1.31.0 — several sessions at once, files made for you, live token counts

Several sessions at once: when a second session is active, a row of tabs appears under the header of the live call, with the state of each one. Click a tab to move the call there; it stays there (📌) until you click its own tab again. When another session stops for you (a question, an edit that waits for your go-ahead, or its turn is over), a visitor leans in at the edge of the call, says which one, and takes you there on a click.

In VS Code the call goes to the Claude Code tab you switch to, so a conversation and its call can sit side by side. A tab does not pin the call: it stays with that conversation for a minute and a half and for as long as it is active, then follows the latest activity again.

Files made for you: when a turn ends, the files the session created for you show in one line under the call: those the lead names, else the documents, not every source file it wrote. A document made by a command (an export, a report written by a script) shows too when the lead names it. A click shows the file in its folder; nothing is read and nothing is run.

The lead's turn ends when Claude Code says so: it is no longer shown as done, with "your turn", each time it thinks for a while. It has a word each time its turn ends, and one when the agents it waited for are all back.

Token counts are live again: during a live call the ⚡ counts on the tiles and in the header stood still until a reload. They now follow the session, and show in the camera of the agent who shares its screen and on each line of the agents list, where an agent whose tile has left the stage keeps its line and its count.

Fixed: after an error message from Claude Code (the API is down, the limit is reached) the lead looked busy for three minutes; a session stopped with Esc read "working" for ten minutes; with Claude Code 2.1.87 and older a session could read "your turn" in the middle of a turn.

Each of the three settings (the visitor, following the Claude Code tab, the files line) is in ⚙ → Display. On 74 real sessions (69,000 events) this version reads everything else exactly as 1.30.0 did.

Full list: CHANGELOG

1.30.0 — Mistral Vibe sessions, and a lead called Margot

Mistral Vibe sessions too: sessions of Mistral Vibe, Mistral's coding agent, appear next to your Claude Code and Codex sessions, in the session list, on the board and in the live call. Nothing to set up: the app reads the folders Vibe already writes in ~/.vibe/logs/session and never writes there. You see what you typed, what the agent says, each command with its real output, the files it changed, its reasoning, the token totals, and its sub-agents, which join the call with their orders and their answers. Both of Vibe's engines are read, each checked against real sessions of Vibe 2.26.1. 45-second video (en français).

Leave it out if you prefer: the VS Code setting agentStandup.vibe, or AGENT_STANDUP_VIBE=off.

The lead is now called Margot, whatever tool runs the session. It was "Claude" in a Claude Code session and "Codex" in a Codex one.

Fixed: a character pack name that is not a pack answered with a server error. Found by @thesubr00t (#7).

Nothing changes for Claude Code and Codex sessions: on 62 real sessions (60,000 events) this version answers exactly as 1.29.0 did, the lead's name aside.

Full list: CHANGELOG

1.29.0 — your own agent names, and French voices that sound French

Your own agent names: a sub-agent you defined yourself (a file in .claude/agents/, for example code-reviewer.md) or named when it was launched now keeps that name on its tile, in the transcript and in the voices: "Code Reviewer" rather than a first name drawn from the list. The agent types Claude Code ships (general-purpose, Explore, Plan…) still get a first name, and ⚙ → Characters → Keep my agents' own names turns it off. Asked for by @Eshinio (#6).

Fixed:

• The slider in Replay and in the demo did nothing when clicked or dragged. It works now.
• French lines with the very natural voices were all read by one system voice, the first in the list. They are now read by your system voices, one per character, French from France first. The speech model itself only has English voices.
• Natural voices on a Mac no longer pick Apple's synthetic voices (Eddy, Flo, Reed, Sandy, Shelley) unless nothing else is installed.
• The first page after the app starts no longer waits a second or more for the list of system voices.
• Header: the status pill is no longer shortened, and the counters lose their words instead of being cut mid-word in a narrow window (seen in Spanish, Portuguese and German).

Also: Live, Replay and Demo are translated in the eight languages that have a catalogue.

Full list: CHANGELOG

1.28.0 — the interface in Korean (한국어)

Korean (한국어): the interface is available in a tenth language: menus, chips, tooltips, the recap, and in VS Code the command titles, the setting descriptions and the status bar. Pick it in ⚙ → Language. Contributed by @jainefer (#5), thank you! Like the other languages apart from English and French it is an AI draft not yet reviewed by a native speaker: corrections are welcome.

Also: the website and the README suggest the very natural voices for the best experience (⚙ → Voice style, a one-time install).

Full list: CHANGELOG

1.27.0 — Codex sessions too, next to Claude Code

Codex sessions too: sessions of OpenAI's Codex (the desktop app and the codex command) appear next to your Claude Code sessions, in the session list under "Codex", on the board, and in the live call when one of them is the most recent. Nothing to set up: the app reads the files Codex already writes in ~/.codex/sessions and never writes there. You see what you typed, what the agent says, each command with its real output, the files it created or changed, web searches, MCP calls, the questions it asks you, its reasoning summaries as inner thoughts, token counts and the end of each turn. Not yet for Codex: its sub-agents (a Codex call is the lead and you) and pictures. To leave Codex out: the VS Code setting agentStandup.codex, or AGENT_STANDUP_CODEX=off. Also fixed: clicking the status bar in VS Code to open the call on a given session showed an empty page (since 1.20.0).

Unofficial: not affiliated with Anthropic or OpenAI.

Full list: CHANGELOG

1.26.0 — a new icon: the lead, in front of a window at night

A new icon: the app icon is now the lead as you see it in the call (the peach character with the gold headset, its focused smile, its desk and its coffee), in front of a window at night. It replaces the old icon in the extension list, the desktop app and the website, and the camera emoji in the header and the browser tab. The lead's room has a window: the three screens with rising curves behind the lead looked like a trading desk, and are replaced by a window on the city at night with a few lit windows that twinkle slowly. The VS Code side bar symbol is now the lead's head with its headset. The desktop app's menu bar or tray indicator has its own small drawing, sharp on high-density screens. For contributors: the drawings are in brand/ with a short guide, and npm run brand renders every icon, the link preview and the README header from them.

Full list: CHANGELOG

1.25.10 — inner thoughts tidied, a wrap-up made for the ear, natural voices keep long lines whole

Inner thoughts: the “N calls in” thought is gone, an agent's own reasoning is shown in whole sentences (it was cut at every dot), a thought sits above the activity line and no longer makes the narrow list jump, and the awkward lines are fixed (doubled punctuation, long file names, regular expressions read out). The wrap-up is made for the ear: “To sum up” now says a few plain sentences in about 18 seconds, without headings run into sentences, list dashes or tables. Very natural voices no longer drop the end of a long line: the model takes about 500 characters per call and the rest was cut without a word; a longer line is now made in parts and joined. Full list in CHANGELOG.

1.25.9 — natural voices recover from a cut download, starter pack no longer shipped

Very natural voices (Kokoro): a model download cut short left a file that failed to load every time after, and the natural voices were gone until the models folder was deleted by hand; the file is now dropped and fetched once more. Contributed by @neondragon (#3). The starter pack is no longer shipped: ⚙ → Characters offers the drawn characters (default), the cat pack and your own folder. Plus everything from 1.25.8: inner thoughts for agents at work, and your own lines always voiced. Full list in CHANGELOG.

1.25.8 — inner thoughts, and your own lines are always voiced

Inner thoughts: in a live call, an agent that has been at work for a while thinks to itself in a lighter bubble that is not said out loud. Its own reasoning comes first, sentence by sentence, when the session keeps it; otherwise the thought is about what it is on (the command it waits for, the files it goes between, its to-do item, a failure, its task). About one in five is mumbled, low. Your own lines are always voiced: an order typed during a long command was shown without a voice because it looked minutes behind; it now takes its place when the lead gets it, and a line of yours is never skipped nor muted. Open VSX is published by the release workflow itself. Full list in CHANGELOG.

1.25.7 — a live call that loses nothing, the lead's real end of turn, readable bubbles, steadier voices

A live call no longer loses what the session writes out of order: orders you type while the lead works, sub-agents' reports (✅, a ding, sometimes thanks) and every tool call of a message. The lead's turn ends when Claude Code says so, not after 25 quiet seconds: no more “your turn”, confetti or “To sum up…” in the middle of a turn. Speech bubbles sit above the tile's own badges; narrow panels get one row per agent and one status line that says who needs you, who is working, or that it is your turn. Voices: a line ends when its sound ends (no more talking over the next speaker), “and/or” is read in full, no voice switch on a short line. Chatter between agents is Normal by default again. Sub-agents launched by workflows are part of the call. Fixes for the lead's answer after a long order, a repeated last line after pause, and a request that could stop the standalone server. The live call is now tested in a real browser on every push. Full list in CHANGELOG.