Long conversations stay responsive
Earlier continuations and capacity headroom keep chat snappy in larger workspaces.
Long conversations now stay responsive even as a workspace grows.
Earlier conversation continuations
Reflection work orders now spawn continuations earlier in the conversation, before the prompt grows large enough to feel sluggish. A long-running thread that was reaching very large prompts before continuation now stays snappy turn over turn.
Reliability: API tier capacity bumped with relaxed liveness thresholds so transient slow responses no longer trigger restarts during peak load. Templates page and Canvas Collectors resolve on demand for mirrors with many instances. Bare artifact IDs in tool calls now resolve to the right handler. OAuth logins no longer accidentally link to submirror accounts. The ref gateway respects the Redis size cap with transparent fallback for larger payloads. New users see group-shared artifacts immediately on first login.