OpenHands Agent Canvas puts coding-agent conversations and automations in one control center. A separate project, lampOS, connects configured agents to robot hardware. The projects target different workloads, but both show why an AI product often needs more than a model: tools to organize its work or connect it to the world around it.
OpenHands manages coding-agent sessions and automations
OpenHands Agent Canvas is a self-hosted developer control center for starting coding-agent conversations and automating recurring tasks. Its README says it runs the OpenHands agent and can connect other backends, including Claude Code, Codex, Gemini CLI, Pi and OpenCode. It describes workflows that run on a schedule or in response to webhook events, with integrations such as Slack, GitHub and Linear.
That makes the project relevant to teams that want one place to manage agent sessions and routine work while choosing among backends. The current README marks Agent Canvas as beta. The v1.27.1 release, dated Oct. 12, lists bug fixes and documentation updates rather than a new product launch.
The local setup requires Node.js 24 or later and uv; Docker is needed for per-conversation sandbox mode. OpenHands warns that a direct local installation gives the agent server access to the host filesystem. Its Docker mode isolates conversation agents, but the Canvas and automation services remain on the host. The repository's MIT license covers the project code; third-party agents and hosted backends have their own terms and costs.
lampOS connects agents to a robot's sensors and controls
lampOS, which calls itself Physical AI Operating System, runs on a robot and coordinates agent tasks with device hardware. Its README describes skills for cameras, microphones, speakers, motors, lights and sensors. The documented starting points include Lamp, Intern and Reachy Mini; a bring-your-own-robot path is also offered. Skills specify which hardware capabilities they need and use operating-system interfaces to access them.
Its Oct. 14 release, tagged models/ambernet, contains NVIDIA AmberNet FP32 ONNX weights and a label file for language identification. It is a model-assets release, not a full operating-system launch. The README also says model inference may use remote services, depending on configuration.
Hardware compatibility remains central to any trial. The repository's top-level license is Apache 2.0, but its README says the hal/ directory is GPL-3.0 and lists third-party components with separate licenses, including YOLOv8 under AGPL-3.0. A team evaluating the stack needs to check the code and services used for its specific device.
For a coding product, Agent Canvas addresses how people manage agent sessions and automations. For a robotics product, lampOS addresses how a configured robot exposes its hardware to agents. The checked project pages describe features and setup, not measured coding quality or field safety, so they are starting points for evaluation rather than evidence of reliable autonomous performance.
Sources
- OpenHands/OpenHands repository
- OpenHands Agent Canvas README
- OpenHands v1.27.1 release
- OpenHands license
- autonomous-ai/lampOS repository
- lampOS README and component-license notes
- lampOS models/ambernet release
- lampOS top-level license
Image: Humanoid Robot at International Exhibitions, by Om Umekar, licensed under CC BY-SA 4.0. The photograph is illustrative and does not depict hardware from either project.
