OpenMonoAgent.ai

OpenMonoAgent.ai is a beta, open-source coding agent from StartupHakk LLC that runs the model and the agent on your own hardware. It pairs a .NET CLI with a bundled llama.cpp inference server and is installed with a single shell command. It also ships a VS Code and Cursor extension and mobile apps for directing a self-hosted inference server.

What It Is

OpenMonoAgent is a local-first coding agent used from the terminal through a full-screen TUI or a classic scrolling mode. The installer detects the hardware (NVIDIA GPU, AMD APU, or Apple Silicon) and picks a suitable Qwen model. The agent works on your project through a built-in toolset, with an approval prompt before destructive operations, and its code stays on your machine.

How the Agent Works

• The agent runs in a Docker sandbox in which the project is mounted as a workspace. Host-level actions go through approved calls only.
• It has 21 built-in tools according to the README (the homepage says 20), plus extension through Model Context Protocol servers.
• Five specialist sub-agents (Explore, Plan, Coder, Verify, general-purpose) run in isolated sessions with locked tool sets and turn budgets.
• Code intelligence comes from Roslyn for C# (type hierarchy, blast radius, callers, diagnostics) and LSP support for languages such as TypeScript, Python, Go and Rust.
• Playbooks are YAML workflows with typed parameters, conditional gates and checkpoint/resume.
• Sessions persist as JSONL transcripts with auto-compaction and snapshot-based file undo.

Deployment and Editor Options

A dual-box mode lets the agent run on a laptop while inference runs on a separate GPU machine, using an outbound tunnel with no port forwarding. Vision input, private web search through SearXNG and anti-bot scraping through Scrapling are optional additions. The VS Code extension offers agent chat, Plan/Build modes, permission management and @ file references, and also works in Cursor. Local llama.cpp is the default provider; OpenAI, Anthropic and Ollama backends are listed as work in progress. The vendor also offers installation and support of private deployments.

Current Status

The project is labeled beta (public beta). It is licensed under the GNU AGPL-3.0 according to the repository's license file.