This is why we need open-source harnesses + local models

i've been thinking about this more after trying different agent setups. the model isn't the only thing that determines how well an agent performs. The harness around the model matters a lot too.
With a managed agent setup, you're often giving up control over things like the agent loop, context management, tool execution, retries, and state.

That's fine when you just want something that works. But if we want to actually optimize agents, I think both parts need to be open:

Open-source model + open-source harness.

With local models, you control the model and where the inference happens.

With an open-source harness, you control what happens around the model.

That gives you room to experiment with things like:

how the agent decides what to do next

how much context gets passed to the model

how tools are executed

when to retry or stop

how state is maintained

which model to use for which task

already seeing this separation become more important, nvidia's sol-pi is an interesting example

and i think we're going to see even more optimization happen at the harness/runtime layer, not just at the model layer.

are you running local models with an open-source harness, or do you still prefer managed agent setups?