A local, offline AI agent powered by Gemma 4 with built-in tools (bash, read_file, write_file). Two implementations — Python and Rust — with identical behavior.
See SPEC.md for the full specification.
uv run python/gemma_agent.py
uv run python/gemma_agent.py --prompt="what time is it?"
uv run python/gemma_agent.py --model=e2b # smaller modelWeights download automatically from HuggingFace on first run.
cd rust
# Build a self-contained binary with embedded weights (~5.4 GB)
./build.sh
# Run it
./gemma-agent-packed
./gemma-agent-packed --prompt="what time is it?"
# Or build without embedding and pass a GGUF
cargo build --release --features metal # macOS
cargo build --release --features cuda # Linux + NVIDIA
./target/release/gemma-agent /path/to/model.ggufbash test.sh