Self-GC: Self-Governing Context for Long-Horizon LLM Agents
2026-07-01 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors address the problem of managing a large, complex memory in AI agents that use language models for long tasks. Instead of just trimming text based on simple rules or summaries, they propose Self-GC, a system that carefully tracks and controls pieces of memory like user inputs and tool results as separate objects. This approach helps decide what parts to keep or remove without losing important information, improving efficiency and reliability. Their tests show that Self-GC removes more unnecessary data while maintaining task quality better than traditional methods.
Large Language ModelsContext ManagementGarbage CollectionTool-augmented AgentsMemory PruningSelf-GovernanceContext ObjectsHeuristicsTask ContinuationLifecycle Management
Authors
Xubin Hao, Hongjin Meng, Xin Yin, Jiawei Zhu, Chenpeng Cao
Abstract
Long-horizon LLM agents accumulate tool results, files, plans, and user constraints that are too structured to be treated as a disposable text suffix. Current systems mostly rely on in-run heuristics such as chronological pruning and tool-output masking, or on final self-summary near a context limit. Heuristics are cheap but blind to future dependencies; summaries preserve narrative state but often hide exact evidence, locators, and editable artifacts. We present Self-GC, where GC denotes self-governing context while deliberately echoing garbage collection: the system does not merely reclaim unused tokens, but governs the lifecycle of agent context objects. Self-GC turns user turns, tool spans, and skill state into indexed objects; asks a side-channel planner to propose fold, mask, and prune actions; and lets the harness enforce recoverable sidecars, safe commit boundaries, and cache-aware commit. On a 33-session Hard Set, Self-GC prunes 43.95% of prefix tokens while leaving 84.85% of future continuations unaffected, compared with no-impact rates of 54.55% to 69.70% for heuristic baselines. On a 332-session production-derived suite, three planner backbones reach no-impact rates of 91.27% to 94.58%, while baselines remain at 77.71% to 87.46%. In production, an online account-level split reduces daytime average input tokens by 10% to 15%, with peak reductions near 20%. These results point to context management as runtime lifecycle control over indexed, recoverable objects rather than post hoc text cleanup.