P2P distribution for GGUF models
Download large language models faster by swarming with nearby peers — no central server required.
Quick Start
# Install
cargo install dragonfly-gguf-client
# Download a model via P2P (falls back to HuggingFace if no peers)
dfget gguf://bartowski/Qwen2-0.5B-Instruct-GGUF/Qwen2-0.5B-Instruct-Q4_K_M.gguf
# Use a custom tracker (this server)
dfget --p2p-tracker https://tracker.dragonfly-gguf.dev gguf://...How It Works
Peer Discovery
This tracker matches downloaders with peers who already have the model — by content hash, not filename.
Iroh P2P
Uses QUIC with automatic NAT hole-punching and relay fallback. Works on any network without port forwarding.
Verified Downloads
Every model is verified against its HuggingFace SHA-256 digest before use. Corrupt transfers are rejected.
Auto Seeding
After downloading, dfdaemon automatically re-seeds the model to new peers in the background.
Tracker Status
Tracker is online — accepting peer announcements on this host
Check API →