Demonstrating integration of openhands with OpenVINO Model Server

This proposal aims to create a production-ready, single-command deployment integrating the OpenHands AI coding agent with the OpenVINO Model Server (OVMS) on Intel AIPC hardware. The project resolves three critical friction layers: the API routing mismatch between OVMS (/v3) and OpenHands (/v1), the failure of structured tool-call execution which leads to model hallucinations, and system crashes caused by memory exhaustion from concurrent Docker sandboxes. By implementing a Docker Compose manifest with shared bridge networking and host Docker socket passthrough, the architecture offloads heavy LLM inference to iGPU or NPU targets while keeping orchestration on the CPU. Key technical deliverables include configuring native tool-guided generation for models like Qwen3-Coder-7B, enabling prefix caching to reduce multi-turn latency by over 50%, and enforcing strict container memory limits to ensure stability on 8 GB and 16 GB RAM profiles. The project concludes with a formal usability gap analysis and a finalized deployment setup to provide a seamless, local, and private AI development environment