Fat-Cat: Document-Driven Metacognitive Multi-Agent System for Complex Reasoning

TL;DR

Fat-Cat introduces a document-driven multi-agent system that enhances reasoning efficiency by using Markdown-based state representations and specialized modules, outperforming traditional JSON-based approaches and even GPT-4o on complex benchmarks.

Contribution

It presents a novel document-driven architecture with a Semantic File System, Textual Strategy Evolution, and a Closed-Loop Watcher, improving reasoning performance over existing rigid state representations.

Findings

Outperforms GPT-4o on HotPotQA benchmark.

Document-based state modeling significantly outperforms JSON-based approaches.

Empirically validates the importance of semantic, document-driven state management.

Abstract

The effectiveness of LLM-based agents is often limited not by model capacity alone, but by how efficiently contextual information is utilized at runtime. Existing agent frameworks rely on rigid, syntax-heavy state representations such as nested JSON, which require models to devote a substantial portion of their limited attention to syntactic processing rather than semantic reasoning. In this paper, we propose Fat-Cat, a document-driven agent architecture that improves the signal-to-noise ratio of state management. By integrating three key components: (1) a Semantic File System that represents agent state as Markdown documents aligned with common pre-training corpora, (2) a Textual Strategy Evolution module that accumulates task-solving knowledge without parameter updates, and (3) a Closed-Loop Watcher that monitors reasoning trajectories to reduce hallucinations. Extensive reasoning, retrieval, and coding benchmarks, Fat-Cat consistently improves agent performance. It enables the Kimi-k2 model to outperform the proprietary GPT-4o baseline on HotPotQA. Replacing the document-based state with JSON leads to performance drop, while empirically validating the critical necessity of document-driven state modeling over rigid syntax. The code is available at https://github.com/answeryt/Fat-Cat.