OpenMono
A free, open-source terminal coding agent that runs entirely on local hardware with a bundled llama.cpp inference server, auto-configuring itself for NVIDIA GPU, CPU or Apple Silicon with no subscription or per-token billing.
OpenMono is an open-source coding agent that runs entirely on the user's own hardware, pairing a .NET 10 CLI with its own bundled llama.cpp inference server so there is no subscription, no per-token billing, and no code or context leaving the local network after a one-time setup. It auto-detects GPU, CPU or Apple Silicon (Metal) hardware and installs the matching model, runtime and Docker containers with one command.
Its agentic loop runs up to 25 iterations per turn, aborting if the same tool sequence repeats three times in a row, checkpointing at 65 percent context fill and compacting at 80 percent. Every tool call passes through a 12-step pipeline of schema validation, path checks, a plan-mode guard, a capability check, caching, hooks, execution and artifact storage, with read-only tools run in parallel, and five specialist sub-agents, including a read-only exploration agent and a no-write planning agent, run in isolated sessions with their own tool sets and turn budgets.
It reports about 45 tokens per second on GPU and about 20 on CPU, is written in C#, and is currently in beta.