Chat

Use Chat to decide what to delegate, develop a plan, and ask an agent to create the resulting work or Library document. The app calls this surface Chats because each conversation keeps its own history, with one selected agent handling each turn.

Start from Home

Home is where Task Machine opens after you sign in, and where the Task Machine logo in the sidebar takes you. It asks what you want to work on with the shared Chat composer.

Starter choices bring visible setup gaps forward, such as connecting tools or improving an agent's instructions. Choices remain available after setup is complete, and only choices without a relevant subject are hidden. Setup you cannot read does not count as missing.

Monday brings planning forward and Friday brings strategy forward, using your account timezone. Home can also offer a shortcut for a Project, Goal or Team you can read.

When recorded work is available, Home offers time and cost summaries for the Workspace or a relevant Project, Goal, Team or Agent. Summaries cover all recorded history unless you ask for a period. Without access to costs, the shortcuts ask for time only. They do not change a Budget or purchase anything.

Home shows the highest-priority starters in three rows below the input, using one, two or three columns as space allows. Choose Show more to browse the remaining suggestions in the scrolling area, or Show less to return to the top rows. Faded edges show where more choices continue.

Choose a starter to fill the editable draft, then adjust it before starting the Chat. Selecting a starter never sends a message or replaces a draft you have already started. The Chats drawer keeps its horizontal starter row and usual order.

A resolved @ mention appears as a named chip in your request and shows the subject's card above the composer. The mention keeps the exact subject selected, even when names repeat. Add or remove mentions to change the cards. Repeated mentions show one card, which stays beside your sent message.

Type @ to choose a Project, Goal, or Repository alongside the existing member, Team, Task, document, and Skill choices. Results respect your current Workspace access. A Repository card shows its name without connection details, and mentioning an item does not assign work or grant access. The one exception is the agent of a new Chat, described under Choose the agent for the next turn.

Several cards form a compact stack with a badge showing the total number of mentioned items. Hover over the stack, focus its badge with the keyboard, or tap the badge to expand it. The preview shows up to six cards, and every mention stays in your message. Existing Playbook and bottleneck attachments keep their own controls.

While your Workspace still has first steps to take, Home keeps a short checklist below the summary note. It lists Set up an agent, Start a chat or task, and Review a first result, and each step gets a check once its records exist, such as an active agent or a Task that reached Done. The top right of the card shows how many steps are done. Only the first open step has a button. It opens the agents settings, focuses the composer, or opens the Tasks list. Without an agent there is no composer to type in, so the checklist takes its place until you set one up.

The checklist shows only the steps your role lets you perform, so a member who cannot manage agents never sees the first one. It has no dismiss button and disappears once every step it lists is done. It returns if a step reopens, for example when the only agent is archived.

Home reads your workspace and shows only what it has to say, so a new workspace stays short and a busy one fills in. Below the composer, a green summary note from Task Machine says in a few sentences what needs you, what your agents are running and finished today, and how your Goals and routing are doing. Each number opens the view it counts, such as your Inbox, the Running now Tasks or Pulse. The note is written from your own data with fixed sentences, so it costs nothing and always agrees with the sections below it.

Under the words, the note shows the day's figures with their change: Tasks done and Usage as hourly charts, Tasks routed by team leads with how many you corrected, and the share of decisions your agents made without review. The figures cover today so far, compared with the same hours yesterday. During the first six hours of your day they cover yesterday instead, so the note never opens on a row of zeros. A figure appears only when you can read it and it has data.

Goals comes next when you can read active Goals. It shows up to three cards, each with the Goal's lead, its first linked Metric with the latest reading, its target and a small chart of recent readings. A Metric with a target value and date also shows whether it is On pace, Behind pace or has Reached its target, judged against a straight line from the reading at the time it was linked to the target date. A Goal without a Metric offers Add a metric, which fills the composer with a draft for you to edit before you send it.

Needs you previews up to eight Inbox items that need action, bringing items from the same source together. A source row shows its group review as a modal over Home, without opening the Inbox drawer. Individual items keep their ⋯ menu with actions such as Approve or Dismiss. Choosing the item opens its decision without leaving Home, and the arrows move through the rest of your Inbox. Open Inbox opens the drawer. Routing decisions wait in the Triage tab and are mentioned in the note rather than listed here.

Today so far lists up to eight changes from today in the same rows as the Pulse Feed, and Open Pulse opens the full Feed. A section that has nothing to show does not appear. Conversations stay in the Chats drawer and the sidebar, and the full list of running Tasks stays on the Tasks page.

Choose the agent for the next turn

A Chat runs one agent turn at a time. A participant can select another active, chat-ready agent from the workspace in the composer before sending the next message. Changing agents does not require a new Chat.

If the agent you need does not exist yet and you may manage agents, type its name into the agent picker of a chat, of the Chats panel or of Home and choose Create agent with the name you typed. The usual guided agent form opens above it with the name filled in. Once you create the agent, it is the agent of your next message and the message you were writing is still there. Closing the form creates nothing.

Task Machine supplies prior conversation context to the incoming agent, including a summary and recent turns as the conversation grows. The agent also reads the workspace knowledge it is entitled to, so the conversation can continue with a different specialist while keeping its history.

Starting a chat keeps the choice small: write the message and pick the agent. A new or empty chat starts with that agent's configured model and reasoning.

You can also pick the agent by mentioning it. In a new Chat, an @ mention of one agent selects that agent in the picker and locks the picker, so the agent you mention is the agent that answers. The @ list offers only agents that can start a Chat, and it stops offering agents once you have mentioned one, because a new Chat has a single agent. Remove the mention to make the picker editable again. It keeps that agent until you choose another. Starting a Chat that mentions several agents asks you to mention only one.

An existing chat keeps its latest available model choice, falls back to the agent's current default when that choice is no longer available, and uses Auto only when neither provides a valid choice.

Auto chooses a concrete model and reasoning level for each new turn from your message and a bounded portion of the conversation so far. Compact model and reasoning controls remain beside the agent picker when you want to override the inherited choice for the next message, without making configuration a prerequisite for the first message.

The Chats drawer also offers starter presets for common conversations such as building an agent, discussing strategy, and planning a project, plus starters for changing what you already have: how an agent works, a workflow or schedule, or which Connectors an agent can use. A preset fills the composer with questions about your goal, current process, constraints, and exceptions before the agent proposes anything. It never replaces a draft you have already started.

Find chats with one agent

The Chats drawer lists every conversation you can read under its Active, Unread and Archived tabs. Choose the Filter button at the right end of the tab row, then Agent, to show only the chats with one or more agents. Search or scroll the agents you may read and select the ones you want. The chosen agents appear as chips under the tabs, the tab counts follow the filter, and the chip's remove button or Clear filters brings every chat back. The filter only narrows what you can already read, so it never shows a chat that was not shared with you, and it ends when you close the drawer. The Filter button needs permission to read agents. From an agent's drawer, Open Chats opens this list with that agent already chosen.

Follow the answer and inspect the work behind it

The conversation shows your messages, the agent's answers, and expandable evidence of the work behind them. Sending saves your message first. Task Machine then prepares the agent's permitted knowledge and conversation context in the background before starting the answer, whether you choose Auto or a fixed model.

The status line distinguishes preparing, queued, starting, and responding. Queued means the turn is waiting to start, not that the agent is already answering. For a Local Worker, it also shows a known manual pause or disconnection. Pausing that Worker prevents new work from starting and leaves turns that are already starting or running alone. A message you send while the Worker is paused is saved and waits in the Chat, then starts once you resume the Worker.

On a Local Worker, your newly prepared Chat message waits alongside comments, reactions and explicit retries, ahead of Urgent and High tasks. Scheduled work and Run next requests still come first. This ordering leaves work already starting or running alone and still respects the Worker's pause and capacity limits.

The agent runs that work using the execution choice selected for the turn, whether that is Auto, a fixed Cloud model, or a model on a selected Local Worker. When you do not make a new choice, Chat keeps a still-available conversation selection, then the Agent profile default, before falling back to Auto.

Your messages and the agent's replies stay in turn order. The internal prompt that framed the turn is recorded for provenance and never appears in the conversation.

While the response is starting or working, its activity stays below incoming messages. Once the response ends, that activity returns to its original place in the history. The collapsed summary covers the whole response rather than only its latest step. Open it to inspect the original thinking and tool groups. An update preserves your choice to keep a group open.

The newest turns load first. Scroll up for earlier turns. Very long conversations keep a bounded window, so scrolling back down loads the later turns again.

Adjacent thoughts appear as one counted row that expands to each complete thought in order. Web activity names one or two public domains and otherwise counts the sites searched, fetched, or browsed, while Skills, Connectors, and Vault activity retain their own recognizable summaries.

When the agent reads or changes a record in Task Machine, the tool call that did it names the action and links the record, such as Read Launch notes or Updated NW-12 Draft the press release. Documents the agent writes, memories it updates and Chats it starts also appear once at the end of its response in the same form. Decisions that need you remain available in the conversation. Proposals from the same source appear together, opening their group review in Inbox without leaving Chat.

