Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Smart-TCP: An Agentic AI-based Autonomous and Adaptive TCP Protocol

Yule Han, Kezhi Wang, Kun Yang

TL;DR

This work treats TCP logic as an autonomous decision-making task by embedding an LLM as a cognitive core and a deterministic ALU as a calculation tool, connected through a dual-agent Client-Server framework. By decoupling reasoning from arithmetic, Smart-TCP overcomes the 32-bit computation limits of pure LLMs and achieves high fidelity in state prediction, error detection, and end-to-end sessions. A retrospective state reconstruction dataset and structured JSON interfaces enable supervised fine-tuning (LoRA/SFT) of the cognitive core, yielding superior performance against several baselines. The results demonstrate the feasibility of AI-native transport protocols with robust error handling and reliable data transfer, suggesting significant practical impact for intelligent, autonomous networks.

Abstract

The Transmission Control Protocol (TCP) relies on a state machine and deterministic arithmetic to ensure reliable connections. However, traditional protocol logic driven by hard-coded state machines struggles to meet the demands of intelligent and autonomous network architectures. Here, we adopt the agentic AI-based paradigm, driven by Large Language Models (LLMs), characterized by context perception, autonomous reasoning, and tool use. Based on this, we propose Smart-TCP, which re-imagines TCP's core control logic as an autonomous agent. Specifically, the proposed architecture employs a context aggregation mechanism to synthesize the protocol context, utilizes the LLM for autonomous logical reasoning, and invokes an Arithmetic Logic Unit (ALU) as a tool for computation. Furthermore, we establish a dual-agent interaction framework based on this architecture and implement TCP protocol interactions. Experiments demonstrate that the Smart-TCP agent excels in static prediction and error detection, achieving a 93.33% success rate in end-to-end sessions. These results strongly validate the technical feasibility of an agentic AI-based TCP protocol.

Smart-TCP: An Agentic AI-based Autonomous and Adaptive TCP Protocol

TL;DR

This work treats TCP logic as an autonomous decision-making task by embedding an LLM as a cognitive core and a deterministic ALU as a calculation tool, connected through a dual-agent Client-Server framework. By decoupling reasoning from arithmetic, Smart-TCP overcomes the 32-bit computation limits of pure LLMs and achieves high fidelity in state prediction, error detection, and end-to-end sessions. A retrospective state reconstruction dataset and structured JSON interfaces enable supervised fine-tuning (LoRA/SFT) of the cognitive core, yielding superior performance against several baselines. The results demonstrate the feasibility of AI-native transport protocols with robust error handling and reliable data transfer, suggesting significant practical impact for intelligent, autonomous networks.

Abstract

The Transmission Control Protocol (TCP) relies on a state machine and deterministic arithmetic to ensure reliable connections. However, traditional protocol logic driven by hard-coded state machines struggles to meet the demands of intelligent and autonomous network architectures. Here, we adopt the agentic AI-based paradigm, driven by Large Language Models (LLMs), characterized by context perception, autonomous reasoning, and tool use. Based on this, we propose Smart-TCP, which re-imagines TCP's core control logic as an autonomous agent. Specifically, the proposed architecture employs a context aggregation mechanism to synthesize the protocol context, utilizes the LLM for autonomous logical reasoning, and invokes an Arithmetic Logic Unit (ALU) as a tool for computation. Furthermore, we establish a dual-agent interaction framework based on this architecture and implement TCP protocol interactions. Experiments demonstrate that the Smart-TCP agent excels in static prediction and error detection, achieving a 93.33% success rate in end-to-end sessions. These results strongly validate the technical feasibility of an agentic AI-based TCP protocol.

Paper Structure

This paper contains 28 sections, 2 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Agent Framework Overview
  3. Context Aggregation
  4. Reasoning and Computation
  5. Segment Assembly and State Update
  6. Agent Implementation
  7. Formalization of Agent Tasks
  8. Dataset Engineering
  9. Data Sourcing and Filtering
  10. Dedicated Error Dataset
  11. Retrospective State Reconstruction
  12. Agent Interfaces and Role Definition
  13. Structured Cognitive Input
  14. Decoupled Decision and Tool Invocation
  15. Agent Role Definition via Prompting
  16. ...and 13 more sections

Figures (2)

  • Figure 1: Dual-Agent Interaction Framework based on Smart-TCP Agents
  • Figure 2: Fine-Grained Field-Level Accuracy