OpenMonoAgent.ai

About

Local-first coding agents usually stop at supporting whatever inference server the user happens to run, leaving setup as the user's problem. OpenMono bundles the whole stack: a .NET 10 CLI paired with a llama.cpp server in Docker that auto-detects NVIDIA GPUs, Apple Silicon, or plain CPU and ships Qwen model defaults tuned per backend, so one install script yields a working agent with no API keys and no network egress. Around the core it layers 20 built-in tools, five sub-agents, Docker-sandboxed execution, and Roslyn-powered code intelligence, with OpenAI/Anthropic/Ollama providers marked work-in-progress. A free relay at app.openmonoagent.ai enables distributed inference — the agent on a laptop, compute elsewhere — and iOS, Android, and VS Code/Cursor clients exist alongside the CLI. The project is a public beta under AGPL-3.0 with hardware demands (64 GB RAM recommended on Apple Silicon, 24 GB NVIDIA on Linux). Privacy-sensitive developers and offline environments are the audience.