# Stateful Inference for Low-Latency Multi-Agent Tool Calling

arXiv: https://arxiv.org/abs/2605.26289 · HTML: https://mlsanity.com/paper/2605.26289/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 5 of its 7 references that are in the map.
- First posted on arXiv 25 May 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 5 of 5)

- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Fast Inference from Transformers via Speculative Decoding](https://mlsanity.com/paper/2211.17192/) (2022)
- [SGLang: Efficient Execution of Structured Language Model Programs](https://mlsanity.com/paper/2312.07104/) (2023)
- [DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving](https://mlsanity.com/paper/2401.09670/) (2024)
- [Attention Once Is All You Need: Efficient Streaming Inference with Stateful Transformers](https://mlsanity.com/paper/2605.13784/) (2026)

## Built on by (most cited first, 1 of 1)

- [Speculative Pre-Positioning: Decoding Stateful Sessions to the Next Decision Point Off the Critical Path](https://mlsanity.com/paper/2606.29565/) (2026)

## Abstract (arXiv)

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(nt) per-turn cost of conventional serving into an O(Deltat) 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× faster per turn on a 6-turn agentic workflow and 4.2× 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.