Memory Compression for High-Fanout Agent Sandboxes
Mengming Li, Ceyu Xu and colleagues (HKUST) build AgentZip, a memory compression system for agent workloads that spawn many concurrent sandboxes from a shared template, and cut sandbox memory by up to 8.7x.
Ask this paper
Observation: Sandboxes from one task start from the same template and run related trajectories, so their memory has large template-relative and cross-sandbox redundancy.
How to compress: AgentZip exploits similarity across non-identical pages, both relative to the template and across sibling sandboxes.
What and when to compress: Any page with a profitable representation is eligible, overhead is controlled by prefetching at restore time instead of conservative page selection, and expensive compression runs while the agent waits on the LLM.
Results: Up to 8.7x reduction in sandbox-owned memory against 2.1x for the Linux configuration. Prefetching and agent-aware scheduling cut the slowdown of aggressive compression from 3.1x to 1.40x.
Abstract
High-fanout agent workloads create a growing memory bottleneck because a single task may spawn many concurrent sandbox sessions. Yet these sandboxes are far from independent: they originate from a shared template and execute related trajectories, exposing substantial template-relative and cross-sandbox memory redundancy. Conventional memory compression is poorly matched to this setting in three fundamental dimensions: how to compress, because they fail to exploit similarity across non-identical sandbox pages; what to compress, because they control page-fault overhead through conservative page selection; and when to compress, because compression is either triggered by memory pressure or performed without awareness of agent execution phases. We present AgentZip, the first memory compression system designed specifically for AI-agent sandboxes. AgentZip introduces compression mechanisms that exploit both the template-relative and cross-sandbox redundancy. It broadens the compression scope to any page with a profitable representation and shifts overhead control from compression-time page selection to restore-time prefetching. It further aligns expensive compression with LLM waiting periods to avoid interfering with foreground tool execution. Across LLM training and inference workloads, AgentZip reduces sandbox-owned memory by up to 8.7x, compared with 2.1x for the Linux configuration. Restore prefetching and agent-execution-aware scheduling reduce the slowdown of aggressive compression from as high as 3.1x to 1.40x while retaining nearly all of its memory-saving benefit.