grekt-keywords 0.1.0

Keyword extraction for grekt artifacts using KeyBERT

grekt-keywords

Keyword extraction for grekt artifacts using KeyBERT.

Installation

# Recommended: CPU-only (~300MB)
pip install grekt-keywords --index-url https://download.pytorch.org/whl/cpu

# With GPU support (~4GB)
pip install grekt-keywords

Usage

CLI

# From argument
grekt-keywords "An agent that helps with Kubernetes deployments and Helm charts"

# From stdin
echo "An agent that helps with Kubernetes deployments" | grekt-keywords

# JSON output (for programmatic use)
grekt-keywords --json "Your artifact description here"

Python API

from grekt_keywords import extract_keywords

keywords = extract_keywords("An agent that helps with Kubernetes deployments")
# ['kubernetes', 'deployments', 'agent', 'helps']

Options

Option	Default	Description
--min	3	Minimum number of keywords
--max	5	Maximum number of keywords
--json	false	Output as JSON array

Why?

grekt artifacts require keywords for discoverability in the lazy-loading index. This tool extracts meaningful keywords from your artifact description using KeyBERT + MiniLM.

Model

Uses all-MiniLM-L6-v2 (~80MB) for keyword extraction.

First run: Downloads the model from Hugging Face Hub and caches it in ~/.cache/huggingface/.

Subsequent runs: Uses the cached model (works offline).

For CI/CD environments, the model will be downloaded on first execution. You can pre-cache it by running grekt-keywords "test" once.

License

MIT