AgentGarten

AgentGarten is an executable environment platform with a real-time neural renderer for evolving AI agents.

AgentGarten provides executable environments with a real-time neural renderer where code keeps the world and a learned renderer shows it. Objects, rules, and goals live in an executable program, while a learned neural renderer turns exported geometry into what the agent sees in real time. Pretrained agents play inside these worlds using rendered camera frames, write down what they learned after each round, and start the next round from those notes. The platform supports tasks such as building shelters, managing a companion dog, navigating a one-lane bridge with cars, herding sheep with dogs, and operating a quarry loader. The renderer starts from a pretrained video model, learns to follow geometry, and is distilled into a block-by-block student that keeps up with the agent.

AgentGarten at a glance

Best for: Software engineer, Researcher
Pricing: Free
Released: October 2026
Category: AI agent tool
Website: mirros-lab.github.io

Key features

Run executable code worlds with rules and goals
Generate real-time video frames via neural renderer
Allow agents to read, play, write notes, and archive
Support pretrained visual agents and self-play
Simulate shelter building and navigation tasks
Simulate multi-agent scenarios like bridge crossing
Simulate herding and quarry loading tasks

Use it for

Train visual agents in code-based worlds
Evaluate agent learning across multiple rounds
Simulate physical tasks like herding and loading

Alternatives to AgentGarten

AgentENV — Scales massive Firecracker environments across machines efficiently.
OmniAgent — It evolves skills in real-time and dynamically hardens security during task execution.
Strata — Runs large AI models locally on consumer gaming PCs without server reliance.
Agent Executor — Declarative runtime for scaling stateful agent workloads with isolation.
BrowserOS — Runs entirely locally and lets AI agents use real user logins to complete tasks.
Beeuni — Agents retain context and improve skills over time through a controlled review process.
Harness2 — It recursively refines agent execution structures using parallel and sequential search.
Composio — Equips AI agents with tools, managed user auth, and sandboxes across 1,500+ apps
Agent Lightning — Train existing artificial intelligence agents with reinforcement learning and zero code changes.
Starnet — Visual agent management with transparent, persistent on-disk logs and state.
Agentcraft — Manage AI coding agents through a visual Minecraft studio with safe git worktrees.
NeMo Molt — It scales fully-async agentic RL to frontier MoE scale with a small PyTorch stack.

All AgentGarten alternatives