AgentR A Stateful and Recovery-Aware Software Architecture for LLM-based Auditable Workflows
Source
arxiv.org
Author
Riya Samanta, Bidyut Saha, Soumya Kanti Ghosh, Rajkumar Buyya
Date
Context
AgentR treats an AI literature review as work that must survive interruptions. Its proposed architecture saves the research intent, generated searches, paper assessments and gap analysis as persistent records. Background workers track processing stages and retry failed jobs, making it possible to inspect what happened and recover work after a failure. Architecture abstract.
The prototype reportedly completed 99.2% of jobs at the language-model stage. That is an operational measure of jobs finishing, not a measure of whether the scientific conclusions were correct. The contribution is a recoverable, auditable workflow around model calls, with intermediate work available for inspection.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.