Crowdbricks

Crowdbricks is a Python toolkit that generates batched character motion for NVIDIA MotionBricks on local NVIDIA GPUs.

Crowdbricks builds the NVIDIA MotionBricks ONNX model into a TensorRT plan with a batch dimension to generate character motion for crowds. It includes encode, decode, and endpoint kernels that process multiple requests at once. The system features a crowd runtime, a GPU motion bank, drivers for behaviors like walking and sitting, and obstacle wrapping for actions such as jumping and ducking. It outputs recorded takes with character poses and foot contacts for a web viewer. It runs on top of a MotionBricks checkout using an NVIDIA GPU.

Crowdbricks at a glance

Best for: Software engineer, Designer
Runs on: Web, Python, API
Pricing: Free
Released: October 2026
Category: AI agent tool
Website: github.com/patch-gg/crowdbricks

Key features

BatchGenerator holds batched plans and motion banks on the GPU
Crowd runtime steps characters and batches requests
Drivers handle walk, stand, approach, interact, and seated behaviors
GaitLibrary blends locomotion targets and styles
Obstacle wraps Smart Primitives for vaulting, hopping, ducking, and sitting
Recorder writes takes of character poses and generated motions

Use it for

Generate motion for 192 characters chasing a moving point at 60 Hz
Fork one character into 48 copies with the same past heading to different obstacles
Simulate crowd behaviors with varied gaits and obstacle interactions

Alternatives to Crowdbricks

MotionSites — Copy-and-paste prompts for building 3D websites with AI code tools.
NPCTalk — Adds real-time cloned NPC voice servers and AI agent skills for game mods.
OmniAgent — It evolves skills in real-time and dynamically hardens security during task execution.
NemoClaw — Provides managed security, networking, and lifecycle operations for AI agents.
Artisan — It uses physical simulation feedback to ensure collision-free motion and gravity stability.
AgentENV — Scales massive Firecracker environments across machines efficiently.
DeerFlow — Executes long tasks using sandboxes, memories, tools, and subagents.
Robot Heads — Visualizes agent states with customizable robot head components.
MuscleMemory — It cuts input tokens, tool calls, and execution time while maintaining task success.
AutoSwarm Shadow Adapter — It prevents engine timeouts during human sign-offs and remote execution.
Qwen MM Plugins — Enables agents to process multimodal data natively rather than via APIs.
AgentGarten — Runs executable code worlds with real-time neural rendering for agents.

All Crowdbricks alternatives