# OpenEAI Complete open-source hardware-software platform for real-world embodied AI from arm to VLA policy. ## Agent Decision Summary - Risk level: elevated - Source confidence: medium - Recommended workflows: Local or private AI stack, Robotics or embodied agent workflow - Permission surface: messages, hardware - Agent JSON: https://www.openagent.bot/bots/openeai.agent.json ## Summary OpenEAI is a fully open-source hardware-software unified platform for real-world embodied manipulation. It consists of two repositories: OpenEAI-Arm, a low-cost 6-DoF desktop robotic arm with complete manufacturing files, and OpenEAI-VLA, an end-to-end vision-language-action policy trained with a two-stage recipe (large-scale pretraining + task-specific fine-tuning). The platform covers the full pipeline — hardware design, low-level control, data collection, dataset processing, VLA training, and real-time deployment. ## Guide OpenEAI is a fully open-source platform that provides everything needed to do real-world embodied AI research — from the robot arm itself to the VLA policy that controls it. ### What it is OpenEAI is a unified platform with two components: OpenEAI-Arm, a low-cost 6-DoF desktop robotic arm with complete manufacturing files, assembly guides, and a full C++/Python control stack; and OpenEAI-VLA, an end-to-end vision-language-action policy with a two-stage training recipe (large-scale pretraining + task-specific fine-tuning). The platform supports multi-modal teleoperation (GELLO, SpaceMouse, VR), LeRobot-format dataset processing, and real-time deployment via a client-server architecture. ### Why it matters Embodied AI research has been limited by two factors: the cost and complexity of robot hardware, and the fragmented nature of training pipelines. OpenEAI addresses both by open-sourcing an entire platform. A research group can build the arm, collect demonstration data, train a state-of-the-art VLA policy, and deploy it — all with publicly available code and designs. This is a model for how open-source can accelerate embodied AI research. ### How it works The hardware is a 6-DoF arm with a 2kg payload, designed for reproducibility with STEP/STL files and manufacturing drawings. Low-level control runs C++ drivers with gravity compensation and feed-forward PID tracking for smooth execution. The VLA pipeline uses dataset adapters to unify heterogeneous state/action conventions from public datasets, trains on large-scale pretraining data, and fine-tunes on task-specific demonstrations. Deployment uses a standard robot-client / policy-server ZMQ interface for streaming observations and receiving action chunks. ## Use Cases - End-to-end VLA research with reproducible hardware: Build the OpenEAI arm, collect data, train a VLA policy, and deploy — all steps are documented, open-source, and reproducible by any other lab. - Cross-dataset VLA pretraining evaluation: Use OpenEAI's dataset adapters to train VLA policies on heterogeneous public datasets and evaluate how well pretraining transfers to real-world hardware. - Teleoperation method comparison: Compare data quality and policy success rates across GELLO, SpaceMouse, and VR teleoperation methods using the same arm and task setup. ## Alternatives - Use LeRobot + SO-100 for a simpler entry point vs OpenEAI: If you only need data collection and imitation learning without VLA training, LeRobot with SO-100 is simpler. OpenEAI is for teams that want the complete hardware-to-VLA pipeline. ### Getting Started - OpenEAI-Arm repository: https://github.com/eai-yeslab/OpenEAI-Arm - OpenEAI-VLA repository: https://github.com/eai-yeslab/OpenEAI-VLA ### FAQ - How much does OpenEAI-Arm cost? - The exact BOM cost is documented in the repository. It is designed to be significantly cheaper than Franka/UR arms while providing sufficient capability for VLA research. - Can I use OpenEAI-VLA without the OpenEAI-Arm hardware? - Yes, the VLA training pipeline works with any robot. Use the dataset adapters to convert your robot's data format and fine-tune the policy. - Is OpenEAI commercially usable? - OpenEAI is licensed under BSD 3-Clause, which permits commercial use with attribution. ## What It Does OpenEAI is a unified platform with two components: OpenEAI-Arm, a low-cost 6-DoF desktop robotic arm with complete manufacturing files, assembly guides, and a full C++/Python control stack; and OpenEAI-VLA, an end-to-end vision-language-action policy with a two-stage training recipe (large-scale pretraining + task-specific fine-tuning). The platform supports multi-modal teleoperation (GELLO, SpaceMouse, VR), LeRobot-format dataset processing, and real-time deployment via a client-server architecture. ## How To Evaluate The hardware is a 6-DoF arm with a 2kg payload, designed for reproducibility with STEP/STL files and manufacturing drawings. Low-level control runs C++ drivers with gravity compensation and feed-forward PID tracking for smooth execution. The VLA pipeline uses dataset adapters to unify heterogeneous state/action conventions from public datasets, trains on large-scale pretraining data, and fine-tunes on task-specific demonstrations. Deployment uses a standard robot-client / policy-server ZMQ interface for streaming observations and receiving action chunks. ## Why It Matters OpenEAI matters because it is one of the few projects that open-sources the complete pipeline from robot hardware to trained VLA policy. Most VLA research uses expensive closed-source arms (Franka, UR) and proprietary training pipelines. OpenEAI provides reproducible manufacturing files for a capable 6-DoF arm (2kg payload, desktop form factor), along with a VLA training recipe that works with public datasets. This lowers the barrier for researchers to enter embodied AI research without industry budgets. ## Best For - Researchers needing a reproducible hardware platform for VLA research - Teams building custom robotic manipulation systems on a budget - Academics teaching embodied AI with open-source tools end-to-end ## Not For - Production deployment (designed for research, not industrial use) - Researchers needing non-manipulation platforms like mobile or humanoid robots ## What It Actually Does - End-to-end open-source pipeline from hardware to VLA: OpenEAI releases the complete stack: CAD files, manufacturing drawings, low-level C++ control, multi-modal teleoperation, dataset processing, two-stage VLA training, and real-time deployment. - Why it matters: Most embodied AI papers release only model weights or simulation code. OpenEAI lets anyone reproduce the full system. - Low-cost 6-DoF arm with 2kg payload: The OpenEAI-Arm is a desktop 6-DoF manipulator with 2kg payload capacity, significantly cheaper than Franka/UR arms while maintaining sufficient capability for VLA research. - Why it matters: Cost has been the primary barrier to entering real-world robotics research. OpenEAI's arm design is reproducible for a fraction of the cost of commercial alternatives. - Two-stage VLA training recipe: The VLA training pipeline uses large-scale pretraining on public robot datasets followed by task-specific fine-tuning with as few as 10-50 demonstrations. - Why it matters: This recipe addresses the data efficiency challenge — you get the benefits of large-scale pretraining without needing to collect millions of your own demonstrations. - Multi-modal teleoperation support: Supports GELLO (puppet), SpaceMouse (delta pose), and VR (absolute pose) teleoperation methods out of the box. - Why it matters: Different data collection scenarios require different teleoperation interfaces. OpenEAI covers the three most common modalities. ## Typical Use Cases - Reproducible VLA research: Use OpenEAI's complete pipeline to conduct VLA research on hardware that any other lab can reproduce, enabling verifiable and comparable results. - Teaching embodied AI from end to end: Build the arm, collect data, train a VLA policy, and deploy — all with open-source tools. A complete embodied AI curriculum in one platform. - Custom manipulation task development: Design a new manipulation task, collect demonstrations via VR teleoperation, fine-tune the VLA policy, and evaluate on real hardware — all within the OpenEAI framework. ## How It Compares - Choose OpenEAI for the most complete open-source VLA pipeline vs other open robot arms: AIRA and SO-100 provide excellent hardware but do not include a full VLA training pipeline. OpenEAI is the choice when you want hardware + training in one open-source platform. ## Fit Matrix - Local or private AI stack: strong. OpenEAI has multiple signals for local or private ai stack, including matching tags, capabilities, category, or positioning. Required check: Verify hardware requirements, data path, storage, and whether all calls stay in your environment. - Robotics or embodied agent workflow: strong. OpenEAI has multiple signals for robotics or embodied agent workflow, including matching tags, capabilities, category, or positioning. Required check: Separate simulator claims from hardware claims and verify safety boundaries before real-world operation. - Browser automation: partial. OpenEAI has at least one signal for browser automation, but should be checked against a real task before adoption. Required check: Run one non-sensitive website task and inspect clicks, waits, retries, and changed URLs. - Coding agent workflow: partial. OpenEAI has at least one signal for coding agent workflow, but should be checked against a real task before adoption. Required check: Run a small repository change and inspect the diff, tests, and rollback path. - Connector or protocol layer: weak. OpenEAI is not primarily positioned for connector or protocol layer in the current metadata. Required check: Connect one low-risk service, then inspect schemas, auth scope, errors, and logs. - Evaluation and observability: weak. OpenEAI is not primarily positioned for evaluation and observability in the current metadata. Required check: Add one repeatable test case and confirm results can run again in review or CI. ## Evidence - verified: OpenEAI is listed as open source. Source: License metadata: BSD-3-Clause - verified: OpenEAI has a recorded GitHub repository: eai-yeslab/OpenEAI-Arm. Source: Resource facts and GitHub source link. - inferred: OpenEAI supports these recorded deployment modes: self hosted, cloud. Source: OpenAgent decision signal metadata. - inferred: OpenEAI is tagged with robotics, messaging capabilities. Source: OpenAgent capability taxonomy. ## Missing Checks - Dedicated docs link is missing. - Repository freshness has not been recorded. ## Next Actions - Inspect repository: https://github.com/eai-yeslab/OpenEAI-Arm - Inspect repository: https://github.com/eai-yeslab/OpenEAI-VLA - Install OpenEAI stack: # Clone both repositories git clone https://github.com/eai-yeslab/OpenEAI-Arm git clone https://github.com/eai-yeslab/OpenEAI-VLA # Follow the hardware assembly guide and training recipe in each repo ## Command Line ### Install OpenEAI stack Clone both repositories and follow the respective READMEs for hardware assembly and VLA training. ```bash # Clone both repositories git clone https://github.com/eai-yeslab/OpenEAI-Arm git clone https://github.com/eai-yeslab/OpenEAI-VLA # Follow the hardware assembly guide and training recipe in each repo ``` ## Facts - Category: bots - Resource type: bot - Open source: yes - License: BSD-3-Clause - Last verified: 2026-06-04 - GitHub repo: eai-yeslab/OpenEAI-Arm - GitHub stars: 622 ## Capabilities - robotics - messaging ## Structured Use Case Tags - self-hosted-ai - robotics-agent ## Getting Started - OpenEAI-Arm GitHub: https://github.com/eai-yeslab/OpenEAI-Arm - OpenEAI-VLA GitHub: https://github.com/eai-yeslab/OpenEAI-VLA ## Links - GitHub: https://github.com/eai-yeslab/OpenEAI-Arm - Homepage: https://github.com/eai-yeslab/OpenEAI-VLA ## Structured Outputs - JSON: https://www.openagent.bot/bots/openeai.json - Markdown: https://www.openagent.bot/bots/openeai.md - Agent JSON: https://www.openagent.bot/bots/openeai.agent.json - Canonical: https://www.openagent.bot/bots/openeai