AgentOCR: Reimagining Agent History via Optical Self-Compression
This addresses a bottleneck in deploying agentic systems for tasks like ALFWorld and search-based QA, offering incremental improvements in efficiency.
The paper tackles the problem of rapidly growing textual histories in agentic systems, which inflate token budgets and memory usage, by introducing AgentOCR, a framework that represents history as compact images and uses segment optical caching and agentic self-compression to reduce token consumption by over 50% while preserving over 95% of performance.
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.