Dot digs deeper into 'why' questions, monitors your data in the background, and turns chat feedback into reviewable proposals.

Deeper answers to your "why" questions

Knowing what changed is useful; knowing why is actionable. Asking why something moved - why revenue dropped, what's driving churn - is usually what turns a chart into a decision.

Dot's always been able to do this kind of key driver analysis, but it now does the work much more thoroughly: exploring high-dimensional spaces it would have skimmed before, and backing each finding with statistical confidence. So when Dot identifies a key driver, that conclusion has been pressure-tested against the rest of your data, not just the first plausible answer.

Schedules become a background agent

Set up a scheduled report with a work gate and a result gate and Dot will monitor the situation for you in the background. The work gate is a natural-language pre-check ("only run if new orders arrived today") that skips the agent run when nothing has changed; the result gate ("only deliver if revenue dropped more than 5%") suppresses delivery when the answer isn't worth telling. Tell Dot what to watch and how to define an interesting change, and it'll only ping you when something matters. See the scheduling docs for details.

Tell Dot to fix itself

During any chat you can now say "remember this", "the orders table was renamed", or "this metric is wrong" - and instead of silently updating, Dot replies "Proposal submitted" and queues a draft change for an admin to review. Admins see pending proposals on the History tab with a counter pill, open a full diff of what would change, then merge or reject with one click. You can also kick off a proposal from any past chat by clicking "Investigate" - Dot reads the conversation and proposes a fix to its own knowledge base.

Also new: Notion, Jira, and Confluence are now live data sources alongside GitLab and Bitbucket, plus TOTP-based MFA with org-level enforcement, per-group connector access controls, and Cursor IDE works as an MCP client.