Assistant Connector Reliability
Your assistant no longer loses sight of the integrations you've already connected — and when something goes wrong upstream, it says so plainly instead of guessing.
If you asked the assistant to do something with a connected integration and were told it "isn't connected," that answer was wrong — and we've fixed the cause. A single slow or failed internal lookup of your workspace's available tools could quietly drop the whole list, leaving the assistant working from a small default set for the rest of the conversation. From the outside it looked identical to a workspace with no integrations at all, which is exactly why the message was so convincing and so misleading.
That lookup is now resilient by design. It has a firm time limit and retries automatically when a hiccup looks temporary, while failures that clearly won't resolve on their own stop immediately instead of stalling you. The tool list is also refreshed at the start of each turn rather than pinned once at the beginning of a session, so a single bad moment can't follow you through a long conversation. And because the retry only applies to reading the list — never to actions the assistant takes on your behalf — nothing gets executed twice.
When every attempt does fail, the assistant now tells you the truth. Instead of pretending the integrations don't exist, it reports that your connectors are still connected and that tool access is temporarily unavailable, so you know to try again in a moment rather than going off to re-authorize something that was never broken. Behind the scenes, those degraded moments now raise a dedicated alert to our team, so we see them as they happen instead of learning about them from a support thread.
None of this adds any waiting to a normal request — if anything, the new time limit puts a ceiling on a hang that previously had none. A small styling correction on a marketing page rode along in the same change. The result is an assistant you can take at its word: when it says something is connected, it is, and when something's temporarily off, you'll hear that instead of a guess.
Trust in an assistant comes down to whether it tells you the truth about what it can see. That's the standard we're holding it to.