Transparency. The harness is the layer between you and the models. We don’t want it to be a black box. You can read the prompts, the tool definitions, and the agent loop and see exactly what we’re doing.It also keeps us honest. If we make a bad call somewhere, you can point at it.
The harness is open. The evals, inference, and provider handling are not.The harness only supports the Magnitude Provider. It’s not designed as a generic shell you plug other providers into. That’s the whole point of the product.
We accept PRs and issues on GitHub. Bug reports are most useful. Feature work is welcome too, but the product direction is opinionated, so we may say no to changes that don’t fit.Join the Discord if you want to talk through an idea before opening a PR.
