Setting up Vikshep

Prerequisites

• +Python 3.10 or later
• +Git
• +No GPU required for the Geant4 Direct Interface or any loader

Path 1 — Loaders (repo-local, no account)

Clone the public repo and install the ingest package in editable mode. This gives you all data loaders, the DisCo recipes, and the benchmark harness. No engine binary is required.

git clone https://github.com/samvardhan03/Vikshep.git
cd Vikshep
pip install -e backend/ingest

vikshep-ingest is not yet on PyPI — install from the repo clone only.

Verify the install

vikshep-ingest --help
vikshep-recipe --help

Both commands should print usage without errors.

Path 2 — Scattering features (requires Rust)

To use the wavelet scattering transform, install the open core from source. A Rust toolchain is required for the build step.

# Install Rust (skip if already present)
curl -sSf https://sh.rustup.rs | sh -s -- -y && source "$HOME/.cargo/env"

pip install maturin
pip install -e "backend/ingest[scattering]"

The [scattering] extra builds vikshep_compute from the open core (samvardhan03/vikshep-compute) at a pinned commit. First build takes 2–4 minutes; subsequent builds use the cargo cache. Without it, the CLI falls back to per-event aggregates; the report states which path ran.

MCP data plane (vikshep-mcp)

Build the MCP data plane binary from the open core and set VIKSHEP_MCP_BIN:

git clone https://github.com/samvardhan03/vikshep-compute.git
cd vikshep-compute
cargo build --release -p vikshep-mcp
export VIKSHEP_MCP_BIN=$PWD/target/release/vikshep-mcp

Add the export to your shell profile (~/.zshrc, ~/.bashrc). The agent reads VIKSHEP_MCP_BIN on every invocation; if it is absent the agent fails with build instructions. GPU backends are in development and not yet released.

Supported platforms

Troubleshooting