{ "schema_version": 1, "builder_plan": { "scenario": "Long customer conversations exhaust memory or limit concurrent requests.", "change": "Change only the attention cache policy, after checking model and runtime support.", "baseline": "The same model with its ordinary attention cache.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Peak cache memory, long-context answer quality, first-token latency, and requests completed.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.28577", "title": "Cache Engineering for Agents", "summary": "It replaces context-engineering approaches (trimming or summarizing history, which invalidates the prefix and forces re-prefill) with direct KV cache operations, drop and repos, that control active cache without altering the trajectory. The cost is operational: you must run or integrate the AKV serving engine to manage stateful cache reuse, eviction, and restoration, which excludes third-party API-only deployments and adds a serving-layer dependency. Failure modes include incorrect cache reuse from history dependence (identical token sequences with different cache states), potential accuracy loss if aggressive drop policies remove needed context, and policy fragility since gains depend on the workload-specific drop policy.", "rationale": "The cost and latency gains are quantified and relevant to agent workloads on a single 24 GB GPU, but adoption requires self-hosting a modified serving engine and validating the drop policy on the reader's own trajectories, so pilot testing rather than immediate adoption is warranted.", "technique": "Direct KV cache drop and reposition operations for agents", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-20", "as_of": "2026-10-21", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.28577", "independent_implementations": 2, "total_implementations": 4, "stars": 142, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "It replaces context-engineering approaches (trimming or summarizing history, which invalidates the prefix and forces re-prefill) with direct KV cache operations, drop and repos, that control active cache without altering the trajectory. The cost is operational: you must run or integrate the AKV serving engine to manage stateful cache reuse, eviction, and restoration, which excludes third-party API-only deployments and adds a serving-layer dependency. Failure modes include incorrect cache reuse from history dependence (identical token sequences with different cache states), potential accuracy loss if aggressive drop policies remove needed context, and policy fragility since gains depend on the workload-specific drop policy.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.28577", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Rolling-drop cache policy reduces average cost by 27.7% versus threshold-triggered summarization while improving benchmark scores by 1.7 points; the AKV engine adds up to 1.2x output throughput and 1.5x P95 latency speedup.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.28577", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }