StartupHakk/OpenMonoAgent.ai

What it solves

OpenMonoAgent is a local-first coding agent designed to eliminate reliance on cloud subscriptions, per-token billing, and the privacy risks of sending code to external servers. It provides a full agentic loop for software development that runs entirely on the user's own hardware.

How it works

The system combines a .NET 10 CLI with a bundled llama.cpp inference server running inside Docker. It auto-detects hardware (NVIDIA GPU, CPU, or Apple Silicon) to configure the optimal model. The agent operates through a multi-step pipeline that includes schema validation, capability checks, and an artifact store, utilizing a set of 20 built-in tools. It employs five specialist sub-agents (Explore, Plan, Coder, Verify, and general-purpose) to handle different stages of the development process with specific turn budgets.

Who it’s for

Developers who want a private, zero-cost AI coding assistant that they own and control entirely on their own machine, including those using VS Code or Cursor.

Highlights

• Local-first Inference: Bundled llama.cpp ensures zero-cost tokens and complete data privacy.
• Deep Code Intelligence: Integrates Roslyn for type hierarchy and symbol search, and LSP for multiple languages (TypeScript, Python, Go, Rust).
• Docker Sandboxing: All file operations are restricted to a /workspace mount to limit the blast radius.
• Specialist Sub-agents: Uses isolated sessions for exploration, planning, coding, and verification.
• Distributed Inference: Supports running the agent on a laptop while the inference occurs on a separate GPU machine via an outbound tunnel.
• Multimodal Capabilities: Supports vision via a multimodal projector (mmproj) for reading images and screenshots.
• Private Web Access: Self-hosted search via SearXNG and anti-bot scraping via Scrapling and Camoufox.
• Editor Integration: Dedicated VS Code and Cursor extension for a seamless sidebar experience.