Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Self-GC: Self-Governing Context for Long-Horizon LLM Agents

Source

arXiv

arxiv.org/abs/2607.00692v1

Read the source

Source

Plain-Language Brief

A research paper titled 'Self-GC: Self-Governing Context for Long-Horizon LLM Agents' presents a system called Self-GC that aims to manage the lifecycle of context objects for long-horizon language models (LLMs). The system indexes user turns, tool spans, and skill states, and uses a side-channel planner to propose actions for folding, masking, and pruning. This leads to reduced input tokens in production, according to the paper.

What the Document Says

  • The system is called Self-GC and aims to govern the lifecycle of context objects for long-horizon LLMs.
  • Self-GC indexes user turns, tool spans, and skill states.
  • A side-channel planner is used to propose actions for folding, masking, and pruning.
  • The harness enforces 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.
  • 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%.

Key Concepts Mentioned

  • Self-GC
  • long-horizon LLMs
  • user turns
  • tool spans
  • skill states
  • side-channel planner
  • folding
  • masking
  • pruning
  • recoverable sidecars
  • safe commit boundaries
  • cache-aware commit

Claims Made by the Document

  • Self-GC prunes a significant percentage of prefix tokens while leaving a high percentage of future continuations unaffected.
  • In production, Self-GC reduces daytime average input tokens by 10% to 15%, with peak reductions near 20%.

Evidence Used

  • On a 33-session Hard Set, Self-GC prunes 43.95% of prefix tokens while leaving 84.85% of future continuations unaffected.
  • 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%.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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.