PLAIground: SLO-Driven Runtime Model Selection for Compound AI Systems in the Edge-Cloud-Space Continuum

A new framework called PLAIground enables runtime model selection for Compound AI systems operating across edge, cloud, and space environments, according to research posted to the arXiv preprint repository on June 12, 2026 [1][2]. The framework addresses a gap in existing distributed and compound AI frameworks, which do not natively support dynamic model switching during workflow execution [2]. Compound AI systems combine multiple tasks—such as object detection, time-series analytics, and natural language processing—and must satisfy stringent Service Level Objectives, or SLOs, for accuracy, latency, and cost [2]. PLAIground introduces a Compoundable AI Model abstraction, called CAIM, that decouples task semantics from AI model implementations using Task and Data Contracts, allowing models to be switched without workflow changes [2]. The framework also includes Pixie, an SLO-driven runtime model selection algorithm that dynamically selects the most suitable model for each task during execution [2]. In an evaluation on two realistic Compound AI workflows, Pixie achieved up to 91.3% accuracy while maintaining SLO compliance [2]. By contrast, fixed-model strategies either violated cost and latency budgets by a factor of up to 21 or missed accuracy targets by 4% [2]. The research appears on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives about 24,000 submissions per month and is not peer-reviewed [6]. The paper is listed under the Computer Science category for Distributed, Parallel, and Cluster Computing [1]. arXiv also hosts a framework called arXivLabs, which allows community collaborators to develop and share experimental tools on the platform, such as bibliographic explorers and code finders [4][5]. arXivLabs was formalized in 2020 to provide a conduit for community innovation while setting guidelines for third-party collaborations that align with arXiv’s values of openness, community, excellence, and user data privacy [4]. The PLAIground paper’s abstract page includes links to several of these Labs tools, including Bibliographic Explorer and Connected Papers [1].