Getting Started

The infotheory binary is feature-gated, so the reliable first build is:

git clone --recursive https://github.com/turtle261/infotheory
cd infotheory
cargo build --release --features cli --bin infotheory

That gives you the current CLI surface under ./target/release/infotheory.

The crate defaults already include the main backend families, but the CLI itself still needs the cli feature. VM-backed environments are a separate optional build because they depend on the Nyx stack and Linux/KVM-oriented infrastructure.

For most documentation and benchmarking work, start with the native CLI build above and only add VM support when you explicitly want Nyx-backed environments.

./target/release/infotheory mi file1.txt file2.txt 8
./target/release/infotheory nte file1.txt file2.txt --rate-backend ctw
./target/release/infotheory ncd a.bin b.bin --compression-backend rate-ac --rate-backend fac-ctw --method 16

The CLI can switch backend families without changing the outer command shape. That is the main idea to keep in mind when reading the rest of the docs: most of the tooling is one binary with backend selection rather than many separate tools.

The agent entrypoint is:

./target/release/infotheory aixi aixi_confs/ui_coin_flip.json

Use planner: "mc-aixi" for the tree-search planner or planner: "aiqi" for the return-prediction planner. The same command is used for both.

See Agents for the planner-specific details and Benchmarks for reproduction against PyAIXI and MC-AIXI C++.

The repository also ships a Python extension crate. That matters for benchmarking, scripting, and parity checks with external baselines, even if the Rust CLI remains the primary interface.