Saved in:
Bibliographic Details
Main Author: Norgren, Victor
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.26289
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914602183294976
author Norgren, Victor
author_facet Norgren, Victor
contents Multi-agent tool calling is becoming the dominant interaction pattern for LLM-based systems, yet existing inference frameworks treat each tool call as an independent request, re-processing the entire conversation from scratch even though 85-95% of the prompt is unchanged from the previous turn. We present a stateful inference architecture that converts the $O(n_t)$ per-turn cost of conventional serving into an $O(Δ_t)$ delta-only cost: a persistent KV cache lives across turns and advances by ingesting only the new tokens, while a radix prefix cache extends this across interleaved multi-agent traffic and a prompt-lookup speculative decoder accelerates structured output. Against vLLM and SGLang on novel, fully-generated workloads, the reference implementation is $2.1\times$ faster per turn on a 6-turn agentic workflow and $4.2\times$ on the median turn of a 35-turn one, halving end-to-end wall time. The advantage comes from stateful reuse and speculation, not caching.
format Preprint
id arxiv_https___arxiv_org_abs_2605_26289
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Stateful Inference for Low-Latency Multi-Agent Tool Calling
Norgren, Victor
Machine Learning
Multi-agent tool calling is becoming the dominant interaction pattern for LLM-based systems, yet existing inference frameworks treat each tool call as an independent request, re-processing the entire conversation from scratch even though 85-95% of the prompt is unchanged from the previous turn. We present a stateful inference architecture that converts the $O(n_t)$ per-turn cost of conventional serving into an $O(Δ_t)$ delta-only cost: a persistent KV cache lives across turns and advances by ingesting only the new tokens, while a radix prefix cache extends this across interleaved multi-agent traffic and a prompt-lookup speculative decoder accelerates structured output. Against vLLM and SGLang on novel, fully-generated workloads, the reference implementation is $2.1\times$ faster per turn on a 6-turn agentic workflow and $4.2\times$ on the median turn of a 35-turn one, halving end-to-end wall time. The advantage comes from stateful reuse and speculation, not caching.
title Stateful Inference for Low-Latency Multi-Agent Tool Calling
topic Machine Learning
url https://arxiv.org/abs/2605.26289