Tabs sit below a Chat's title once the conversation has more to show: Chat, Library, Mentions, Browser use, and Details. A tab appears only when it has something for you to see, so a new conversation has no tab bar, and the bar updates while you watch.

Library lists the Library documents the conversation worked with, newest update first. A document is marked Created when an agent in this Chat wrote its first version, and Referenced when an agent read or changed it or a message in the Chat mentioned it. Each row shows when the document was last updated and opens it. You only see documents you can read, and an archived document leaves the list until someone restores it. The list loads more documents as you scroll and updates while you watch.

Mentions lists everything else the agents referenced, newest first: Tasks, memories, Chats, Goals, Projects, Agents, Teams, Workflows, Skills and Budgets, plus your Inbox items from the conversation. Documents appear only in Library. Each entry opens its record, and an Inbox item opens in your Inbox. Records you cannot open are left out, and the list updates while you watch.

Browser use lists the browser journeys the agents recorded, and its own section below explains how to review one. Details keeps the conversation's usage and the storage notice.

Retry and resolve a paused turn

You can send corrections and attach files while the agent works. For example, add “Focus on European customers” without stopping the answer. The same agent receives your message between steps, after its current tool call finishes. Agent, model, and reasoning choices stay fixed until that turn ends.

Your message appears as soon as it is saved. Queued means it is waiting for the agent, and Received confirms delivery without claiming that the agent followed it. New files become available before their message reaches the agent. A turn still being prepared, or a Worker without live-message support, keeps the input for later.

Choose Unqueue on one of your messages before the agent starts receiving it to return it for editing. Only that message is withdrawn. The agent keeps working, and nothing is resent until you choose Send.

Its text goes before your current draft, separated by a blank line, and both sets of files remain attached. Mentions in the returned text restore their subject cards. Returning several messages keeps all their mentions. The card stack keeps its preview compact without hiding any mention from your message or changing what reaches the Agent.

Once delivery has started, send a correction instead.

A Chat keeps one active turn. After a successful turn, still-queued input the agent has not received continues in a follow-up. Unqueued messages are excluded. Stopping, failure, and timeout preserve that input for an explicit retry or your next message. Stop never starts another turn automatically. Conversations take fair turns across the workspace when several are ready.

If the latest reply fails, times out, or you stop it, its original message shows Retry beneath it for a participant who can send. Task Machine keeps that message in place and starts one fresh linked attempt instead of posting it again.

When a turn creates a proposal, question, approval, or other Chat-visible Inbox decision, the same durable recipient-scoped item remains reachable from the conversation for the person who can decide it. Shared proposal reviews bring their captured items together, while individual requests keep their native cards. Both open their complete decision context without losing the thread.

Resolving a decision updates the item without starting another Chat turn. Send a new message when you want the agent to continue the discussion. The completed decision remains available in Chat and Inbox.

The original task proposal and your decision stay in the conversation. Task execution specs are reviewed in Inbox, not Chat, even when prepared during the conversation. They never replace the original proposal or its approval.

If credential or Connector authorization blocks a running turn, approval still resumes that same paused run. If the agent paused before an action that needs a decision, approving or rejecting that action resumes the run. The composer stays visible but unavailable while it waits. Background Playbook generation also delivers its result when ready. An Inbox decision about that result does not start another turn.

Local and Cloud conversations share one Chat capacity lane, separate from Task execution capacity. A Cloud Chat retains its conversation files between turns without creating a synthetic Task.

Keep inputs and replies in the conversation

Attachments remain available to later turns, and suggested responses remain editable shortcuts rather than automatic replies. Sharing and context explains persistence, retries, filename reuse, and when response choices disappear.

Choose the microphone beside Send to dictate in English into the editable draft. Dictation replaces Send while you grant microphone access, record or wait for the final words. Stop and Cancel restore Send without submitting your message.

Browser use stays reviewable inside the conversation

After an Agent uses the browser, the Chat's Browser use tab lists its named journeys, newest first. The tab appears as soon as the first journey is recorded, and the list loads more journeys as you scroll. Only an active journey carries a Live badge.

Select a journey to inspect its captured frames and chronological action log.

You can move through the evidence with:

• Previous and Next, or the Left and Right keys.
• A swipe across the viewer.
• A thumbnail or an action in the log.

The image and highlighted action stay synchronized. Missing images retain their action entries, so an incomplete capture does not erase the log.

Viewing this history requires Browser use access as well as access to the Chat, and the tab stays hidden without it. Frames are masked before storage, but can still contain other confidential page content. They follow the conversation's retention and are not public artifacts.

Browser use explains capture controls, masking, missing frames, and navigation in detail. See Privacy and data for the wider data boundary.

Continue with Planning and creating work or Sharing and context.