conventions for command wrappers (like sudo, timeout, time):

• Executes commands as subprocesses
• Inherits and propagates exit codes exactly
• Maintains clean stdout for composability
• Enables reliable integration with automated workflows

Development

For development documentation, see:

• Architecture Overview

License

MIT