AgentOCR: Reimagining Agent History via Optical Self-Compression

2026-01-08

arXiv.org

AI Summary

This work addresses the high token cost and memory overhead in multi-turn interactive agents caused by rapidly growing textual history. To mitigate this, the authors propose compressing the observation-action history into compact images, leveraging the high information density of visual tokens to reduce the language model’s input burden. A segmented hashing cache is introduced to avoid redundant rendering, and a reinforcement learning–driven adaptive compression mechanism enables the agent to autonomously adjust its compression rate, jointly optimizing task performance and computational efficiency. Evaluated on ALFWorld and search-based question answering tasks, the method maintains over 95% of the original performance while reducing token consumption by more than 50% and achieving a 20× speedup in rendering.

Technology Category

• Multiagent Systems: Agent Communication
• Search and Optimization: Learning to Search
• Machine Learning: Reinforcement Learning

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction trajectories, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token budgets and memory usage. We introduce AgentOCR, a framework that exploits the superior information density of visual tokens by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, results demonstrate that AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Our further analysis validates a 20x rendering speedup from segment optical caching and the effective strategic balancing of self-compression.

Problem

• agent history
• token efficiency
• memory usage
• textual history
• scalability

Innovation

• AgentOCR
• optical self-compression
• segment optical caching
• visual token representation
• compression-aware reward