The article introduces VISTA, a training-free layer designed to address the context window limitations of long-horizon tool agents by exposing their internal state. It argues that frontier models are blind to their own context usage and proposes an interface that surfaces working memory details rather than relying on learned compression policies.
- VISTA represents working memory as typed, addressable blocks and provides a runtime dashboard showing per-block token usage, recency, and access history.
- The system archives blocks as recoverable full-fidelity payloads without requiring model training.
- On LOCA-Bench, the interface improved four backbones, lifting Gemini-3-Flash performance from 22.7% to 50.7%.
- Performance gains increase with context pressure and transfer across million-, 100K-, and 10K-scale trajectories on LOCA-Bench, BrowseComp-Plus, and GAIA.
This approach allows models to make informed keep-or-drop decisions by providing visibility into their own context state, addressing the gap left by previous system-controlled management methods.