# Agent Live Activity

> When an agent works on one of your saved servers, databases or Kubernetes clusters through AgentsRoom, you see what it does as it happens: every command it runs on a server (with its output streaming live), every SQL query and its first rows, every kubectl call.

- Area: Desktop, mobile and web
- Plans: All plans
- Last checked against the product: 2026-11-07
- Web page: https://agentsroom.dev/docs/agent-live-activity

## What it does

When an agent works on one of your saved servers, databases or Kubernetes clusters through AgentsRoom, you see what it does as it happens: every command it runs on a server (with its output streaming live), every SQL query and its first rows, every kubectl call. One journal per connection keeps the recent calls of every agent, so you can follow an agent on a production machine without opening a terminal yourself.

**Video (1 min 23 s):** [Agent Live Activity, shown in the real app](https://youtu.be/_USubsg2EWM). An agent is asked to check a staging server; the commands land in the Activity tab one by one, then the connection journal shows each call with its output, its exit code and how long it took.

## Where to find it

- Right panel, **Activity** tab (next to Files, Changes and Tests): a filter bar (**All** with the project's total number of calls, then one pill per agent that ran something, with its avatar and its own count; it follows the agent you are looking at until you pick one, and clicking the active pill goes back to **All**), a search field ("Search a command, a connection, an output…", with **Clear search**; nothing matching says "No call matches “<your words>”"), a small help icon with a play badge at the end of that filter bar ("Watch the video tutorial on YouTube", shown once at least one call exists) that opens the video, then **CONNECTIONS** (one card per connection, with its last command, the agent, the time, a **Live** badge while a call runs and an arrow **Open in Servers**) and **RECENT CALLS** (each call with its duration, or **Running…** and a green dot while it is going). Before any agent has touched a machine it reads "No agent activity yet", with the three kinds (Server, Database, Cluster) and an example prompt.
- **Commands**, **Servers** and **Databases** buttons (dev command bar): the avatars of the agents working there, up to three then `+N`, their names at the top of the tooltip. Commands shows the agents whose saved command is running; Servers and Databases show the agents with a call running or finished in the last two minutes. Kubernetes counts on the Databases button, its window opens from there.
- Above the message box: a chip with the connection and the command (for example `Rocky  ls -la  Running…`), shown while the call runs and for 30 seconds after. Click it to open the journal.
- The **Activity** tab shows a green dot while a call runs and you are on another tab. The four tabs of the panel are pills, like the places of the left column: when the panel is too narrow, the inactive ones shrink to their icon (their name in a tooltip) instead of scrolling out of sight, so the Activity tab and its green dot are always there. The panel's close cross only takes room while you hover the header.
- **Servers**, **Databases** and **Kubernetes** windows (dev command bar): on a connection, a green **Agent at work** badge with the agent's avatar while it runs something there, then a neutral **Recently here** badge for two minutes. Click it to open the journal.

## How to use it

1. Let an agent use a saved connection (it calls `ssh_exec`, `db_query` or `kube_kubectl`).
2. Click the chip above the message box, or open the **Activity** tab.
3. The journal follows the bottom while the call streams; scroll up to read, it stops following.

A click on a **connection card** of the Activity tab opens the same journal, as a large window over the panel: header `<connection> · Server · N calls`, then every call as a terminal shows it (`<agent> · <time>`, `$ <command>`, its output, `Done · exit 0 · 1.0 s`). Its header has a button to copy the whole journal and a button **Open in Servers**, which closes the journal and opens the Servers window with its search already filled in with the connection name. Escape (or a click beside it) closes the journal.

A click on a section title (**IN BACKGROUND**, **CONNECTIONS**, **RECENT CALLS**) folds that section to its title and count; click again to unfold it. The choice is saved on your account: the section stays folded on your other computers and on the phone's Activity screen.

A click on a line of **RECENT CALLS** opens the same journal, scrolled to that call and with it outlined. The journal header also offers **Copy the journal**, and the button is named after the kind of connection: **Open in Servers**, **Open in Databases** or **Open in Kubernetes**.

### Encoded or very long commands

Agents often send a script as base64 (`echo aW1w… > run.b64`, `powershell -EncodedCommand …`). The lists and the chip show `‹base64›` in its place, and the journal shows the command shortened with a **Decoded** block under it: the text of the script, even when the agent split it over several `echo`. **Show the raw command** gives the agent's exact text back, **Show all** unfolds what is long, the search finds words inside the decoded text, and **Copy the journal** copies both. Only what really is text gets decoded: a hash, a path or a binary stays as it is, and a compressed (gzip) payload is not decoded. A command is kept up to 16 KB: past that, the block says **Decoded, start only: the command was cut off**.

## Settings

- `uiElements.rightPanel.tabActivity` (global, Settings > Interface elements > Right panel > Activity): shows the Activity tab. Default: shown.
- `activityCollapsedSections` (global, set by clicking a section title of the Activity tab): the folded sections, `{ background?: true, connections?: true, timeline?: true }`. Default: `{}`, every section unfolded. Saved on the account, phone included.

## Agent tools (MCP)

Nothing new for agents: the journal records the calls of `ssh_exec`, `ssh_transfer`, `db_query`, `db_schema` and `kube_kubectl` as they happen.

