AI Agents
harbor
A framework for evaluating and improving AI agents.
★ 5.9k starsApache-2.0
View guide →SimWorld: an Unreal Engine 5 simulator for evaluating LLM/VLM agents in physical and social worlds.
SimWorld is an open-source ai agents project by SimWorld-AI (SimWorld-AI/SimWorld on GitHub). It has 793 GitHub stars and 90 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 26 June 2026.
Check your hardware against the requirements (dedicated GPU, ample RAM and disk) and decide which environment package you need.
Clone the repository, create a Conda environment, and install the Python client in editable mode.
Download and extract the Base UE server package for your OS, and optionally add extra map packs.
Start the UE server, then run an example such as the gym-interface demo so an LLM agent navigates the world.
Realistic, open-ended environments with accurate physics and a rich asset library for embodied agent research.
Language-based procedural generation, a city layout generator, and a live asset retrieval and placement editor create varied scenarios.
LLM/VLM agents receive multimodal observations (including RGB, segmentation, and depth) and act through a local action planner for pedestrians, vehicles, and robots.
Built-in traffic simulation and weather modules add dynamic, social-world complexity to scenarios.
Start with the lightweight Base package, expand with 100+ pre-built maps, or import your own Unreal environments, assets, and agent models.
A single configuration file controls global simulator settings and module behaviour for city generation, traffic, assets, and agents.
Use Python 3.10 or later in a fresh Conda environment.
git clone https://github.com/SimWorld-AI/SimWorld.git
cd SimWorld
conda create -n simworld python=3.10 && conda activate simworld
pip install -e .Get the Base package for your OS from the README's Hugging Face links and extract it. # see README setup section
# download Base UE server package (Windows or Linux) per READMELaunch the UE server first, then run a Python example from the examples folder.
./SimWorld.sh # from the extracted UE server directory on Linux
# then run an example such as examples/gym_interface_demo.ipynbIt is maintained by the SimWorld-AI organisation, with the underlying research presented at venues including NeurIPS and CVPR demo tracks per the README.
Yes, SimWorld is Apache-2.0 licensed. The agents you run inside it use your own LLM or VLM access, such as an OpenAI key for the example agent.
The simulator itself needs no key, but its LLM/VLM agents call model APIs, so you provide a key for whichever model you plug in.
Simple grid-world or purely text simulators are cheaper to run; SimWorld trades that for Unreal-grade visual realism, physics, and multimodal perception for embodied agents.
Clone the repo, pip install -e . in a Python 3.10 Conda env, download the Base UE package, start the server, and run an example.
A framework for evaluating and improving AI agents.
A full-lifecycle platform for developing, debugging, evaluating and monitoring AI agents.
Research-backed lifelong memory for LLM agents, text and multimodal.
Make your agents learn from experience with a dedicated context engine.
Agentic memory for LLM agents, from a published research project.
Open-source LLM observability: trace, evaluate and monitor apps and agents
The most-starred other AI Agents projects in the LibreAI catalogue — the realistic shortlist if SimWorld is not the right fit.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Free and open — grab the source, star the repo, and start creating.