Three tools: snipe_search (GPU-scored trust-ranked), snipe_enrich (deep BTF scraping), snipe_save (persist search to Snipe UI). GPU inference scoring uses VRAM + arch tier weighted composite. LLM-condensed output trims verbose listing dicts to trust/price/GPU/url. Configure via ~/.claude.json with SNIPE_API_URL env var pointing at local or cloud API. |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| formatters.py | ||
| gpu_scoring.py | ||
| server.py | ||