I built a harness and now it is my daily driver for local agents

I tried a lot of open source harnesses, such a pi, oh my pi, open code, prime agent. Wrote a whole extension kit for pi. none of them really completely satisfied me for the way I work and none of them really felt like they were truly built for local first.

So like everyone else I built my own agent runtime + harness. It’s zero bloat, event driven, cache friendly, and supports concurrent tool calls. Out of the box it ships with 18 native tools including a persistent python kernel, extensible plugin system, scoped memory + todos, and a scheduler for deferred jobs. subagents are supported through the delegation tool, which supports parallel fan out of tasks over the number of slots available on your gpu.

I have been dogfooding it for the past month and once I had the core + loop written I built most of the rest of the harness from within the harness, lol. It is now my daily driver for my local agents (I drive the whole setup with tailscale + termius in tmux sessions from my workstation). Plus I think the tools I work with should be fun to use.

Currently ships with OpenAi compatible provider for local models with other providers planned. Currently only CLI and TUI frontends are built. Binary available for installation on Linux/MacOS (I don’t have windows sorry). Open source, MIT license. Open to feedback, issues on the GitHub welcome.

Rig, a minimum runtime for your agent.