What shipped

For you: typing two words into search now returns results in roughly a third of the time it used to. On a 10,000-item workspace, that's 1–5 milliseconds — under the threshold of perception.

Under the hood: the new posting layout splits doc IDs, term frequencies, and field flags into parallel vectors, which the CPU prefetcher loves.

• P50 of 1.6ms, P99 of 4.8ms against a 10k-item community on iPhone.
• BM25-F scoring no longer needs to decode positions for filter-only queries.
• Existing indexes are upgraded in place on first launch — no resync required.

For you: when someone joins a channel, they can't read messages from before they were added. When they're removed, they can't read what comes next. The keys rotate automatically on every change.

Under the hood: MLS group key rotation is now automatic on member add/remove. Past messages stay readable only by the people who were in the group at the time they were sent.

• Adding a member rotates the group key before the new member sees any history.
• Removing a member drops them out of the next epoch — no stale access window.
• Built on OpenMLS (RFC 9420). The server never sees plaintext, ever.

For you: log an hour against a sub-task and the cycle burndown moves immediately. No more stale chart on Friday afternoon, no more spreadsheet to update by hand.

Under the hood: hours logged on issues are now reflected in cycle burndown charts in real time. Logging an hour on a sub-task immediately updates the parent's effort estimate and the cycle's remaining work line.

• WorkSlots aggregate up the issue tree without an explicit refresh.
• Burndown rendering is local-first; the chart updates on the device that logged the hour first.

For you: when a teammate joins a room or starts typing, you see it within a tenth of a second. Closer to "in the same room" than the multi-second lag most chat tools have.

Under the hood: presence and signalling moved from JetStream to Core NATS. Latency from change to peer-visible: under 100ms.

• Ephemeral state (typing, cursor, active room) no longer touches durable storage.
• JetStream remains the substrate for messages and durable state — Core NATS is just for "right now".
• The EphemeralStreamFeature handles the transport so individual features stay decoupled.

For you: if one issue is blocking three others (which are blocking five more), it shows up at the top of the blocker list automatically. Find the root cause without anyone configuring a query.

Under the hood: a new live view ranks blocking issues by chain length. If issue A blocks B blocks C blocks D, A floats to the top of the ledger.

• Computes transitively-blocked-issue counts on every blocker-edge change.
• Updates incrementally — no full re-scan of the planning graph.
• Available in any view that supports a "sort by blocker chain" key.

For you: two people editing the same issue description on different devices — both offline, both writing — both edits land cleanly when they reconnect. No "your version vs theirs" dialog.

Under the hood: issue descriptions are now edit-anywhere CRDTs. Two people editing the same issue offline merge cleanly when they reconnect — no last-write-wins, no merge conflict UI.

• Built on Loro for rich-text and tree-shaped CRDTs at a wire format small enough for mobile sync.
• Cursor position and selection follow the merged result, not the pre-merge view.
• Existing issue bodies are migrated lazily on first edit.