Stringman: The $795 Open-Source Robot That Turns Your Ceiling Into a Crane Neufangled Robotics released Stringman, a $795 cable-driven parallel robot that mounts four ceiling-corner anchors with NEMA 17 stepper motors to move a suspended two-finger gripper through a workspace of up to 5m × 5m × 3m, with a 1.9 kg average payload and roughly 0.4 m/s traverse speed. Stringman runs Hugging Face LeRobot models including π0.5, JEPA and Multitask DiT, exposes an MCP server so an AI coding agent can watch its cameras and send movement commands, and ships its full stack — CAD, PCB designs, firmware and control software — under OSHWA open-source certification, controlled by the nf-robot 7.6.2 Python package released November 13, 2026. The launch drew 830 points and 245 comments on Hacker News, where commenters confirmed the kinematics while flagging that cables snag on furniture, doors and pets in real homes. Neufangled Robotics’ Stringman is a cable-driven robot that installs in your ceiling corners and covers your entire room. It costs $795, runs LeRobot AI models, exposes an MCP server so your coding agent can control it, and every hardware design is published under OSHWA open-source certification. The Hacker News thread https://news.ycombinator.com/item?id=50354358 this week hit 830 points and 245 comments. For developers who have been watching the robotics space from a safe distance, Stringman is the first actual reason to get in. Four Cables, No Wheels, Room-Scale Workspace Stringman is a cable-driven parallel robot https://en.wikipedia.org/wiki/Cable robots CDPR . Four anchors mount in the ceiling corners, each with a NEMA 17 stepper motor that spools and unspools wire to move a suspended two-finger gripper through 3D space. No balancing, no rolling base, no articulated arm segments — just geometry and tension. The workspace scales with your room: up to 5m × 5m × 3m. Payload is 1.9 kg average across the full area. Traverse speed is roughly 0.4 m/s. Three Raspberry Pi Zero 2 W units handle distributed control — one per anchor pair, one in the gripper — while a central host machine runs vision and decision-making. The trade-off is worth naming: CDPRs require a clear path between anchors. Add a couch, an open door, or a dog, and the cables snag. In the HN thread, a PhD-level commenter confirmed the kinematics are sound while also identifying the workspace collapse problem in real homes. The design accepts this constraint and bets that some spaces — a playroom, a small workshop, a maker lab — are constrained enough to work. That bet is what makes the $795 price point possible. MCP Server: Your AI Agent Can Drive It The motion controller exposes an MCP server. That means any AI coding agent can connect, watch the robot’s cameras, send movement commands, and load new Python behaviors into the running controller — the same interface pattern you use to give an LLM access to a web API or a file system, now working for a physical actuator in your ceiling. Three control modes are available: - Gamepad: Direct teleoperation; used to record imitation learning datasets - Python maneuver API: Write policies as code using the nf-robot package - MCP server: Connect an AI agent and let it reason about the physical space The Python package is on PyPI https://pypi.org/project/nf-robot/7.6.2/ . Install with: pip install --upgrade "nf-robot host,dev " Version 7.6.2 shipped November 13, 2026 and is actively maintained. Train Your Own Robot Policies With LeRobot Stringman is a supported platform in Hugging Face’s LeRobot framework https://github.com/huggingface/lerobot — the open-source robotics library often called “Transformers for robotics.” The practical workflow: 1. Teleop the robot with a gamepad while the UI records episodes 2. Upload the dataset to Hugging Face or keep it local 3. Train a policy on Google Colab using provided notebooks 4. Deploy the trained policy back to Stringman Supported models include π0.5, JEPA, and Multitask DiT — the same architectures the robotics research community uses. LeRobot-tagged datasets on Hugging Face grew from roughly 1,000 to 58,000+ in a single year, making robotics the largest dataset category on the Hub. The infrastructure for sharing robot policies already exists. Stringman gives developers an affordable hardware target to contribute to it. Fully Open Source, Runs Air-Gapped OSHWA certification covers the full stack: mechanical CAD, PCB designs, firmware, and control software. Buy the kit from neufangled.com https://neufangled.com/stringman or pull the plans and build your own. The firmware repository is public on GitHub. The system runs on local Wi-Fi with no cloud dependency. No subscription, no data leaving your network — which is the right default when cameras are watching your living space. What This Actually Signals The commercial humanoid robot landscape in 2026 has 33 robot families; 29 are enterprise or research only. The one accepting home deposits starts at $30,000+. Developers who want to experiment with physical manipulation have had no practical entry point under a few thousand dollars. Stringman does not solve all of that. Neufangled is explicit that it is not yet a reliable home appliance. But it is the first device that combines sub-$1K price, open-source everything, MCP integration, LeRobot support, and local-first operation. That combination is new. AI agents are getting progressively better at manipulating files, calling APIs, and navigating the web. Physical manipulation is the next layer, and right now most of that work sits behind expensive hardware and proprietary stacks. Stringman is a small but concrete step toward the agent-to-physical frontier being accessible to any developer with a room to spare.