Projects

Work worth pointing to.

These are the public projects I want the site and the X profile to point at when I talk about agentic AI, local models, and harnesses that have to survive reality.

multiharness

multiharness is the main proof point: a local-first agentic harness that can use local and cloud models in the same workflow.

  • routes work to the right model instead of pretending one model fits everything
  • keeps failure modes visible instead of burying them behind a polished demo
  • makes the harness the thing you can inspect, replay, and trust
  • case study explains the routing layer in plain language

ds4

I also got it working with antirez/ds4, which is a good reminder that local models get interesting when the workflow is real enough to test.

  • local inference is only useful when the loop is fast enough to matter
  • the model label matters less than how legible the failure path is
  • good harnesses make model choice a practical decision, not a belief system