downcraft

Generic HTTP/HTTPS downloader with cross-session resume via HTTP Range headers. Survives power loss, process crashes, and days-long gaps between sessions.

Quick start

pip install downcraft

# Download any URL
downcraft url https://example.com/bigfile.iso /tmp/bigfile.iso

# Or a HuggingFace model
downcraft hf Qwen/Qwen2.5-0.5B-Instruct

# Check status
downcraft status https://example.com/bigfile.iso

# List all tracked downloads
downcraft list

How resume works

1. Download starts → writes to file.sgpart
2. Process crashes or connection drops → .sgpart + ~/.downcraft/state.json survive
3. Same URL requested again → detects .sgpart, sends Range: bytes=N- to server
4. Server sends only remaining bytes → appended to .sgpart
5. On completion → atomically renames to final filename

Python API

from downcraft import download

# Generic URL
result = download("https://example.com/bigfile.iso", "/tmp/bigfile.iso")
print(result["status"])  # "complete"

# With progress callback
def on_progress(downloaded, total, speed_bps):
    pct = int(downloaded / total * 100) if total else 0
    print(f"\r{pct}% @ {speed_bps/1e6:.1f} MB/s", end="")

download("https://...", "/tmp/file", on_progress=on_progress)

# HuggingFace model (requires huggingface-hub)
from downcraft import download_hf_model

result = download_hf_model("gpt2")
print(result["cache_dir"])  # path to HF cache

CLI

Dependencies

• requests (required) — HTTP with Range header support
• huggingface-hub (optional) — only needed for HF model commands

Install with HF support: pip install downcraft[hf]