0.24.4 The phone sends, agents read and send, idle apps wake

Owner, 2026-11-04: moved here from 0.24.3 when the owner chose the cut-down 0.24.3 ("Yes, cut-down version"). Plan: side notes 2026-11-04 re-plan and review.

The phone page sends like the computer's page (#866, step 3), after the owner approves its design: pairing by QR code from the computer's page, a credential per phone, the hub delivering each send, an allow-list, and local network only until 0.25.
Agents read the other sessions' progress and send to each other through the board: #721 (the MCP server, PR #842), #723 (setup adds the MCP entry), #317 and #869. Agents get a JSON read, not the text board.
Idle apps wake: #722 (board_wait); #318 (Codex idle at once); #858 (the Claude desktop Code tab: check e.surface, then receive); #686 (OpenCode, DeepSeek and OpenClaw, when they can be verified).
Setup writes into another agent's config (Codex trust, agy allow rule) only after a consent design: the preview names the file and the exact rule, there is no wildcard, and uninstall removes it (#812, #774).
The remaining usersim FRICTIONs, #401, #402, #511, #851, #852 and #803.

Navigation Menu

0.24.4 The phone sends, agents read and send, idle apps wake

page: keep the notification switch, side pick, theme and filter across restarts without reusing the port

test(messaging): two messaging tests fail on and off on Windows, on main too

feat(messaging): an agent sends to another agent with --send, labelled as not from the person

feat(messaging): Send to Codex — a Codex CLI session on the shared daemon receives a board message, idle or busy

fix(pane): show an error only when the person must act, in plain words with no file paths; heal the rest quietly

feat(messaging): in-process SDK live receivers for OpenCode, DeepSeek Harness, and OpenClaw

feat(messaging): board_wait and --wait-message park on the store and wake the host when a message arrives (#685, part 2 of 3)

feat(setup): connect and setup add the agent-board MCP entry for ten hosts (#685, part 3 of 3)

perf(windows): a setup preview took 75 s once, and the board and page opened over their 5 s targets

test(messaging): one live run on each mainstream AI, recorded for the support table

docs(readme): over SSH the board command prints the board once instead of the live board the Quick start promises

test(usersim): judge every release for a CLI user and a UI user, per AI, by the steps they take

docs(messaging): the support table lists surfaces as a person sees them, one row each

feat(phone): the phone page sends like the computer's page, and personal agents (Muse, dot) connect through it

feat(retention): keep 7 days of history in hand-kept tasks, short steps, 3 months of archives, and a board that stays readable with many sessions

feat(page): the first screen says at a glance whether anything needs you, and whether the page is current

feat(plan): Cursor's todo list reaches the board as the session's steps

docs(video): a 45-second README cover that sells the board, with MP4s for social media

perf(windows): measure what agent-board leaves running or growing over long use, and whether it slows git

review(ux): evaluate two feature ideas, then walk the board as a new user against the principles

perf(stability): what agent-board adds when a computer is overloaded, how it falls back, and how it recovers after a crash

docs(images): the README, cover, pictures and social preview show the new page and the Bubble logo, and hook a first-time visitor within 45 seconds

test(page-open): "the progress open command and the pane page button request a browser" timed out on plugin (ubuntu-latest), and the job hit its 15-minute limit

test(journeys): L1 latency "a new ask is visible within 3 s" failed once on Windows and passed on rerun

feat(messaging): Codex without the daemon gets a board message when its current step ends

Issues