NeuroClaw FAQ

Yes. The NeuroClaw repository is publicly available and can be used according to the project license in the repository. You do not pay a platform fee to inspect the code or run it locally; your own compute, storage, and third-party service costs still apply.

Yes. NeuroClaw can be installed as a skill or rule pack for Codex, Claude Code, Cursor, OpenClaw, Hermes, WorkBuddy, and QClaw. The host-agent AutoResearch loop also lets those agents use their own built-in models as NeuroClaw's hypothesis generator, critic, and experiment supervisor.

NeuroClaw runs locally unless you explicitly connect it to external services. Your datasets, files, and outputs remain under your control, and any prompts or artifacts sent to a third-party model depend on the backend you configure.

NeuroBench evaluates executability, output validity, planning quality, tool/skill reasonableness, and reproducibility readiness. It is designed for realistic neuroimaging workflows, not just one-shot question answering.

Use the Web UI with python core/agent/main.py --web --benchmark, or run the CLI batch runner with python core/agent/main.py --benchmark. To compare skill-enabled and baseline runs, add --benchmark-compare-skills.

Yes. NeuroClaw is designed for iterative experiment loops and can sustain long-running sessions. You can run separate sessions in different terminals or browser tabs, but the actual capacity depends on your local resources and the model backend you choose.

The bundled installer guides Python environment setup, related dependencies, neuroimaging toolchain paths, LLM backend selection, default directories, and Web UI attachment support.