g4f-go

A single self-contained executable for gpt4free: a small Go launcher with an embedded, relocatable CPython runtime that has gpt4free pre-installed. No system Python required.

g4f-go client "What is gpt4free?"
g4f-go api --port 8080

How it works

1. fetch-python.sh downloads the official CPython embeddable zips from python.org for every supported platform, copies the g4f package into the interpreter's python-home/, drops prebuilt wheels in wheels/, patches python311._pth for relocation, and re-packs everything into embed/<os>/<os>-<arch>-embed-<ver>.zip.
2. go:embed (build tags per OS) bakes that archive into the launcher binary.
3. On first run the launcher extracts the archive next to itself, finishes the g4f install from the bundled wheels (offline), then execs python -m g4f <args>.
4. build-all.sh cross-compiles the launcher with CGO_ENABLED=0 for every target and zips each release.

┌──────────────────────────────┐
│  g4f-go (Go, ~10 MB binary)  │
│  ┌────────────────────────┐  │
│  │ CPython 3.14 runtime   │  │  ← baked in via go:embed
│  │ + g4f + wheels          │  │
│  └────────────────────────┘  │
└──────────────────────────────┘
   │  first run: extract next to binary
   ▼
g4f-go.exe/ + python-home/ + wheels/ + .g4f-runtime/.installed
   │
   ▼
exec python -m g4f <args>

Build

Requires Go 1.22+ and (for fetch-python.sh) bash, curl, unzip, zip, python3.

cd g4f-go

# 1. Download & embed the Python runtimes (all platforms, ~200 MB on disk)
./fetch-python.sh
./fetch-python.sh linux          # or just one platform

# 2. Build launchers for every OS/arch
./build-all.sh                   # -> dist/g4f-go-<ver>-<os>-<arch>.zip
# or a single one:
go build -o g4f-go .

Release zips live in dist/. Each zip is a portable folder: drop it anywhere, run g4f-go, and the runtime is extracted next to it on first launch.

Size note: CPython runtimes are ~40-60 MB per platform, so release zips are large (Windows ≈ 160 MB, Linux ≈ 820 MB). The .zip is stored inside the Go binary, and extracted on first run; the Go executable itself stays small.

Usage

g4f-go <g4f args...>        run gpt4free (e.g. g4f-go client "hello")
g4f-go api --port 8080      start the OpenAI-compatible API server
g4f-go gui                  launch the web GUI
g4f-go status               show embedded runtime status
g4f-go install g4f          (re)install the g4f package (network)
g4f-go help                 show help

Environment: G4F_PYTHON_ONLY=1 g4f-go --version prints the embedded python path (useful for wrapping the runtime from other tools).

Supported platforms

fetch-python.sh can be extended to other OSes; add a row and a matching runtime_<os>.go build-tagged file.

Layout inside a release folder

g4f-go (binary)
python.exe / python (launcher)
python-home/  g4f package + site-packages
wheels/       prebuilt g4f dependency wheels
.g4f-runtime/.installed   (stamp written on first run)

Limitations

• The runtime is extracted to disk next to the executable (this is what makes it relocatable — CPython cannot run a 100%-in-memory interpreter reliably).
• Bundled dependency versions are pinned by requirements-min.txt at build time; use g4f-go install g4f for network installs.
• macOS builds must be signed/notarized by the distributor for Gatekeeper.