Please report security issues privately through GitHub's private vulnerability reporting rather than a public issue. This is a small alpha project, so expect a best-effort response.
benchrig is a local, single-user tool.
- The coding suite executes code generated by the model in a subprocess (
benchrig/core/sandbox.py) with a timeout (the whole process group is killed) and no other isolation. This is not a security boundary: run benchmarks only with models and scenarios you trust, and preferably inside a container or VM. - It talks to Ollama, Microsoft Foundry Local and ONNX Runtime GenAI over local endpoints configured in
config.yaml.
Supported versions: the latest release only.