## Providers

All providers, no difference: every CLI reaches the connections through the same AgentsRoom MCP.

## Mobile

Yes, for a whole project and per agent. In a project, the **…** menu has an **Activity** row with the same number as the desktop's Activity tab (calls of this copy of the project plus background tasks running now, shown as "99+" beyond 99; it needs an up-to-date desktop). Tap it for the project's screen: **In background** first, with the agent behind each task, then **Recent calls**, newest first, the latest 40 with the agent's name on each, live while the screen is open. Read-only. Tap a section title to fold it, as on the desktop: the same choice, saved on the account.

Per agent, since 2026-10-31. Open an agent's console and tap its avatar or name at the top: the agent card has an **Activity** section ("What agents run on your servers, databases and clusters, live") with the number of calls, **Live** while one runs, and the connection and the latest command. With no call yet it shows "No agent activity yet". Tap it to open the journal of that agent: every call as on the desktop (connection and time, `$ command`, its output streaming live, `Done · exit 0 · 1.2 s` or **Failed**), oldest first, following the bottom until you scroll up. Read-only. An encoded command shows as on the desktop: shortened, with its **Decoded** block and **Show the raw command**.

The **Servers**, **Databases** and **Kubernetes** lists of the phone show the same badges as the desktop on a connection: **Agent at work** with the agent's avatar during a call, then **Recently here** for two minutes. Tapping the row opens the connection as usual; the agent's journal stays in its card.

While one of those lists is open, the phone only receives who is on which connection (agent ids, no command). Opening the **…** menu reads one number. Otherwise, it only receives the activity while the project's Activity screen or an agent card is open, and only for that project or that agent: the last 40 or 30 calls, the last 6,000 characters of output per call, 2,000 of command and 4,000 of decoded text, live output grouped every half second. Before it leaves the computer, what looks like a secret (API keys, `KEY=value` lines, passwords in URLs, tokens, values of your vault) is masked, on a best-effort basis, decoded text included; the desktop journal itself masks nothing. Not on the phone: the per-agent filter, the search, the per-connection journals, the avatars on the Commands, Servers and Databases tabs, **Open in Servers** and **Copy the journal**.

## Limits

- Kept in memory on this computer only, for the current session of the app: the last 120 calls per project, 32 KB of output per call. Nothing is stored on a server; the only copy that leaves the computer is the one shown on your phone while the Activity screen or an agent card is open there, through the end-to-end encrypted connection and with secrets masked (see Mobile).
- Live output is streamed for server commands (SSH, SSM). Windows hosts (WinRM), database queries and kubectl calls show their result when they finish.
- Read-only: you watch, the agent drives. To type yourself, open the connection's terminal from the Servers window.

## Common questions

- **What is the number on the Activity tab?** Since 2026-11-05, the calls of this copy of the project (the **All** count of the tab, "99+" beyond) plus the background tasks running now. The green dot stays next to it: the number says how many, the dot says "right now".
- **Can I see what my agent is doing on my server?** Yes: click the chip above the message box, or open the right panel **Activity** tab. Commands and their output appear live.
- **How do I see which agents use my servers, databases or commands?** Their avatars show on the **Commands**, **Servers** and **Databases** buttons of the dev command bar, and on the connection in the Servers, Databases and Kubernetes windows.
- **The command is unreadable, it is all base64. Can I decode it?** Yes, it is automatic: open the journal (click the call), the decoded text is under the command, and **Show the raw command** gives the original back.
- **Where did the server pill in the agent header go?** It was removed on 2026-11-02: the chip above the message box and the avatars on the buttons replace it.
- **Does it open a terminal tab per command?** No: an agent's `ssh_exec` never opens a terminal tab at the bottom. You follow it through the journal or the chip above the message box.
- **Is the output stored in my account?** No, it stays in memory on your computer and is gone when the app quits.
- **Can I follow it from my phone?** Yes. For the whole project: **…** menu > **Activity**. For one agent: open its console, tap its avatar or name, and the card's **Activity** section shows its calls; tap it for the live journal.
- **Can I collapse a section, like Recent calls?** Yes: click its title (**In background**, **Connections** or **Recent calls**). It folds to its title and count, on every computer and on the phone, until you click it again.
- **I don't see the Activity tab.** It may be hidden: Settings > Interface elements > Right panel > Activity.
- **Is there a video that shows it?** Yes, an agent checking a staging server while every command appears in the Activity tab, then the connection journal: https://youtu.be/_USubsg2EWM (also one click away: the play icon at the end of the Activity tab's filter bar).

## Related

- [SSH Connections](https://agentsroom.dev/docs/ssh-connections.md): the servers agents reach with `ssh_exec`.
- [Database Connections](https://agentsroom.dev/docs/database-connections.md): the databases agents query with `db_query`.
- [Kubernetes Clusters](https://agentsroom.dev/docs/kubernetes-clusters.md): the clusters agents drive with `kube_kubectl`.
- [Interface Elements](https://agentsroom.dev/docs/interface-elements.md): hiding the Activity tab.
- [Agent Background Work](https://agentsroom.dev/docs/agent-background-work.md): the sub-agents and commands an agent still runs, listed at the top of this tab.
