Chat, tasks and questions

At the top of the sidebar, a row of buttons opens the pages that gather every agent’s work: New project, Bookmarks, Replies, Questions, Tasks and Schedules. On a phone the same buttons sit at the right of the top bar.

Questions from agents

Agents keep their questions for you until you answer them, so none is lost in a long chat.

In the chat: a question shows as a card with its choices, a box for your own answer, and Dismiss. Once answered, withdrawn or dismissed, the card says so and still shows the choices it offered, the one chosen marked.
Answering: a choice or your own words go to the agent as your message, quoting the question. In the chat your answer shows as a card with the question and your choice or words. Dismiss tells the agent the question was closed without an answer.
Withdrawn: an agent can withdraw a question that no longer applies.
In the chat header: the Questions button counts what waits for your decision from this agent: its open questions and its proposed definition changes. It opens a dialog with two tabs. Waiting lists them, oldest first, each with its answers, or with Approve and Reject for a definition change, and a link to its card. Past lists what was answered, withdrawn, dismissed or decided, newest first.
Questions page: the Questions button at the top of the sidebar, with its count, opens a page listing every open question of every active project’s agents, grouped by project and agent, with the same answers and a link to each in its chat. A second tab lists the last 100 closed ones. Proposed definition changes are there too, with Approve and Reject, and so are permission prompts waiting in a chat, with a link to answer them there. A row of tabs (All, then one per agent) narrows the list. An agent’s count in the sidebar opens this page on that agent’s tab.

Unread replies

Each administrator sees how many of the agents’ replies they haven’t read yet.

What counts: only an agent’s replies to you (what it posts to you, and the last text of each of its turns), not what it writes along the way, its thinking, tools, sub-agents, ProjectStart’s notes or messages between agents.
When a reply is read: in the chat, once at least half of it (or most of the screen, for a long one) has stayed in view for a second in a visible browser tab. Each reply counts on its own.
Counts: the Replies button at the top of the sidebar shows your total; each agent’s row shows its own in a soft pill; a folded project group shows its total. The browser tab’s title starts with the total, as “(3)”.
Replies page: lists every unread reply in full, oldest first, each with its time and a link to its place in the chat. Tabs: All, then one per agent. Mark read marks one and scrolls to the next; Mark all read marks an agent’s. Replies marked there stay listed, dimmed, until you open the page again. Showing a reply on this page never marks it read by itself.
In the chat header: the Replies button opens a dialog with Unread (with Mark read, Show in chat and Mark all read) and All (every reply, newest first).
In the chat: a chat with unread replies opens at the first of them, under a New replies line, each marked New. While there are unread ones, an Unread replies pill with their count and ↑ ↓ takes you to the previous or next one.
On phones: the top bar’s Replies button shows the total, and the agent’s name in the chat header carries the unread count of the other agents.

Reactions

You can react to an agent’s message with an emoji: a small smiley shows when you point at the message (on a phone, in the long-press menu) and offers 👍 👎 ❤️ 🎉 😄 😕 👀 🙏. Tap one to add it, tap it again to remove it.

The agent learns about your reactions with your next message to it, never on their own, and may take them as feedback (a 👎 as a sign to check what it said).

Allow agents to use emoji, in a project’s Settings (off by default), decides whether its agents may put emoji in what they write.

Tasks

Agents record each request you make as a task. The chat header’s Tasks button opens the agent’s task list, with three tabs: Open, Completed and Timeline. Every task has a number, shown as “Task 12”; agents name their tasks that way, and you can too. In a chat, “Task 12” in a message is a link that opens the task.

Blocked tasks

An agent marks a task Blocked only when it can’t go on until you act: an answer, a decision, access, a machine to test on. A task that waits on another task, a sub-agent or a job stays being worked on.

In the task list, a blocked task carries a yellow Blocked mark and the reason under its request.
The chat header’s Tasks button shows the number of blocked tasks in a yellow badge at its bottom right, beside the open count at its top right.
Starting, pausing, finishing or cancelling the task clears the mark.

Start and pause

Start: each open task that isn’t being worked on and isn’t blocked has a Start button: the agent is asked to start it now.
Pause: each task being worked on has a Pause button: the agent is asked to stop at a safe point, keep what is done, and mark the task paused, saying where it stands.
Either shows in the chat as a card (Task requested or Pause requested). The task’s status changes only when the agent changes it.

Timeline

The Timeline tab is a chart of the agent’s tasks over time:

One row per task: a thin line from when it was asked to when it ended, bars for when the agent worked on it, and a mark for how it ended. A Status menu chooses which statuses show.
Under each task, the sub-agents that worked for it, as thinner bars numbered #n; ones not tied to a task are under Other sub-agents. A failed or stopped one is faded with a cross.
A row at the top counts how many were working at once.
It shows the last 24 hours, or one day you pick (‹ › for the day before and after).
Clicking a task or bar opens its details: its times, time worked and paused, tokens and cost, its sub-agents, and links to the request and to the task.

Tasks and Schedules pages

Two pages show what every agent is on and what will run, for every active project, grouped by project and agent, each agent with its state and a link to its chat. A row of tabs (All, then one per agent) narrows the list, and a line at the top sums it up.

Tasks has Open (tasks being worked on first, then blocked, paused and not started, each with its status, kind, line, time worked, and Start or Pause) and Completed (the last 100 done or cancelled).
Schedules has Active (still to run, paused ones too, soonest first, each with what it tells the agent, when it runs, its runs so far and the next one, with Pause or Resume and Delete) and Done (the last 100 one-time jobs that ran).

Schedules are timers and recurring jobs you ask an agent to set; at its time the agent gets the message.

Bookmarks

Bookmark sits under each message you can reply to (beside its time, Copy and Reply) and under an agent’s thinking. A bookmarked message shows a filled mark; pressing it again removes the bookmark.
Files have Bookmark too, in the chat’s Files dialog, which also has a Bookmarked filter.
The Bookmarks button at the top of the sidebar opens your bookmarks, most recent first, each with its project, agent, time and a link to its place, and Remove.
In a chat, the header’s Bookmarks button opens the same list for the project, with a choice of All or one agent, and Show in chat.
Your bookmarks are your own. A project’s bookmarks are hidden while it is in the Trash.

Notifications

The installed app (or the browser) can notify you on a device, even when ProjectStart isn’t open, when an agent needs you.

What notifies: a new question, a definition change waiting for your decision, an agent’s reply, and usage: a Claude or Codex login reaching 80%, 90%, 95% and 100% of one of its usage windows. Tapping a notification opens the agent’s chat at the message (usage: the logins in Settings).
Nothing shows on a device whose app is open and in front; a usage step shows there as a short notice instead.
Turning it on: Settings › General › Notifications on this device: Turn on asks the browser’s permission; four switches choose the kinds (New questions, Definition change requests, Replies and Usage limits); Turn off stops them. On iPhone and iPad, notifications work only once ProjectStart is added to the Home Screen.
Each administrator’s devices are their own; signing out of a device ends its notifications.
The text is encrypted for the device, and no outside notification service or account is used.