Local models improve cloud-written agent programs by keeping data private
Should SLMs be in the Driver's Seat? A Study of Local Models as Interpreters in Cloud-Written Agent Programs
Artificial IntelligenceDistributed, Parallel, and Cluster Computing
Summary
Big AI models run on the cloud and can solve tasks very well but need to see private user data. Smaller models running on the user's device keep data private but don't perform as well. The authors studied a middle ground where a big cloud model writes a program that uses a small local model to answer certain questions at runtime without sharing data. This approach makes the program more successful while keeping user data on the device.
What this means in practice
- •For mobile app developers: Build apps that run AI-generated programs locally while keeping user data private using small models for runtime decisions.
- •For privacy engineers: Create systems where sensitive user data stays on device yet leverages cloud-composed agent logic to improve accuracy.
Authors
Boris Radovič, Marco Canini, Veljko Pejović
Abstract
The most accurate agents executing user-provided tasks are driven by large, cloud-hosted models, which solve them by reading the private data of the user. Agents powered by models small enough to run on edge devices keep that data local, but yield far worse results. A third option is to have cloud models write a program that devices execute: the data stays local, but as the program is written before any data is seen, it fails whenever a step depends on interpreting content that appears only at run time. We explore an intermediate architecture, in which the cloud-written program defers bounded questions over user data to a small local model through an ordinary function call. On AppWorld this improves the success rate of a code-only agent in all six combinations of backbone model and test partition we evaluate, by up to 14.9pp, recovering a substantial part of the gap to a cloud-hosted ReAct loop while keeping user data on the device.