English

Atomix: Timely, Transactional Tool Use for Reliable Agentic Workflows

Machine Learning 2026-02-17 v1 Artificial Intelligence Distributed, Parallel, and Cluster Computing Multiagent Systems

Abstract

LLM agents increasingly act on external systems, yet tool effects are immediate. Under failures, speculation, or contention, losing branches can leak unintended side effects with no safe rollback. We introduce Atomix, a runtime that provides progress-aware transactional semantics for agent tool calls. Atomix tags each call with an epoch, tracks per-resource frontiers, and commits only when progress predicates indicate safety; bufferable effects can be delayed, while externalized effects are tracked and compensated on abort. Across real workloads with fault injection, transactional retry improves task success, while frontier-gated commit strengthens isolation under speculation and contention.

Keywords

Cite

@article{arxiv.2602.14849,
  title  = {Atomix: Timely, Transactional Tool Use for Reliable Agentic Workflows},
  author = {Bardia Mohammadi and Nearchos Potamitis and Lars Klein and Akhil Arora and Laurent Bindschaedler},
  journal= {arXiv preprint arXiv:2602.14849},
  year   = {2026}
}
R2 v1 2026-07-01T10:38:41.078Z