TECHNICAL REPORT / IN PROGRESS

An on-device agent harness for Android.

Method, benchmark results, ablations and failure cases. Published when measured.

The study behind the system
Concept illustration of separate work and review panels with an inspection lensConcept artwork · Illustration, not product footage
010203
01Method02Ablations03Failure cases

The report we’re working toward

The planned technical report will cover the exact model and harness settings, repeated runs, cost and latency, completion-check and procedure ablations, failure cases and reproducible scripts.

MobileWorld evaluation results are not yet published. We have not announced a publication date.

Get research updates ↗