Vanity Ed25519 key generator for MeshCore. Find keys whose public key starts with a chosen hex prefix.

mc-keygen <PREFIX>... [OPTIONS]

Options:

• -t, --threads <N> — worker threads (default: all cores)
• --json — output result as JSON (no TUI, no color)
• --cpu-only — force CPU-only search, skip GPU even if available (requires cuda or metal feature)
• --gpu-only — force GPU-only search, no CPU threads (requires cuda or metal feature)
• --verify — cross-check GPU keygen against CPU (requires cuda or metal feature)
• --benchmark <SECS> — run GPU search for N seconds, tally every match with a no-early-exit kernel, and compare observed vs expected count to validate the reported rate (requires cuda or metal feature)

When built with GPU support (cuda or metal feature), the default mode is hybrid: both CPU threads and GPU run concurrently, and the first match from either wins. If no GPU is detected at runtime, the tool falls back to CPU-only with a warning.

Examples:

mc-keygen AB             # find a key starting with AB (hybrid if GPU available)
mc-keygen AB CD EF       # find a key matching ANY of these prefixes
mc-keygen DEAD -t 4      # use 4 threads
mc-keygen AB --json      # machine-readable output
mc-keygen ABCDEF --gpu-only   # GPU only for longer prefixes
mc-keygen AB --cpu-only       # force CPU even when GPU is available

Multiple prefixes can be passed in a single invocation — every key is checked against all of them, so searching for N prefixes is ~N× more efficient than N separate runs. The JSON output includes a matched_prefix field.

Each hex character multiplies expected attempts by 16:

¹ CPU: Ryzen 9 7950X3D, 32 threads (~1.3M keys/s). GPU: RTX 4090 (~68M keys/s). See docs/gpu.md for details.

cargo build --release                    # CPU only
cargo build --release --features cuda    # with NVIDIA GPU support
cargo build --release --features metal   # with Apple Metal GPU support

CUDA support requires the NVIDIA CUDA Toolkit. Metal support requires macOS with an Apple Silicon or AMD GPU. See docs/gpu.md for details.

1. Draw 64 random bytes from the OS CSPRNG
2. Clamp the first 32 bytes to form an Ed25519 scalar
3. Multiply the base point by the scalar to get the public key
4. If the hex starts with the target prefix, return; otherwise repeat

Keys starting with 00 or FF are skipped (reserved by MeshCore). On GPU, only the starting scalar per thread is drawn from the CSPRNG — successive candidates come from repeated +8B point addition (see docs/gpu.md).

• MeshCore — the mesh networking firmware these keys are for
• MeshCore mc-keygen web tool — reference implementation of the key generation algorithm
• Ed25519 / RFC 8032 — the signature scheme spec
• curve25519-dalek — Rust Ed25519 elliptic curve library
• ratatui — TUI framework for the progress display