Research into domain-specialized models for narrow, high-value tasks, with constrained tools, versioned inputs, and measurable operating boundaries.
Research & development · Not commercially available
Give a general foundation model tools and it becomes a principal with enormous leverage, running on an engine no one can fully predict. This work focuses on narrow tasks with defined inputs and outputs, so their operating boundaries can be tested.
The aim is a model tuned to the language, policies, and data of one industry, rather than a general model guessing at that world from the public internet. Narrow scope, deep competence.
The proposed design runs each model in a dedicated, isolated tool environment with scoped, short-lived permissions, and decomposes complex work across specialized models coordinated by an orchestrator that follows tested patterns instead of improvising.
Controlled inference settings and versioned environments are intended to make behavior reproducible and loggable. Fisher would test these models the same way it tests any agent, so claims about their behavior can be checked against evidence.
Small Models is part of mace AI’s Deep Model Trust research and development. It is not commercially available.
Explore the research → Assess your current agent with Fisher