Get an answer when a model provider stumbles
Chat turns now retry through temporary provider outages and finish with follow-up suggestions, whichever way the work is routed.
You ask Chief a question, wait, and the turn dies with an error or lands with no follow-up chips under it. Both of those had causes you could not see, and both are fixed.
Retries on provider outages
A chat that hit a temporary outage at the model provider used to end the turn with an error straight away. Chief now recognises those failures for what they are and tries again, so you get your answer instead of a dead turn.
The same applies to a failure the provider reports partway through a response. Chief used to record it as a success and carry on quietly; it now stops, reports the real reason, and backs off that model until it recovers.
Long reasoning answers that keep going
Some chats broke mid-answer when the model paused to reason between steps of a longer task. Chief now sends the model's own reasoning back exactly as it arrived, so those turns run to completion.
Follow-up suggestions on every answer
Follow-up suggestion chips appeared under some answers and not others, with nothing to explain the difference. You now get them under every answer.
Turns that no longer die on connected tools
A turn could crash with no response at all while Chief checked whether your Slack, Notion and Google connections were reachable. Those checks now report an unavailable connection correctly, and the turn answers.
Tool calls also failed when Chief's own storage was slow to respond. That path no longer takes the turn down with it.
Nothing to turn on. Ask your next question and the turn should hold together where it used to break.