Horizon Code
A local-first coding harness. Typed tools, bounded agent loops, and a verification gate before the work is called finished.
Independent studio / Human-directed, machine-assisted
Local AI, useful software, and machine-made media. A workshop for turning new capabilities into things worth using.
Explore the workModels on your own hardware. Harnesses, inference, and the boundaries that make them usable.
Small, inspectable software. Tests that tell us what happened, not what the agent said happened.
Creative systems for moving images and sound. Human direction, explicit inputs, deliberate assembly.
On the workbench
A local-first coding harness. Typed tools, bounded agent loops, and a verification gate before the work is called finished.
A show engine that treats sound and picture as things to assemble carefully: frozen inputs, explicit timelines, and checked outputs.
One model, one code-golf problem, one independently judged program. A small test bench for inspecting agent results rather than trusting the transcript.
Graphic studies by the studio. These are visual identities, not software screenshots or measured results.
The name is an ambition. Not a claim.
Automation is useful when you can see what it did. We build, test, inspect, and keep the human decisions visible.
How we work ↗Earlier lab notes ↗