Skip to main content
ADE 1.2.109 makes long chats, project opening and background sync noticeably lighter, and makes sync to unreachable machines give up sooner.

​
Speed

  • Opening a very long chat drops from 1,174 ms to 94 ms (12.5× faster). Across five real chats the average open falls from 259 ms to 36 ms (−86%).
  • A turn that touched thousands of files no longer stalls the chat (#1711).
  • New chats appear without waiting on git (#1701).
  • Older chat history pages load about 2.3× faster (#1692: 2,551 → 1,107 ms over a 54,000-event backfill).
  • The chat composer stops re-rendering on every streamed event: renders per event 249 → 181, script time per event −12% (#1691).
  • Chat windows trim old events in steps instead of one per update (#1690).
  • Closed dialogs stop re-rendering while a chat streams: dialog renders over 300 streamed events drop 502 → 185 for the handoff dialog and 158 → 102 for the remaining dialogs (#1715, #1717).
  • Every ade command starts about 145 ms faster (657 → 511 ms) by reusing Node’s compile cache for the CLI bundle (#1718).
  • Opening a terminal or editor, or checking sign-in, no longer freezes the app for about a second while the login shell re-runs (#1709: ~1,050 ms → 1 ms).
  • Credential lock waits no longer freeze the background service (#1706: 183 ms blocked → 0).
  • Recent projects summarize 50× faster (#1661: 25.3 → 0.5 ms per pass).
  • Account settings sync stops asking the relay when nothing changed: background relay requests drop about 20× (2,880 → ~144 per day per machine) (#1682), and remote changes still apply as they arrive (#1696).
  • Background cleanup of automation history drops from 150 statements / 241 ms to 3 statements / 6.3 ms per 50 relay events (#1708).
  • The sync poll wakes about 5× less when idle (#1674), and the roster’s warm build is 38% faster (#1712).
  • The Your Macs card stops polling while nothing can see it (#1667).
  • Lane file counts are cached by HEAD and index state (#1707), and live durations share one timer (#1660).

​
Sync & multi-machine reliability

  • A sync to a dead or unreachable host gives up in 5 s instead of about 75 s, and stale hosts are reclaimed (#1694).
  • Opening a project waits out a peer’s database lock instead of failing on it (#1693).
  • Account sync and sync status finish on slow links (#1705).
  • A project that follows another machine is reported as such, instead of timing out (#1685).
  • Recovery controls stay available when sync activation fails (#1695), and a failed sync-host activation stops re-running for every caller (#1664).
  • Sync-host activation no longer waits on the status snapshot (#1680).
  • Sync status reads wait for the first readiness check (#1697).
  • Shared project brains check their identity before use (#1698).
  • A loopback probe timeout no longer reads as a shadowed port (#1672).
  • When pairing stalls, the machine and route are named (#1703), and Repair appears beside the sync-host outcome that explains it (#1710).

​
Fixes

  • Search no longer wedges the background service, and chats can be re-parented and keep their links across a project move (#1716).
  • Account settings and local sync status can no longer hang the app for 30 s (#1666, #1671).
  • Opening a project reports why it failed and groups errors by the real cause (#1678).
  • Moving a project releases old log files, moves proof paths, and carries closed lanes’ Claude history (#1677).
  • Progress notes no longer raise false “Needs you” hands in chat (#1679).
  • Branch import works from the web client (#1687).
  • Lane actions accept the lane-id argument they document (#1673).
  • A turn that settles after shutdown no longer reads the move store once it has closed (#1669).
  • A missing folder or file no longer counts as an allowlist violation, and a missing path outside the allowlist no longer reveals that it is missing (#1668, #1670, #1702).
  • The Mac desktop driver keeps retrying after a transient crash instead of giving up, and stops retrying once nothing wants it; explicit screenshots get a longer capture budget (#1681, #1704).
  • Runtime endpoint ownership is verified before repair, and a service-owned endpoint is no longer mistaken for another copy of ADE (#1699, #1676).
  • Teardown stops the producers before it disposes the pool (#1675).

​
Features

  • ⌘-chords for the Work page, with a shortcut sheet on ? (#1686).
  • Sign in to an MCP server once, for every agent (#1688).

​
Diagnostics

  • Error log lines name the failure rather than the event (#1684), and Error values in log details include their message and stack (#1700).
  • Expected shutdown states no longer open Sentry issues (#1683), and Sentry noise is filtered out with run_ade_action errors in their own bucket (#1663).