Finding / agent-zero

2026-10-13-agent-zero-responses-histories-retain-cache-boundaries-and-providers-use-native-routes

Responses histories retain cache boundaries and providers use native routes. Completed local Responses turns retain masked extras at their original history boundaries instead of discarding cacheable prefixes as conversations grow. Replay eligibility and secret masking remain, and extras count toward compression. This is distinct from v2.13’s Chat Completions cache fix; Chat Completions remains the fallback when no Responses alias is configured. Nebius, Venice chat, Z.AI/Coding and OpenRouter embeddings move to native LiteLLM routes while documented overrides/fallbacks remain. Upstream cache-hit and startup percentages are upstream measurements, not independently reproduced results here.

Channel: tagged-release. Half: both. Date: 2026-10-13 (first stable; implementations 2026-10-07).

Operator consequence: Re-test long Responses sessions and affected custom/gateway providers. Inspect effective overrides and actual cached-token usage before claiming a cost reduction. Native provider routing removes some provider-specific scaffolding, but a general dependency update still needs compatibility checks.

Receipt

Finding metadata

Run: 2026-10-13-profile-refresh-2026-09-21_2026-10-13-frontier-v0

Finding ID: 2026-10-13-agent-zero-responses-histories-retain-cache-boundaries-and-providers-use-native-routes

Profile citations

  • Agent Zero / claim / responses-cache-and-native-routes

Source links

Primary links, including exact changelog lines when available.