RagLeapragleap-physical 0.0.1 (pre-alpha, simulation only)
Install: pip install ragleap-physical==0.0.1 (optional extras: mujoco, test, interop).
A simulation-first, observe-first safety bridge between AI agents and simulators. The AI can only propose a change; a named person approves
the exact change; the bridge checks the limits again when it applies it; the real outcome is verified in the simulator.
What is in this release
- A safety envelope and bridge: exact-proposal approval, dry run, runtime monitoring, stop and an audit log.
- Four AI-facing tools and a small MCP server (loopback by default, optional bearer token).
- MuJoCo adapters (optional extra), workspace limits, a two-link and a three-joint arm, an offline ROS 2 observe-only mock,
a simulated machine tool with a tool-offset envelope, and a repeatability experiment. - Hardening after independent review: stricter input and prediction checks, and no self-approval.
Verified: 185 tests passed on the published source archive with MuJoCo and the MCP SDK installed (Colab, 2026-11-02); GitHub CI on Python 3.10, 3.11 and 3.12.
Not claimed: anything about real robots, real machines, safety certification or regulatory compliance. Nothing here controls real hardware.
MuJoCo results are repeatable within one installed environment only. The wear-study and machine-tool numbers are invented.
AI-assisted. Changelog: https://github.com/ragleap/ragleap-core/blob/main/packages/ragleap-physical/CHANGELOG.md