agent-environment

OpenART is an open-source framework designed to evaluate the safety and robustness of autonomous AI agents in dynamic, long-horizon, and stateful environments. It stress-tests agent runtimes against multi-step state poisoning, privilege escalation, and tool-use vulnerabilities across 10,000+ benchmark scenarios.

Branchable agent containers: checkpoint a running sandbox (files, memory and live terminal sessions) and fork it into many independent live copies. Built on CRIU and OverlayFS.

AgentX Registry API for workspace manifests, policies, artifacts, and devices

AgentX CLI for reproducible AI agent environments

Companions Online: A sandbox survival MMO -gather, craft, build, and explore alongside your AI companions. MCP-enabled. Research prototype, early alpha

PLATO MUD server — the room is the interface. Text-first agent environment.

Diagnoses the gap between configured state and what AI coding agents actually load at runtime. Read only.

Agnostic Shell Controller (ASC) — scaffolding bash CLI for local human and agent project workflows

Language world model reference implementation (Qwen-AgentWorld style): one latent backbone predicts the next textual observation across 7 agent environments (MCP/Search/Terminal/SWE/Web/OS/Android) with cross-env transfer. Deterministic oracle + monotone benchmark. Pure Python, zero deps.

RegTriage is an OpenEnv RL environment that trains agents to perform regulatory compliance auditing on financial services contact center transcripts

OpenEnv-compatible ticket-booking environment for Hugging Face Spaces, featuring FastAPI APIs, interactive web UI, and multi-task evaluation (movie, train, flight booking).

Historical archive. Development moved to rsgcsg/STS2-The-Pefect-Defect-Project.

An AI-playable historical technology simulator about rebuilding civilization's tech tree from different eras, societies, roles, and constraints.

OpenEnv-compatible AI agent environment for reviewing LLM-generated content — hallucination, bias, toxicity & incoherence detection

Sandbox protocol, adapters and MCP bridge that let people and AI agents play IOCALC

Playable CUHK-Shenzhen 3D campus world for human and AI-agent navigation, interaction and reproducible research.