PUMPFUN HACKATHON WINNER
Robotics, accessible to everyone.
We're building open-source tools to build, simulate, deploy and train robots.
SimArena is the first one — V1 is live.
We're building open-source tools to build, simulate, deploy and train robots.
SimArena is the first one — V1 is live.
Robotics today takes weeks of setup. Every step of the work uses a different tool, and every tool comes with its own wall.
CodecFlow is the foundation underneath. One place to build for the physical world, with the layer of compute, runtime, and intelligence already in place.
Loved by leaders of the industry
Browser-based robot simulator.
Building a robot should feel like vibe-coding a website. Describe a scene, drop in a robot, watch it move. Export to MuJoCo or Isaac Sim when you're ready to scale.
A student, an indie builder, a lab, or a startup. Everyone has the same entry point.
No install. No license. No GPU.
Serverless compute layer.
Serverless GPU compute for simulation, training, and inference. Pay per second. Routes between cloud and edge based on what the task needs and how fast the robot needs to react.
Runtime for robot intelligence.
The runtime that connects your robot to the cloud. A vision-language-action loop that runs on the chip you ship. Same operator works on a desktop arm or a humanoid. Open source. Plugs into existing ROS stacks.
Browser-based robot simulator.
Building a robot should feel like vibe-coding a website. Describe a scene, drop in a robot, watch it move. Export to MuJoCo or Isaac Sim when you're ready to scale.
A student, an indie builder, a lab, or a startup. Everyone has the same entry point.
No install. No license. No GPU.
If you're building something in robotics, we want to hear about it.
TALK TO USCodecFlow is the execution engine that turns AI into action.
It lets AI models operate real systems, including robots, desktops, simulations, and digital environments through intelligent agents called Operators.
If large language models are the brain, CodecFlow is the system that lets that brain act.
Our goal is to become the execution layer of the robotics and automation economy.
No. Robotics is a core focus, but CodecFlow supports any environment where AI needs to act, including:
The long-term vision is a universal execution layer for embodied and digital AI.
CodecFlow is built for:
An Operator is an AI agent that runs in a continuous loop of:
Operators adapt to changing conditions and behave more like intelligent workers than fixed bots.
Traditional automation relies on fixed rules and scripts.
Operators rely on perception and context.
RPA breaks when a button moves. Operators understand what they're seeing and adjust in real time.
That makes CodecFlow suitable for dynamic software, robotics, and physical systems.
No.
CodecFlow supports:
The goal is to make building AI Operators as accessible as building software, without requiring deep robotics expertise.
The OPTR SDK is the developer toolkit for building and running AI Operators on CodecFlow.
It lets teams integrate robotics or automation models in minutes, connect them to real environments, and deploy them through the CodecFlow runtime.
Instead of wiring together simulation, inference, and execution from scratch, OPTR gives you a unified way to turn models into acting agents.
Creators earn by:
When an Operator is deployed, the platform handles execution and routes a share of usage fees back to the creator.
This turns robotics components into sustainable revenue streams.
No. CodecFlow complements frameworks like ROS.
ROS handles low-level communication and hardware control. CodecFlow acts as a high-level execution and coordination layer.
Teams can integrate Operators into existing ROS stacks to add modular intelligence without rebuilding their systems.
CodecFlow focuses on reusable components, not porting entire stacks.
Developers can use specific Operators representing discrete abilities like grasping or detection.
These plug-and-play components can be shared, monetized, and integrated across different robotic platforms.
SimArena is CodecFlow's browser-based simulation environment for robotics.
It lets builders create environments, run Operators, collect data, and test behaviors without needing physical hardware.
Instead of setting up heavy local infrastructure, teams can iterate directly in the browser and move from simulation to real-world deployment faster.
Fabric optimizes where and how AI workloads run.
It prioritizes on-device compute for real-time tasks and routes heavier models to cloud resources based on location and network conditions.
This reduces latency and ensures robots respond safely and smoothly in real time.
