LLM Agents Are Latent Context Managers: Eliciting Self-Managed Context via a Proprioceptive Dashboard

2026-06-29

AI Summary

This work addresses the challenges faced by long-context agents, which often lose critical information due to context inflation or rely on opaque external mechanisms that hinder self-awareness of their internal state. To overcome these limitations, the authors propose VISTA, a training-free, model-agnostic middleware for context management that represents working memory as typed, addressable blocks. VISTA introduces a runtime “proprioceptive” dashboard that dynamically visualizes each block’s token consumption, recency, and access history, thereby revealing— for the first time—that large language models can autonomously manage their own context. Experimental results demonstrate that VISTA substantially improves performance across LOCA-Bench, BrowseComp-Plus, and GAIA benchmarks; for instance, it boosts Gemini-1.5-Flash’s score on LOCA-Bench from 22.7% to 50.7%. The gains intensify under higher context pressure and generalize well across diverse models.

Abstract

Long-horizon tool agents are bottlenecked by how their context grows toward the limits of the context window. Recent systems make context management agent- or system-controlled, but they either learn a compression policy that discards evidence or manage context in a layer the agent never sees. We argue both leave a more basic gap unaddressed. Frontier language models are proprioceptively blind to their own context. From the prompt alone they cannot see how large, how old, or how used each block is, the signals a keep-or-drop decision needs. We hypothesize that competent context management is already latent in capable models, and that what is missing is not a learned policy but an interface exposing this state. We introduce VISTA (Visible Internal State for Tool Agents), a training-free, model-agnostic layer that represents working memory as typed, addressable blocks, surfaces a runtime dashboard of per-block token usage, recency, and access history, and archives blocks as recoverable full-fidelity payloads. On LOCA-Bench, BrowseComp-Plus, and GAIA, the same untrained interface transfers across million-, 100K-, and 10K-scale trajectories. On LOCA-Bench it improves four backbones and lifts Gemini-3-Flash from 22.7 to 50.7%. The lift grows with context pressure and transfers across backbones. Ablations further confirm that the dashboard matters beyond archive and recovery tools.

Problem

• context management
• long-horizon agents
• context window
• proprioceptive blindness
• working memory

Innovation

• context management
• proprioceptive dashboard
• LLM agents
• working memory
• model-agnostic interface