This update addressed a core usability problem in long-running agent collaboration: too much reporting makes the important work harder to see.
Reports Became Less Noisy
Agent reports are now treated more like informational content and less like hidden requests for confirmation.
- Reports no longer need to create extra planning-chat threads just to be visible
- Navigation badges can focus more tightly on asks that actually need action
- Report delivery is summarized more aggressively before it reaches the user
- Telegram updates avoid repeating heavy prefixes when the content itself is what matters
The goal is a calmer signal: when the agent needs you, it should be obvious; when the agent is merely reporting, it should not flood the same attention channel.
Planning Chat Uses A More Compact Memory Layer
Planning chat now uses a ReMe-backed memory layer to carry forward useful context without reloading every detail into every turn.
That helps long-running planning conversations stay grounded while reducing unnecessary repetition. The memory path also gained diagnostics and caching improvements, so recall issues are easier to investigate when context does not show up as expected.