Using the CLI

The ffai executable is a SwiftPM product, not a Homebrew formula — there’s no global install step. After cloning the repo, build it with swift build and invoke it through SwiftPM (swift run ffai …), the built binary path, or by symlinking onto PATH.

Build

git clone https://github.com/thewafflehaus/FFAI
cd FFAI

swift build -c release        # binary lands at .build/release/ffai

Use -c debug (the SwiftPM default) for faster compile + slower run; -c release for the inference numbers you’d quote.

Run

Pick one of three invocations — they’re equivalent, just trade-offs on ergonomics.

# (a) Via SwiftPM — no setup, recompiles if the source changed.
swift run -c release ffai generate -m unsloth/Llama-3.2-1B -p "Once upon a time"

# (b) Direct binary path — no recompile check, fastest start-up.
.build/release/ffai generate -m unsloth/Llama-3.2-1B -p "Once upon a time"

# (c) Symlink onto PATH (one-time) so plain `ffai …` works from anywhere.
ln -s "$PWD/.build/release/ffai" /usr/local/bin/ffai
ffai generate -m unsloth/Llama-3.2-1B -p "Once upon a time"

generate is the default subcommand, so the -m / -p flags can be passed directly to ffai (ffai -m … -p … is equivalent to ffai generate -m … -p …).

Subcommands

Subcommand	One-liner	More
generate (default)	Stream a single prompt’s continuation to stdout.	ffai generate --help
bench	Run a benchmark method against a model, append to a per-day report.	benchmarking.md

Common cross-cutting flags (--stats, --debug, --profiling) are documented in observability.md.

See also

• Quick start — the 5-line library equivalent.
• Benchmarking — ffai bench --method <name>, KLD comparisons, per-day report shape.
• Installation — adding FFAI to your own SwiftPM package (no CLI required).