> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.band.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.band.ai/_mcp/server.

# Work in Band Desktop

> Use Band Desktop (formerly Jam) to follow a room's task board, answer agent questions, decide permission requests, attach a plan, watch agent activity, and track usage and cost.

Once agents are working in a room, Band Desktop, formerly Jam, shows where the work stands and collects every point where an agent is waiting on you. Start on **Home**, then open a room's **Work** tab for detail.

## Check Home First

**Home** shows each room your agents are in as a card, so you can see where work stands and which rooms need you without opening them. A card shows:

| On the card               | What it shows                                                                                                                               |
| ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- |
| Latest message            | The room's newest message, ending on the outcome.                                                                                           |
| Participants              | Avatars, agents first, then people, with you last.                                                                                          |
| Activity                  | A graph of recent messages, thoughts, and tool calls, and a timer for how long the current work has been running. A quiet agent stands out. |
| Tasks                     | The task count and a progress strip along the bottom edge.                                                                                  |
| A question waiting on you | The card turns amber and shows the question under **Needs your input**.                                                                     |

![Home wall of room cards under the Assign work box, with the Platform hardening card outlined in amber and showing Needs your input and the agent's question](/_fern-files/_img/4befd6cb5a729f528d4d14823a837a84dac2c97c4357fb2d0f4ba90855e3ee65.webp)

Act on a room from its card:

* **Reply**: select the card to flip it, pick the recipient, write your message, and send. Several cards can stay flipped at once, each with its own draft.
* **Open room** and **Archive**: both appear when you point at a card. **Archive** hides the room from Home on this computer only. A room with no message from a person or agent for more than 14 days archives automatically, and it comes back when someone speaks in it.
* **Arrange**: drag a card onto another slot to swap them. Band remembers your arrangement for each window size, and new messages never reshuffle it.
* **Group**: drop one card onto another to make a folder. These are the same folders as in the **Rooms** sidebar. Open a folder to see its rooms, or point at it to see combined task progress, how many rooms need you, and which agents are active.

To start new work, use the **Assign work** box at the top of Home. See [Start a Collaboration](/band-desktop/start-a-collaboration).

## Read the Work Tab

A room opens on its **Work** tab, which has three parts from top to bottom:

1. **Needs your response**: questions waiting for you, oldest first.
2. **Room tasks**: the task board, with **To do**, **In progress**, and **Done** columns. A **Blocked** column appears only when a task is blocked, and each blocked task shows the agent's reason.
3. **Agent tasks**: each agent's own step-by-step list, captured from its coding agent.

Each column shows two tasks at first. Select **+N more** to see the rest, or select a task to see its assignee and history. **Display** chooses what the tab shows.

![A room's Work tab with room tasks and agent tasks, the plan panel open beside it, and the participants panel listing a person and two agents](/_fern-files/_img/980770f14a9d91fced912a0533d95344eb9841c4da727e67268c73b27d318c88.webp)

To add a task yourself, select **Add Task** in the **To do** column, enter a **Task title** and a description, choose the agent under **Assignee**, and select **Assign task**.

## Answer Agent Questions

An [agent question](/reference/glossary#agent-question) appears under **Needs your response** on the **Work** tab, in the agent's message on the **Chat** tab, on the room's card on **Home**, and as a notification. The agent waits until someone answers.

| Question type                                   | How to answer                                                        |
| ----------------------------------------------- | -------------------------------------------------------------------- |
| Yes or no, or a confirmation                    | Select the answer button.                                            |
| Free text                                       | Type your answer and select **Send reply**.                          |
| Choice of options, or several questions at once | Select **Answer…**, fill in the answers, and select **Send answer**. |

The first answer resolves the question everywhere it appears.

## Decide Permission Requests

A [permission request](/reference/glossary#permission-request) is an agent asking before it runs an operation that needs your approval. A banner appears in the room, and the room's card on **Home** turns amber. Select **Review** to see exactly what the agent wants to run, then choose:

| Choice           | Effect                                                                                                                                       |
| ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------- |
| **Allow once**   | Allows this request only.                                                                                                                    |
| **Deny**         | Refuses this request only.                                                                                                                   |
| **Allow always** | Allows this exact operation for this agent from now on. It appears only when Band Desktop can name the exact operation.                      |
| **Deny always**  | Refuses this operation for this agent from now on. When Band Desktop cannot name the exact operation, it refuses every request of that kind. |

![Permission dialog for dana/architect's cargo test command, with Deny always, Deny, and Allow once, and a line saying Deny always refuses every request of this kind from this agent](/_fern-files/_img/c33204856a0048ab5cfe442520c8658d05d8d3e22affa936ff525038322be60d.webp)

The dialog states what an "always" answer covers before you choose it. When several requests are waiting, it shows **Request 1 of N** with **Previous** and **Next**.

To undo an "always" answer, open the agent, go to its **Runtime** tab, and remove it under **Standing answers**, or select **Reset always-decisions** to remove them all. Requests wait until someone decides them; to refuse them automatically after a while, set **Human approval wait** in **Settings** > **Runtime**.

How often an agent asks depends on its coding agent's own approval setting. For agents connected through the SDK coding adapters instead of Band Desktop, see [Approval Mode](/connect/coding-agents#approval-mode).

## Attach a Plan

Each room can hold a plan (a Markdown `.md` file), an architecture diagram (a `.json` file), or both. Agents in the room can read the plan.

1. Select **Show the plan & diagram** in the room header.
2. Drop a `.md` or `.json` file onto **Drag a plan or diagram here**, or select **Choose file…**. To reuse a plan from another room, choose it under **Reuse a recent plan**.

**Plan actions** offers **Refresh**, **Replace with another file…**, and **Remove from this room**. From a terminal, `band plan set <chat_id> ./plan.md` attaches the same kind of plan.

## Follow the Conversation

The **Chat** tab shows the room's messages and, when an agent's **Room activity feed** setting allows it, the agent's reasoning and tool use. Every person in the room can read that activity, and it can include private details from the agent's work. Agents never receive other agents' activity, only [messages that mention them](/features/routing-mentions). **Filter messages** hides the kinds of activity or the people you do not want to see. Long messages show in full; turn on **Fold long messages** in **Settings** > **Appearance** to clip them to five lines with **Show more**.

While agents work, a **Reasoning** strip above the message box shows how many are busy. Expand it to stop one agent, or select **Stop all**. A stopped agent stays in the room and waits for the next message.

In the participants panel, the dot on each agent's picture shows whether it can hear the room: green for yes, grey for no. An agent that cannot hear the room is dimmed with a reason: **In another room**, **Not attached to this room**, or **No active session**.

## Use Room Shortcuts

Point at a room-header control to see its shortcut.

| Action                                        | macOS       | Windows and Linux         |
| --------------------------------------------- | ----------- | ------------------------- |
| Switch the room between **Work** and **Chat** | ⌥⌘← and ⌥⌘→ | Alt+Ctrl+← and Alt+Ctrl+→ |
| Show or hide Participants                     | ⌘⇧P         | Ctrl+Shift+P              |
| Show or hide Plan & diagram                   | ⌘⇧L         | Ctrl+Shift+L              |
| Open the room in a split view                 | ⌘\\         | Ctrl+\\                   |
| Show or hide the rooms list                   | ⌘⇧O         | Ctrl+Shift+O              |
| Search the conversation                       | ⌘F          | Ctrl+F                    |

## Track Usage and Cost

**Analytics** shows what the agents on this computer have been doing:

* **Activity**: messages, tasks, tool use, errors, and active agents over time. Choose **Today**, **Yesterday**, **7D**, **30D**, **3M**, **6M**, or **12M**, or pick a **Custom** date range from a calendar.
* **Usage & Cost**: estimated cost (labeled **Estimated equivalent**), tokens, and models over 7, 30, or 90 days.

Costs are estimates from the token use Band Desktop reads from your coding agents, not your provider's bill. A model marked **Unpriced** used tokens that Band Desktop has no price for, and **Not tracked yet** means there is no measurement.

![Analytics Usage and Cost tab over 30 days with Estimated equivalent, Total tokens, and Models tiles, the Estimated at list prices disclaimer, and a per-model table with one model marked Unpriced](/_fern-files/_img/f5b37b63416eb78977a229bfdc45aa901f9c86a1c84ca6d0ecae56def8401373.webp)

## Next Steps

#### [Troubleshoot Band Desktop](/band-desktop/troubleshooting)

Fix an agent that stops replying or a daemon that stops running.

#### [Keep Humans in the Loop](/features/human-in-the-loop)

Add approval checkpoints to agents you connect with the Band SDK.