Most of the effort in AI goes into the machine. Far less goes into the person on the other end.

Agents, harnesses and evaluations are well covered. The context they would need to actually help is not. Danbi takes more of it in, and keeps all of it on your own machine.

See what actually happens

·✓across the tools you already use

Notice what keeps repeating

·✓you did that by hand six times — automate it?

Take it off your hands

·✓once you have said yes enough times

Keep it from rotting

·✓your work changes, so the software has to

What it sees

How you actually work, across the tools you already use. The coach keeps one record of it, because anything built to help you without one is guessing.

Browser, phone and audio, all feeding that record.
Facts worked out by code; the model only comments.
Local only, for privacy and security.

What it does

Software you describe rather than shop for. The coding agent turns what you want into working code, checks the result, and waits on your call.

Open models, running on your own machine.
Nothing lands unless you say so.
Re-checked later, and reverted if it drifts.